<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Zapolu Blog</title><description>Building and running the systems businesses depend on: custom software, e-commerce platforms and infrastructure.</description><link>https://zapolu.com/blog/</link><language>en</language><atom:link href="https://zapolu.com/rss.xml" rel="self" type="application/rss+xml"/><item><title>PHP2Shell: which PHP versions are safe and what to do now</title><link>https://zapolu.com/blog/php2shell-upgrade/</link><guid isPermaLink="true">https://zapolu.com/blog/php2shell-upgrade/</guid><description>Two critical PHP vulnerabilities, CVE-2026-109618 and CVE-2026-109619, are fixed in PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. Who is affected, which versions are safe, what to check on your servers, and what to do if you still run PHP 8.1 or older.</description><pubDate>Wed, 21 Oct 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;On October 21, 2026, the PHP project released
&lt;a href=&quot;https://www.php.net/archive/2026.php#2026-10-21-5&quot;&gt;PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35&lt;/a&gt;
to fix two critical vulnerabilities, CVE-2026-109618 and
CVE-2026-109619. Both let an attacker run code on a web server without
logging in. The pair is widely known as PHP2Shell, the name of the
exploit tool the researchers built from both bugs
(&lt;a href=&quot;https://sansec.io/research/php2shell&quot;&gt;Sansec’s summary&lt;/a&gt;). PHP’s
advisory states that “only an update is a complete solution.”&lt;/p&gt;
&lt;p&gt;Every vendor fact below was checked against the linked source on
October 21, 2026.&lt;/p&gt;
&lt;h2 id=&quot;who-is-affected&quot;&gt;Who is affected&lt;/h2&gt;
&lt;p&gt;According to the
&lt;a href=&quot;https://www.php.net/archive/2026.php#2026-10-21-5&quot;&gt;PHP announcement&lt;/a&gt;,
the risk covers “all PHP installations serving web requests”,
whichever way PHP is connected to the web server. That includes
Magento, Mage-OS, WordPress, Drupal, Laravel and Symfony applications
and any custom PHP code. Sansec reports that the researchers ran code
without a login on Magento 2.4.9, WordPress, Drupal, Nextcloud and
other applications. PHP processes the attacker’s request before the
application sees it, so a store’s own code and extensions cannot
block the attack.&lt;/p&gt;
&lt;p&gt;The second vulnerability can also expose what the server holds in
memory. A server that ran a vulnerable version while reachable from
the internet may have leaked passwords, API keys or session data. Once
it is updated, consider changing the credentials it uses.&lt;/p&gt;
&lt;p&gt;Merchants on Shopify have no PHP installation of their own to update,
because Shopify hosts the platform
(&lt;a href=&quot;https://www.shopify.com/faq&quot;&gt;Shopify’s FAQ&lt;/a&gt;). Custom apps written in
PHP and hosted on your own servers are affected like any other PHP
code.&lt;/p&gt;
&lt;h2 id=&quot;safe-versions&quot;&gt;Safe versions&lt;/h2&gt;
&lt;div class=&quot;table-scroll&quot;&gt;


































&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;Branch&lt;/th&gt;&lt;th&gt;Safe from&lt;/th&gt;&lt;th&gt;Security support ends&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;&lt;tr&gt;&lt;td&gt;8.5&lt;/td&gt;&lt;td&gt;8.5.12&lt;/td&gt;&lt;td&gt;December 31, 2029&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;8.4&lt;/td&gt;&lt;td&gt;8.4.27&lt;/td&gt;&lt;td&gt;December 31, 2028&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;8.3&lt;/td&gt;&lt;td&gt;8.3.36&lt;/td&gt;&lt;td&gt;December 31, 2027&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;8.2&lt;/td&gt;&lt;td&gt;8.2.35&lt;/td&gt;&lt;td&gt;December 31, 2026&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;8.1 and older&lt;/td&gt;&lt;td&gt;no fix from the PHP project&lt;/td&gt;&lt;td&gt;ended&lt;/td&gt;&lt;/tr&gt;&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;The support dates come from PHP’s
&lt;a href=&quot;https://www.php.net/supported-versions.php&quot;&gt;supported versions page&lt;/a&gt;.
A system on PHP 8.2 needs this update now and a move to a newer branch
before the end of the year, unless it runs Debian 12’s own PHP
packages, which Debian’s long-term support team maintains
(&lt;a href=&quot;https://security-tracker.debian.org/tracker/DLA-4848-1&quot;&gt;DLA-4848-1&lt;/a&gt;).&lt;/p&gt;
&lt;h2 id=&quot;what-to-check-on-your-servers&quot;&gt;What to check on your servers&lt;/h2&gt;
&lt;p&gt;Installing the update is usually quick. Most of the risk is in
believing it was installed when it was not. These are the questions
to ask whoever runs your servers:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Which PHP version does the web server actually run?&lt;/strong&gt; It can
differ from the version shown on the command line, and one server
can run several PHP versions side by side.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Does the version number prove the fix?&lt;/strong&gt; Ubuntu, for example,
adds the fix to its older PHP package, so a patched Ubuntu 24.04
server still reports PHP 8.3.6
(&lt;a href=&quot;https://ubuntu.com/security/notices/USN-8948-1&quot;&gt;Ubuntu’s notice&lt;/a&gt;).
There, the package version is what has to match the vendor’s notice.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Was PHP restarted after the update?&lt;/strong&gt; A running PHP process keeps
the old code in memory until it is restarted.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Were the Docker images rebuilt?&lt;/strong&gt; A running container does not
update itself. The official PHP images on Docker Hub were rebuilt with the fixed versions later on October 21, so an image built from them before that still contains a vulnerable version.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Is the hosting provider handling it?&lt;/strong&gt; On Adobe Commerce on Cloud,
Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-on-cloud/user-guide/configure/app/php-settings&quot;&gt;PHP settings page&lt;/a&gt;
does not say when the PHP patch release is updated, and we found no
Adobe notice about these vulnerabilities on October 21. Check the
version after the next deployment, and ask Adobe Support if it is
still below the safe release.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;magento-stores-on-older-versions&quot;&gt;Magento stores on older versions&lt;/h2&gt;
&lt;p&gt;Magento limits which PHP a store can run, per Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/installation-guide/system-requirements&quot;&gt;system requirements&lt;/a&gt;.
Stores on 2.4.8 and 2.4.9 can move to a safe release today. A store on
2.4.7 can run PHP 8.3, which is a better target than 8.2.&lt;/p&gt;
&lt;p&gt;The latest patches of 2.4.4 and 2.4.5 support only PHP 8.1, which no
longer receives fixes from the PHP project, and these lines are past
the end of Adobe’s security support
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/planning/lifecycle-policy&quot;&gt;lifecycle policy&lt;/a&gt;).
Such a store depends on a fix from its operating system vendor (next
section) until the Magento upgrade is done.&lt;/p&gt;
&lt;h2 id=&quot;php-81-and-older&quot;&gt;PHP 8.1 and older&lt;/h2&gt;
&lt;p&gt;The PHP project no longer releases versions 8.1 or older. It
published minimal fixes as source code for 8.1, 8.0 and 7.4 and points
users to their vendors. Some vendors have already shipped them:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Ubuntu&lt;/strong&gt; fixed PHP 8.1 on Ubuntu 22.04 in its regular updates
(&lt;a href=&quot;https://ubuntu.com/security/notices/USN-8948-1&quot;&gt;USN-8948-1&lt;/a&gt;), and
older PHP versions on older Ubuntu releases for Ubuntu Pro customers
(&lt;a href=&quot;https://ubuntu.com/security/notices/USN-8948-2&quot;&gt;USN-8948-2&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Ondřej Surý’s repositories&lt;/strong&gt;, a common source of PHP on Debian and
Ubuntu, published fixed builds of PHP 5.6 to 8.1
(&lt;a href=&quot;https://launchpad.net/~ondrej/+archive/ubuntu/php&quot;&gt;PPA&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Red Hat&lt;/strong&gt; had fixed the PHP 7.4 packages on RHEL 8 and had no fix yet for the other PHP versions on RHEL 8, 9 and 10 (&lt;a href=&quot;https://security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-109618.json&quot;&gt;Red Hat’s data&lt;/a&gt;).&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The PHP project describes these fixes as a “Minimal mitigation for EOL
version.” They close the known attack paths on a version that receives
no other security fixes. The upgrade to a supported PHP version is
still required.&lt;/p&gt;
&lt;h2 id=&quot;if-the-update-has-to-wait&quot;&gt;If the update has to wait&lt;/h2&gt;
&lt;p&gt;PHP lists configuration changes that lower the risk, and states that
they “reduce exposure to specific attack paths but do not make an
installation safe.” They include turning off session upload progress
and lowering the maximum request size, and the full list is in the
&lt;a href=&quot;https://www.php.net/archive/2026.php#2026-10-21-5&quot;&gt;PHP announcement&lt;/a&gt;.
Cloudflare has also
&lt;a href=&quot;https://blog.cloudflare.com/php-vulnerabilities-waf-protection/&quot;&gt;deployed firewall rules&lt;/a&gt;
for these vulnerabilities. Both lower the risk until the update is
installed.&lt;/p&gt;
&lt;p&gt;Blocking multipart requests, the format browsers use for file
uploads, is sometimes suggested as a stopgap. On a Magento store it also blocks product image uploads in
the Admin and, on some stores, the cart or checkout, per
&lt;a href=&quot;https://sansec.io/research/php2shell&quot;&gt;Sansec’s analysis&lt;/a&gt;. Place a
test order after any such change.&lt;/p&gt;
&lt;h2 id=&quot;exploitation-status&quot;&gt;Exploitation status&lt;/h2&gt;
&lt;p&gt;Cloudflare reports that the details leaked before the release and that
it “observed attack traffic targeting our customers”
(&lt;a href=&quot;https://blog.cloudflare.com/php-vulnerabilities-waf-protection/&quot;&gt;Cloudflare’s post&lt;/a&gt;).
When we checked on October 21, neither CVE was yet in CISA’s
&lt;a href=&quot;https://www.cisa.gov/known-exploited-vulnerabilities-catalog&quot;&gt;Known Exploited Vulnerabilities catalog&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;If you are not sure which PHP version your servers run, or whether the
update reached every one of them, we can check and update them for
you. Keeping the PHP runtime current is part of our
&lt;a href=&quot;/services/ongoing-care/&quot;&gt;ongoing care retainer&lt;/a&gt;. For a one-off update, &lt;a href=&quot;/contact/&quot;&gt;send us a message&lt;/a&gt;.&lt;/p&gt;</content:encoded><category>security</category><category>operations</category><category>php</category></item><item><title>From October 30, Adobe Commerce on Cloud stores on outdated services face suspension</title><link>https://zapolu.com/blog/adobe-commerce-cloud-deadlines/</link><guid isPermaLink="true">https://zapolu.com/blog/adobe-commerce-cloud-deadlines/</guid><description>Adobe can now lock the Admin or take Commerce on Cloud stores offline when they run unsupported services or versions. What is due on October 30, 2026, what follows in 2027 and 2028, and what to do for the version you run.</description><pubDate>Tue, 20 Oct 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;On October 30, 2026, Adobe starts enforcing a new
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/planning/security-enforcement-policy&quot;&gt;security policy&lt;/a&gt;
on Adobe Commerce on Cloud. An environment that still runs
Elasticsearch, MariaDB 10.5 or older, or RabbitMQ 3.9 or older after
that date “may have Admin access suspended, while the storefront
remains online, and/or have inbound traffic suspended, taking the
storefront offline.” More deadlines follow in May 2027, and after that the older
Magento versions themselves have to be upgraded.&lt;/p&gt;
&lt;p&gt;The policy applies to Adobe Commerce on Cloud (PaaS) on versions 2.4.4
through 2.4.9. It does not apply to Adobe’s SaaS product or to stores
you host yourself. If that is you, the short section near the end is
the relevant part. Dates and quotations below were checked against
Adobe’s own pages on October 20, 2026.&lt;/p&gt;
&lt;h2 id=&quot;due-dates&quot;&gt;Due dates&lt;/h2&gt;
&lt;p&gt;These are the services Adobe requires you to move off:&lt;/p&gt;
&lt;div class=&quot;table-scroll&quot;&gt;

















































&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;If you run&lt;/th&gt;&lt;th&gt;Move to&lt;/th&gt;&lt;th&gt;By&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;&lt;tr&gt;&lt;td&gt;Elasticsearch (any version)&lt;/td&gt;&lt;td&gt;OpenSearch (2.19 on 2.4.4 and 2.4.5, 3 on 2.4.6 and later)&lt;/td&gt;&lt;td&gt;October 30, 2026&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;MariaDB 10.5 or older&lt;/td&gt;&lt;td&gt;MariaDB 10.6 or newer&lt;/td&gt;&lt;td&gt;October 30, 2026&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;RabbitMQ 3.9 or older&lt;/td&gt;&lt;td&gt;RabbitMQ 3.13 or newer&lt;/td&gt;&lt;td&gt;October 30, 2026&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;PHP 8.1 or older&lt;/td&gt;&lt;td&gt;PHP 8.2 or newer&lt;/td&gt;&lt;td&gt;May 31, 2027&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;MariaDB 10.6 to 10.10&lt;/td&gt;&lt;td&gt;MariaDB 10.11 or newer&lt;/td&gt;&lt;td&gt;May 31, 2027&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;OpenSearch 1.x&lt;/td&gt;&lt;td&gt;OpenSearch 2.19 or 3, as above&lt;/td&gt;&lt;td&gt;May 31, 2027&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Redis 5 or older&lt;/td&gt;&lt;td&gt;Valkey 8 or newer&lt;/td&gt;&lt;td&gt;May 31, 2027&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;RabbitMQ 3.10 to 3.12&lt;/td&gt;&lt;td&gt;RabbitMQ 4.3 or newer&lt;/td&gt;&lt;td&gt;May 31, 2027&lt;/td&gt;&lt;/tr&gt;&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;This applies to every Magento version, including 2.4.8 and 2.4.9, and
to every environment, not only production.&lt;/p&gt;
&lt;p&gt;The Magento version has its own deadline. Stores on 2.4.4 or 2.4.5
must upgrade to 2.4.9, or migrate to Adobe’s SaaS, by June 1, 2027.
Stores on 2.4.6 or 2.4.7 have until June 1, 2028. Adobe has set no
deadline for 2.4.8 and 2.4.9. Cloud stores older than 2.4.4 are not
listed in the policy, but Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/planning/lifecycle-policy&quot;&gt;lifecycle policy&lt;/a&gt;
applies the same suspension to any unsupported version from June 1,
2027.&lt;/p&gt;
&lt;p&gt;If you are on 2.4.4 or 2.4.5, confirm your dates with Adobe. The
policy page gives these two versions October 30, 2026 for all service
upgrades in one table and the dates above in another. Ask in your
support ticket and get the answer in writing.&lt;/p&gt;
&lt;h2 id=&quot;what-happens-if-you-miss-it&quot;&gt;What happens if you miss it&lt;/h2&gt;
&lt;p&gt;Adobe may suspend Admin access for all users while the storefront
stays online, suspend traffic so the storefront goes offline, or both.
If the environment is still not compliant after a traffic suspension,
“Adobe may terminate
the cloud services”, and decommissioning permanently deletes all
data, environments and branches. Adobe states no notice period before
either step.&lt;/p&gt;
&lt;p&gt;If you cannot make a deadline, the policy says to “contact your
account team or Adobe Support as soon as possible” for “a short
extension”. It does not say how long an extension can be.&lt;/p&gt;
&lt;h2 id=&quot;check-what-you-run&quot;&gt;Check what you run&lt;/h2&gt;
&lt;p&gt;You need two things: the Magento version and the version of every
service in every environment.&lt;/p&gt;
&lt;p&gt;The Magento version is in the bottom-right corner of any Admin page,
or comes from &lt;code&gt;bin/magento --version&lt;/code&gt;. For the services, Adobe’s
policy points to the Cloud Console: open the environment and check
each service on its “Containers” tab, which shows the version
currently running.&lt;/p&gt;
&lt;p&gt;On Pro, check the running service rather than the repository. Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/experience-cloud-kcs/kbarticles/ka-26668&quot;&gt;knowledge base&lt;/a&gt;
describes Staging and Production environments where changes to
&lt;code&gt;services.yaml&lt;/code&gt; “are ignored” and the old versions keep running. The
file shows what was requested. The running service shows what you
have. Also note that New Relic keeps calling the cache “Redis” even
after a move to Valkey, so do not read the service from monitoring
labels.&lt;/p&gt;
&lt;h2 id=&quot;the-upgrade-procedure&quot;&gt;The upgrade procedure&lt;/h2&gt;
&lt;p&gt;On Pro, Adobe upgrades Staging and Production services for you,
through a support ticket. The policy asks for that ticket “at least
30 days before your enforcement date”. For October 30 that point
passed on September 30, so if you have not filed yet, file now and ask
about an extension in the same ticket. Production changes also need
at least two business days’ notice, not counting weekends, and more
lead time during busy periods.&lt;/p&gt;
&lt;p&gt;On Starter, Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-on-cloud/user-guide/configure/service/services-yaml&quot;&gt;services page&lt;/a&gt;
describes the change as self-service: merging an updated
&lt;code&gt;.magento/services.yaml&lt;/code&gt; to the production branch upgrades production.
Services cannot be downgraded afterwards, so take a backup first. The
policy itself tells all Cloud customers to open a ticket, so confirm
with Adobe who makes the change on your project.&lt;/p&gt;
&lt;p&gt;Either way, some of the work stays with you:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;test the upgrade on a non-production environment first, at least
checkout, search, cart and your integrations,&lt;/li&gt;
&lt;li&gt;redeploy after Adobe finishes, which also rebuilds the RabbitMQ
queues,&lt;/li&gt;
&lt;li&gt;run a full reindex after an OpenSearch change,&lt;/li&gt;
&lt;li&gt;have someone from your team or partner online for the whole
Production window, because Adobe does not test your application.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Adobe’s services page suggests validating service upgrades one at a
time and warns that
“Database upgrades carry the highest operational impact and deserve
the most caution.” On Pro, rehearse on Staging with production-sized
data. The time it takes there is your best estimate for the
Production window.&lt;/p&gt;
&lt;h2 id=&quot;a-service-upgrade-as-a-magento-upgrade&quot;&gt;A service upgrade as a Magento upgrade&lt;/h2&gt;
&lt;p&gt;Each Magento patch supports specific service versions, listed in
Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/installation-guide/system-requirements&quot;&gt;system requirements&lt;/a&gt;,
and Adobe supports only the combinations listed there. The version
you are asked to reach may therefore not run on the Magento version
you have. These cases follow from that table:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;2.4.4 and 2.4.5 cannot run PHP 8.2.&lt;/strong&gt; Every patch of both
versions supports PHP 8.1 only. Meeting the May 31, 2027 PHP
deadline means upgrading Magento by then, a day earlier than the
June 1 version deadline.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;2.4.6 cannot go beyond PHP 8.2,&lt;/strong&gt; and PHP 8.2 stops receiving
security fixes on December 31, 2026, per
&lt;a href=&quot;https://www.php.net/supported-versions.php&quot;&gt;php.net&lt;/a&gt;. Adobe’s own
lifecycle policy marks this as a PCI risk from the end of 2026. The
only way off it is a Magento upgrade.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Older patches need a patch upgrade for the October deadline.&lt;/strong&gt;
For example, OpenSearch 2 needs at least 2.4.4-p13 or 2.4.5-p12, and
RabbitMQ 3.13 needs 2.4.4-p18. On an older patch, the patch upgrade
comes first.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;OpenSearch 3 on 2.4.6 and 2.4.7 needs the latest patch&lt;/strong&gt;
(2.4.6-p15 or 2.4.7-p10). Those patches also require Valkey instead
of Redis, RabbitMQ 4.3 and MariaDB 10.11 or newer, so reaching
OpenSearch 3 can pull those changes forward too. Earlier 2.4.6 and
2.4.7 patches support OpenSearch 2. Whether that is enough for
October 30 is a question for your ticket.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The jump from an old version to 2.4.9 takes several steps.&lt;/strong&gt; No
single patch supports both the old and the new versions of every
service, and RabbitMQ has to be upgraded one version at a time. Ask
Adobe how many Production windows to plan for.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;two-changes-that-are-bigger-than-they-look&quot;&gt;Two changes that are bigger than they look&lt;/h2&gt;
&lt;p&gt;Adobe itself warns that “Elasticsearch to OpenSearch and Redis to
Valkey are migrations to different software rather than version
upgrades.”&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Elasticsearch to OpenSearch.&lt;/strong&gt; On Cloud, the search engine is set
in &lt;code&gt;.magento.env.yaml&lt;/code&gt; (&lt;code&gt;SEARCH_CONFIGURATION&lt;/code&gt;) and in the service
relationships in &lt;code&gt;.magento.app.yaml&lt;/code&gt;. Both need to point at the new
service. Search extensions need checking too. Smile’s ElasticSuite,
for example,
&lt;a href=&quot;https://github.com/Smile-SA/elasticsuite/blob/86cbfc3a246ae046b76a03129e265f901af81597/README.md&quot;&gt;supports&lt;/a&gt;
OpenSearch 3 only from Magento 2.4.8, so a 2.4.6 or 2.4.7 store using
it should settle the OpenSearch version with Adobe before October 30.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Redis to Valkey.&lt;/strong&gt; Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-on-cloud/user-guide/configure/service/valkey&quot;&gt;Valkey page&lt;/a&gt;
says to treat the switch as a fresh cache: sessions may not survive,
and “users are logged out after the migration completes.” Remove the
old Redis definition, because if both are available,
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/implementation-playbook/best-practices/planning/redis-valkey-service-configuration&quot;&gt;Magento keeps using Redis&lt;/a&gt;. The latest patches of 2.4.5 to 2.4.7, and all of 2.4.8
and 2.4.9, no longer support Redis at all, so a store that keeps its
patches current makes this move before the May 2027 deadline.&lt;/p&gt;
&lt;h2 id=&quot;steps-per-version&quot;&gt;Steps per version&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Every Cloud store, now:&lt;/strong&gt; check the running version of every
service in every environment. If anything in the October rows
applies, or you are on 2.4.4 or 2.4.5, open a support ticket now.
Ask which dates apply to you, how many Production windows the work
needs, and for an extension if October 30 is out of reach.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;2.4.4 and 2.4.5:&lt;/strong&gt; upgrade to 2.4.9, or migrate to Adobe’s SaaS,
and finish by May 31, 2027. File the service tickets well before
May 1, which is the last possible date, not a starting point.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;2.4.6:&lt;/strong&gt; start the upgrade to 2.4.9 now, because PHP 8.2 loses
security support at the end of 2026.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;2.4.7:&lt;/strong&gt; move to PHP 8.3, which receives security fixes until
December 31, 2027, and plan the upgrade to 2.4.9 before then.
Adobe’s deadline is June 1, 2028.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;2.4.8 and 2.4.9:&lt;/strong&gt; check the service table above. There is no
version deadline yet. Standard support for 2.4.8 ends on May 31, 2028.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Older than 2.4.4:&lt;/strong&gt; upgrade to 2.4.9 before June 1, 2027, and ask
Adobe whether the October deadlines apply to you. OpenSearch
support starts at 2.4.4-p13, so leaving Elasticsearch already
requires a Magento upgrade.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;if-you-host-magento-yourself&quot;&gt;If you host Magento yourself&lt;/h2&gt;
&lt;p&gt;Nothing in this policy takes a self-hosted store offline. It does not
cover on-premises Adobe Commerce, Magento Open Source or Mage-OS. The
same software still reaches end of life, though: Elastic’s support
for Elasticsearch 7.17 ended on January 15, 2026, per its
&lt;a href=&quot;https://www.elastic.co/support/eol&quot;&gt;end-of-life page&lt;/a&gt;, and PHP 8.2
follows at the end of 2026. Our &lt;a href=&quot;/blog/magento-hosting/&quot;&gt;hosting post&lt;/a&gt;
covers which hosts keep up with Adobe’s requirements.&lt;/p&gt;
&lt;h2 id=&quot;the-alternative-adobe-offers&quot;&gt;The alternative Adobe offers&lt;/h2&gt;
&lt;p&gt;Every version deadline comes with a second option: moving to Adobe
Commerce as a Cloud Service, which Adobe calls its “recommended
long-term destination” for Cloud customers. Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce/cloud-service/migration/overview&quot;&gt;migration overview&lt;/a&gt;
makes clear that this is a rebuild: customizations become App Builder
applications and storefronts are rebuilt on Edge Delivery Services.
The SaaS
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce/cloud-service/overview&quot;&gt;does not support&lt;/a&gt;
Luma storefronts. For most stores this is a
replatforming project, and our
&lt;a href=&quot;/blog/replatform-or-fix/&quot;&gt;replatform-or-fix post&lt;/a&gt; covers how to
weigh it against an upgrade.&lt;/p&gt;
&lt;p&gt;This is the version and service upgrade work our
&lt;a href=&quot;/services/magento/&quot;&gt;Magento service&lt;/a&gt; covers.&lt;/p&gt;</content:encoded><category>magento</category><category>operations</category><category>security</category></item><item><title>Who owns which field: the store, the ERP, or the PIM</title><link>https://zapolu.com/blog/product-data-ownership/</link><guid isPermaLink="true">https://zapolu.com/blog/product-data-ownership/</guid><description>Stock is the easy field. Price, attributes, media, alt text, per-locale copy and URL keys are the ones two systems both write, and the ownership rule is limited by what your platform can express.</description><pubDate>Thu, 06 Aug 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Every store that connects an ERP eventually settles who owns the
stock number. The answer is
&lt;a href=&quot;/blog/erp-eshop-stock-mismatch/&quot;&gt;almost always the ERP or the warehouse system&lt;/a&gt;,
because that is where goods are booked and where they physically
move, and once that is written down the synchronization bugs become
mechanical.&lt;/p&gt;
&lt;p&gt;Then the same question arrives for every other field, and the tidy
answer stops working. Stock is one number with one meaning. Price is
one number per website, or per market, or per customer company.
Product copy is one value per language. An image is one file, but its
alt text is a sentence per language. Nobody argues about who owns
stock for long. The rest of the record is where the ownership
question stays open.&lt;/p&gt;
&lt;p&gt;This post gives a field-by-field ownership rule for prices,
attributes, media, per-locale copy, B2B pricing and URL keys, then
checks each rule against what Magento and Shopify can actually
express, where the ERPs used in Central Europe stop, and when a PIM
earns its place. The platform half is Magento and Shopify. A Czech or
Slovak store is at least as likely to be on Shoptet, and the
field-ownership question there is the same even though the mechanisms
are not. Every vendor fact was checked against the vendor’s
own documentation on August 4, 2026, and the load-bearing ones carry
their link. Where something could not be verified, we say so.&lt;/p&gt;
&lt;h2 id=&quot;one-owner-per-field-at-one-scope&quot;&gt;One owner per field, at one scope&lt;/h2&gt;
&lt;p&gt;The principle from the stock post still holds: one system of record
per fact. What changes is that a field is not a fact until you have
also named its scope. “The ERP owns price” is not a rule an
integration can implement. “The ERP owns list price per website, and
nothing owns price per store view because the platform cannot store
one” is.&lt;/p&gt;
&lt;p&gt;The vocabulary for this predates e-commerce. Pat Helland’s
&lt;a href=&quot;https://www.cidrdb.org/cidr2005/papers/P12.pdf&quot;&gt;Data on the Outside versus Data on the Inside&lt;/a&gt;
(CIDR 2005) states the rule in one sentence: “Reference data refers
to a type of information that is created and/or managed by a single
service and published to other services for their use … For each
piece, there is exactly one publishing service.”&lt;/p&gt;
&lt;p&gt;The data-management literature says the same thing with different
words, and it is worth borrowing them because they name the two roles
a store’s systems play. The DAMA-DMBOK, second edition, chapter 10
(Reference and Master Data), defines a System of Record as “an
authoritative system where
data is created/captured, and/or maintained through a defined set of
rules and expectations,” and a System of Reference as “an
authoritative system where data consumers can obtain reliable data to
support transactions and analysis, even if the information did not
originate in the system of reference.” One of its six guiding principles for
reference and master data is Authority: “Master Data values should be
replicated only from the system of record.” Another is Ownership,
which it frames at the level of the organization rather than the
field. Your storefront is a system of reference for most of the
product record, and the argument this post is about is which system
is the record for each field.&lt;/p&gt;
&lt;p&gt;What the literature stops short of is the granularity. The second
edition names the designation, and names match, merge, survivorship
and trust rules among the business rules master-data management
needs, but it addresses entities and data elements in general rather
than the specific question of which system owns a storefront’s alt
text in Polish. The
four MDM styles usually cited as registry, consolidation, coexistence
and transaction are widely attributed to Gartner. Gartner’s own pages
refused every client we tried, and the openly readable source we
found attributes the styles to
&lt;a href=&quot;https://www.techtarget.com/data-technologies/definition/What-is-master-data-management-MDM&quot;&gt;data-management consultants and MDM vendors&lt;/a&gt;
rather than to Gartner, so we use them as a common description and
not as a definition.&lt;/p&gt;
&lt;p&gt;Propagation has better-named patterns. A push feed is an
event-driven consumer and a nightly job is a
&lt;a href=&quot;https://www.enterpriseintegrationpatterns.com/patterns/messaging/PollingConsumer.html&quot;&gt;polling consumer&lt;/a&gt;.
Tolerating a repeated message is
&lt;a href=&quot;https://www.enterpriseintegrationpatterns.com/patterns/messaging/IdempotentReceiver.html&quot;&gt;Idempotent Receiver&lt;/a&gt;,
whose two documented routes are explicit de-duplication and
semantics that make reprocessing harmless. Restoring order in a
stream that arrived out of order is a
&lt;a href=&quot;https://www.enterpriseintegrationpatterns.com/patterns/messaging/Resequencer.html&quot;&gt;Resequencer&lt;/a&gt;.
Helland’s
&lt;a href=&quot;https://www.cidrdb.org/cidr2007/papers/cidr07p15.pdf&quot;&gt;Life beyond Distributed Transactions&lt;/a&gt;
(CIDR 2007) states the requirement these patterns exist to meet:
“the application must tolerate message retries and the out-of-order
arrival of some messages.”&lt;/p&gt;
&lt;h2 id=&quot;what-your-platform-can-express&quot;&gt;What your platform can express&lt;/h2&gt;
&lt;p&gt;An ownership rule the platform cannot express cannot be implemented,
so check the rule against the data model before agreeing it with
anyone.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Magento scopes by store hierarchy.&lt;/strong&gt; Adobe documents a
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-admin/start/setup/websites-stores-views&quot;&gt;cascade of global, website, store and store view&lt;/a&gt;,
but a product attribute is scoped at only three of those four: the
store level carries the root category rather than attribute values,
so an attribute is global, per website or per store view. Two limits
decide most ownership questions. Price is not freely scopable: the
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-admin/catalog/products/pricing/catalog-price-scope&quot;&gt;Catalog Price Scope&lt;/a&gt;
setting offers Global or Website only, the page states plainly that
“Commerce does not allow setting a product price for each store,”
and the admin exposes the setting at Default Config scope only, so
it is one switch for the whole installation. Some attributes cannot
be rescoped at all: Magento locks the scope selector on &lt;code&gt;sku&lt;/code&gt; and
&lt;code&gt;category_ids&lt;/code&gt; in
&lt;a href=&quot;https://raw.githubusercontent.com/magento/magento2/2.4-develop/app/code/Magento/Catalog/etc/eav_attributes.xml&quot;&gt;eav_attributes.xml&lt;/a&gt;
and disables it for &lt;code&gt;sku&lt;/code&gt; and &lt;code&gt;media_gallery&lt;/code&gt; in
&lt;a href=&quot;https://raw.githubusercontent.com/magento/magento2/2.4-develop/app/code/Magento/Catalog/etc/adminhtml/di.xml&quot;&gt;the admin form’s configuration&lt;/a&gt;
(2.4-develop, checked August 4, 2026). An attribute used as a
configurable option
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-admin/catalog/product-attributes/product-attributes-add&quot;&gt;has to be Global&lt;/a&gt;,
though its option labels stay translatable per store view.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Shopify separates pricing from translation, and they are different
mechanisms.&lt;/strong&gt; A variant carries one base price on the product
record. An authored per-market price lives outside the product
entirely, on a price list attached to a catalog attached to a market.
A market can also diverge from the base price without any price list,
through currency conversion and rounding rules. Magento has no
equivalent: an authored price per market means a website per market,
with the store views beneath it sharing that price. Translation is a
separate surface, and on the product resource itself it is narrow:
only title, body, handle, product type, meta title and meta
description are translatable, and
&lt;a href=&quot;https://shopify.dev/docs/apps/build/markets/manage-translated-content&quot;&gt;a resource’s tags cannot be translated&lt;/a&gt;
at all. Option names, option values and publicly accessible
metafields are translatable too, as their own resource types. A
translation can also be scoped to one market through the &lt;code&gt;marketId&lt;/code&gt;
on the translation input, so the same language can read differently
in two markets. URL handles are the exception: Shopify states they
cannot be customized per market. Two mechanics matter for whoever
owns the write path.
writing a translation requires the source’s
&lt;code&gt;translatableContentDigest&lt;/code&gt;, and
&lt;a href=&quot;https://shopify.dev/docs/api/admin-graphql/latest/objects/Translation&quot;&gt;&lt;code&gt;Translation.outdated&lt;/code&gt;&lt;/a&gt;
reports whether the original content has changed since the
translation was written. That makes staleness detectable, but it does not
resolve it. And &lt;code&gt;productSet&lt;/code&gt;
reconciles list fields against the payload you send: for variants,
metafields and collections it creates and updates what is in the
payload and deletes what is not, which makes a naive partial “sync
everything” job destructive. Behavior above is the 2026-07 Admin API,
checked August 4, 2026.&lt;/p&gt;
&lt;h2 id=&quot;the-field-by-field-answer&quot;&gt;The field-by-field answer&lt;/h2&gt;
&lt;div class=&quot;table-scroll&quot;&gt;










































































































&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;Field&lt;/th&gt;&lt;th&gt;System of record&lt;/th&gt;&lt;th&gt;Scope it can actually hold&lt;/th&gt;&lt;th&gt;On conflict&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;&lt;tr&gt;&lt;td&gt;Product identity&lt;/td&gt;&lt;td&gt;ERP, as a surrogate id&lt;/td&gt;&lt;td&gt;Keep the ERP id in a Magento attribute or a Shopify metafield. Do not join on SKU, which gets renamed&lt;/td&gt;&lt;td&gt;ERP id wins. A changed SKU is a rename, not a new product&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;List price&lt;/td&gt;&lt;td&gt;ERP&lt;/td&gt;&lt;td&gt;Magento: global or per website only. Shopify: one base value per variant&lt;/td&gt;&lt;td&gt;ERP wins. Store edits are overwritten on next sync&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Tax basis and tax class&lt;/td&gt;&lt;td&gt;ERP&lt;/td&gt;&lt;td&gt;Magento: &lt;code&gt;tax_class_id&lt;/code&gt; is a website-scoped attribute, and net-or-gross display is separate configuration. Shopify: a product flag plus per-market tax settings&lt;/td&gt;&lt;td&gt;Decide which side of the VAT multiplication is authoritative and write it down. A Czech or Polish B2C catalog is authored gross, so deriving gross from a net ERP price puts rounding between the ERP and the price the customer sees&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Market or channel price&lt;/td&gt;&lt;td&gt;Store&lt;/td&gt;&lt;td&gt;Shopify price lists per catalog per market, plus currency conversion. Magento cannot separate the two: the website price is the only value, so whoever owns list price owns market price&lt;/td&gt;&lt;td&gt;Store wins for display. The ERP still receives the sold price on the order, and must not re-price an imported order from its own list&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Promotional price&lt;/td&gt;&lt;td&gt;Store, unless the ERP holds margin approval&lt;/td&gt;&lt;td&gt;Magento: same envelope as list price, or a catalog price rule per website and customer group. Shopify: compare-at price, or a catalog adjustment&lt;/td&gt;&lt;td&gt;If margin approval sits in the ERP, the ERP owns it and the store must not also run catalog rules on the same products&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;B2B or contract price&lt;/td&gt;&lt;td&gt;ERP&lt;/td&gt;&lt;td&gt;Magento: customer-group tier prices per website. Shared catalogs are available only with Adobe Commerce B2B. Shopify: catalogs per company location&lt;/td&gt;&lt;td&gt;ERP wins. Above a few dozen price profiles neither platform models per-customer contracts well&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Physical stock&lt;/td&gt;&lt;td&gt;ERP, or the WMS where one exists&lt;/td&gt;&lt;td&gt;Per source. Salable quantity is computed by the platform from sources minus reservations and is not writable&lt;/td&gt;&lt;td&gt;Never write salable quantity. Write source quantity and let the platform derive&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Variant set&lt;/td&gt;&lt;td&gt;PIM, else store&lt;/td&gt;&lt;td&gt;Magento configurable products need global option attributes. Most ERPs hold each variant as its own item with no parent&lt;/td&gt;&lt;td&gt;Whoever owns the parent grouping owns it. The ERP usually cannot express one&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Attributes and options&lt;/td&gt;&lt;td&gt;PIM, else store&lt;/td&gt;&lt;td&gt;Configurable-option attributes have to be global on Magento&lt;/td&gt;&lt;td&gt;PIM wins. The store is a display copy&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Media files&lt;/td&gt;&lt;td&gt;DAM, else PIM&lt;/td&gt;&lt;td&gt;One global file set. On Magento the base, small and thumbnail roles are per store view&lt;/td&gt;&lt;td&gt;DAM wins, but only if it keeps a stable identity per asset. Renaming on re-sync orphans every per-store-view role&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Alt text&lt;/td&gt;&lt;td&gt;PIM, else store&lt;/td&gt;&lt;td&gt;Per store view on Magento, per locale on Shopify&lt;/td&gt;&lt;td&gt;Copy owner wins&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Per-locale copy&lt;/td&gt;&lt;td&gt;ERP up to its language ceiling, else PIM, else store&lt;/td&gt;&lt;td&gt;Magento store views. On Shopify, five of the six translatable product fields, plus option names, option values and publicly readable metafields (the handle is the URL row)&lt;/td&gt;&lt;td&gt;Source locale wins. A stale translation is flagged, not replaced&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;URL key or handle&lt;/td&gt;&lt;td&gt;Store&lt;/td&gt;&lt;td&gt;Per store view on Magento, per locale on Shopify, but handles cannot be customized per market&lt;/td&gt;&lt;td&gt;Store wins. Never regenerate from the product name&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Publication and delisting&lt;/td&gt;&lt;td&gt;ERP for eligibility, store for channel&lt;/td&gt;&lt;td&gt;The ERP flag (&lt;code&gt;isInternet&lt;/code&gt;, &lt;code&gt;exportNaEshop&lt;/code&gt;) gates the catalog. The store publishes per website or channel&lt;/td&gt;&lt;td&gt;Disable rather than delete: a deleted product loses its URL, its reviews and its rewrite history&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Category or collection membership&lt;/td&gt;&lt;td&gt;PIM, else store&lt;/td&gt;&lt;td&gt;One global assignment on Magento, collections on Shopify. An ERP product-group tree is an input, not the record&lt;/td&gt;&lt;td&gt;PIM wins. The ERP tree is mapped, not mirrored&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Regulatory fields&lt;/td&gt;&lt;td&gt;Store, filled from supplier records&lt;/td&gt;&lt;td&gt;Per market, in a language that market’s Member State accepts&lt;/td&gt;&lt;td&gt;Nobody may overwrite an attestation automatically&lt;/td&gt;&lt;/tr&gt;&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;Two rows need spelling out: media and URL keys.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;An image is global. Its alt text is not.&lt;/strong&gt; Adobe states that “New
product images are &lt;strong&gt;&lt;em&gt;always&lt;/em&gt;&lt;/strong&gt; uploaded and visible in &lt;strong&gt;&lt;em&gt;all&lt;/em&gt;&lt;/strong&gt;
store views, even if the &lt;code&gt;All Store Views&lt;/code&gt; scope is not used for
upload”
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-admin/catalog/products/digital-assets/product-image&quot;&gt;product images and video&lt;/a&gt;,
checked August 4, 2026), and the gallery stores one file per image
with no store column. A DAM therefore owns the asset set for every
locale at once. What is per store view is the row beside each image:
label, position and a disabled flag, which makes alt text, image order
and which image is hidden all per-store-view fields. One caveat that
complicates the tidy split: the image &lt;em&gt;roles&lt;/em&gt; (base, small,
thumbnail, swatch) are store-view-scoped attributes holding a file
path, so which of the global files a storefront renders can itself
differ per locale. Shopify draws the line in the same place from the
other direction: on a media image the only translatable field is
&lt;code&gt;alt&lt;/code&gt;. Its
&lt;a href=&quot;https://help.shopify.com/en/manual/international/translate-adapt-app&quot;&gt;Translate &amp;#x26; Adapt documentation&lt;/a&gt;
lists product images among the content types the app cannot translate,
while other images and videos on a translated resource can be swapped
per language by uploading an alternate file (checked August 4, 2026).&lt;/p&gt;
&lt;p&gt;That split matters more than it looks, because alt text is not a
technical field. The
&lt;a href=&quot;https://eur-lex.europa.eu/eli/dir/2019/882/oj&quot;&gt;European Accessibility Act&lt;/a&gt;
covers e-commerce services sold to consumers, and Member States have
had to apply its measures since June 28, 2025. The Directive itself
requires those services to be perceivable, operable, understandable
and robust rather than naming alt text. The concrete
text-alternative requirement is expressed by EN 301 549, which points
at WCAG. Note that the version in force, V3.2.1, was harmonized under
the Web Accessibility Directive rather than under the Accessibility
Act, and the EAA-facing V4.1.0 was still not cited in the Official
Journal when we checked. Its microenterprise exemption is
&lt;a href=&quot;/blog/accessibility-act-ecommerce/&quot;&gt;narrower than it looks&lt;/a&gt;, and a
separate disproportionate-burden assessment is open to any provider. Alt
text is per-language content with a
compliance regime attached, authored wherever the rest of the copy is
authored. Nothing in either platform’s upload path prompts for it
per locale.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The URL key usually belongs to the store, but not because nothing
else claims it.&lt;/strong&gt; Most ERPs have no opinion about storefront URLs,
and the temptation is to conclude the field has no competing
claimant. It does. Comarch ERP Optima’s own e-Sklep documentation
gives the ERP a Pozycjonowanie (positioning) tab holding the page
title, a dedicated &lt;code&gt;Link&lt;/code&gt; field for building URLs, keywords and a
meta description, for product groups and individual items alike
(&lt;a href=&quot;https://pomoc.comarch.pl/optima/pl/2026_5/dokumentacja/wspolpraca-z-comarch-e-sklep/&quot;&gt;Comarch Optima help, version 2026_5&lt;/a&gt;,
checked August 4, 2026). Connectors and PIMs write handles too,
whatever the ERP core documents.&lt;/p&gt;
&lt;p&gt;The operational rule is better than the scope argument, because
Comarch shows an ERP can express the field. Whoever owns the URL key
owns every consequence of changing it, and those consequences land
outside the system that made the change. The decision that matters is
therefore what happens on rename. Magento makes that a setting,
“Create Permanent Redirect for URLs if URL Key Changed,” and whoever
owns the key inherits a growing rewrite table.&lt;/p&gt;
&lt;p&gt;Letting an external system churn that field has a direct cost.
Heureka’s
&lt;a href=&quot;https://sluzby.heureka.cz/napoveda/xml-feed/&quot;&gt;feed specification&lt;/a&gt;
marks four elements unconditionally mandatory, product name, item ID,
price with VAT and URL, and states, in Czech, that when product URLs,
names or categories change in the XML all products are unpaired from
their Heureka product cards and then wait to be paired again, which
it says can take about four business days. A PIM that regenerates handles
from product names will
therefore unpair the whole catalog from the comparison engine for
several days, without anything in the store appearing broken.&lt;/p&gt;
&lt;h2 id=&quot;where-the-erp-stops-per-locale-copy&quot;&gt;Where the ERP stops: per-locale copy&lt;/h2&gt;
&lt;p&gt;For a store selling in Czech, Slovak, German and Polish, the
dividing line between “the ERP can own product copy” and “something
else must” is usually the number of languages the ERP’s data model
admits, and it is lower than the number of markets a store of this
size sells into.&lt;/p&gt;
&lt;p&gt;ABRA Flexi owns multilingual copy in its core price-list entity, but
as a fixed number of columns rather than a list: one primary name and
description plus three more, visible in
&lt;a href=&quot;https://demo.flexibee.eu/c/demo/cenik/properties.json&quot;&gt;the entity’s own published field schema&lt;/a&gt;
(checked August 4, 2026, where the demo company labels them English,
German and French). That is four language slots in the
core entity, and ABRA’s own support article confirms more languages
are possible only by adding per-language custom fields, which is
integration work rather than configuration. Money exposes localized
names and
descriptions through the
&lt;a href=&quot;https://developer.moneyerp.com/docs/eshop-konektor/vycitani-dat-e-shopem/&quot;&gt;PLUS views of its e-commerce connector&lt;/a&gt;,
and Comarch ERP Optima carries per-language names and descriptions on
the same e-Handel tab as its SEO fields. Business Central exports item
translations to Shopify one way, and was the only system in this group
whose documentation described a custom-field bridge, mapping its
fields to Shopify metafields. For SAP Business One and InsERT we found
no documented multi-locale product-copy export, though SAP’s
documentation portal renders only to a JavaScript client and we could
not read all of it, so treat that as unconfirmed rather than absent.&lt;/p&gt;
&lt;p&gt;Two more ERP-side limits break integrations quietly, because both
are hard ceilings rather than errors. ABRA Flexi encodes quantity
tiers as fixed columns rather than rows, capping volume pricing at
five tiers. Helios expresses each price level in at most six
currencies, one domestic and five foreign. A tier grid or currency
list that outgrows the ERP’s
shape stops being expressible in the system of record, without
raising an error.&lt;/p&gt;
&lt;p&gt;Worth finding before anything else: the flag that decides which SKUs
the storefront may see. Every system has one and each names it
differently, &lt;code&gt;isInternet&lt;/code&gt; on a
&lt;a href=&quot;https://www.stormware.cz/schema/version_2/stock.xsd&quot;&gt;POHODA stock card&lt;/a&gt;,
&lt;code&gt;exportNaEshop&lt;/code&gt; on an ABRA Flexi price-list item, documented in its
&lt;a href=&quot;https://demo.flexibee.eu/c/demo/cenik/properties.json&quot;&gt;published field schema&lt;/a&gt;
and used as the filter in ABRA’s own integration guide.
Whoever owns that flag controls the catalog, whatever the rest of
the design says. Helios settles the direction of authority in its API
surface instead: of the seventeen paths in its
&lt;a href=&quot;https://public.helios.eu/inuvio/api/eshop/Inuvio_doc_api_eshopv2.json&quot;&gt;published eShop API description&lt;/a&gt;,
only orders and customers accept writes. Products, categories, price
lists and inventories are read-only, which makes the ERP
authoritative by construction.&lt;/p&gt;
&lt;h2 id=&quot;the-fields-the-law-added&quot;&gt;The fields the law added&lt;/h2&gt;
&lt;p&gt;Some of the contested fields are new, and they arrived from
regulation rather than from merchandising, which is why no system
was designed to own them.&lt;/p&gt;
&lt;p&gt;The
&lt;a href=&quot;https://eur-lex.europa.eu/eli/reg/2023/988/oj&quot;&gt;General Product Safety Regulation&lt;/a&gt;
(Regulation (EU) 2023/988, applicable since December 13, 2024)
requires that a distance-sales offer carry the manufacturer’s name
and both postal and electronic address, the same for a non-EU
manufacturer’s responsible person, information identifying the
product including a picture and its type, and any warning or safety
information “in a language which can be easily understood by
consumers, as determined by the Member State in which the product is
made available on the market.” That last clause makes safety text a
per-market content field, and the language is the destination
Member State’s call rather than yours. The regulation also requires
an EU-established economic operator responsible for each product.&lt;/p&gt;
&lt;p&gt;Energy labeling adds more.
&lt;a href=&quot;https://eur-lex.europa.eu/eli/reg/2017/1369/oj&quot;&gt;Regulation (EU) 2017/1369&lt;/a&gt;
names online distance selling explicitly in the dealer’s duty to
display the label (Article 5(1)(a)). The product information sheet has
to be made available to customers under the following point, without
that qualifier. Article 6(a) separately requires a visual
advertisement for a specific model to carry both the efficiency class
and the range of classes available on the label. Its EPREL database has
required suppliers to register each new model since January 1, 2019.
The consequence reaches the feed:
&lt;a href=&quot;https://support.google.com/merchants/answer/7052112&quot;&gt;Google Merchant Center&lt;/a&gt;
has required an EPREL registration code in the certification
attribute for affected EU-targeted products since April 2025, and
now limits its energy efficiency class attribute to Switzerland,
Norway and the United Kingdom.&lt;/p&gt;
&lt;p&gt;The channels have already modeled these fields.
&lt;a href=&quot;https://developer.allegro.pl/documentation&quot;&gt;Allegro’s API&lt;/a&gt; carries
responsible persons and responsible producers as their own
resources, per-offer safety information and a flag for products
marketed before the GPSR obligation. Heureka documents dedicated
manufacturer postal and electronic address tags. So these fields
need a system of record that can attest to them per market and per
language, and often no existing system holds one: an ERP records a
supplier rather than a responsible person, and a storefront records
copy rather than an attestation. The practical placement is a
metafield or attribute set per market on the storefront, filled at
supplier onboarding rather than at listing time, with the empty-field
check applied to
new and edited products rather than retroactively. Run it as a gate on
new listings and as a report on the existing catalog: applied to a
catalog of thousands of SKUs from dozens of suppliers, a hard block
unpublishes the store on the day it is switched on.&lt;/p&gt;
&lt;h2 id=&quot;b2b-pricing-is-a-different-mechanism-on-each-platform&quot;&gt;B2B pricing is a different mechanism on each platform&lt;/h2&gt;
&lt;p&gt;Adobe puts B2B pricing in shared catalogs, which are part of the
Adobe Commerce B2B module and therefore unavailable on Magento Open
Source or Mage-OS. Enabling them adds a Shared Catalog option to
the Customer Group selector on the product’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-admin/catalog/products/pricing/pricing-advanced&quot;&gt;Advanced Pricing&lt;/a&gt;
page, so the same tier grid is edited from either side.&lt;/p&gt;
&lt;p&gt;Shopify routes B2B through the same catalog and price-list machinery
as markets, with the catalog scoped to a company location. The plan
limits are the constraint. Shopify’s
&lt;a href=&quot;https://help.shopify.com/en/manual/b2b/catalogs&quot;&gt;B2B catalog documentation&lt;/a&gt;
states that on the Basic, Grow and Advanced plans “you can assign up
to 3 active catalogs across all your B2B markets,” while Plus offers
an unlimited number and direct assignment to companies and locations
(checked August 4, 2026). Where several catalogs apply to one company
location, the lowest price is the one displayed. Microsoft’s
&lt;a href=&quot;https://learn.microsoft.com/en-us/dynamics365/business-central/shopify/synchronize-items&quot;&gt;Business Central connector documentation&lt;/a&gt;
states that its B2B price synchronization requires a Shopify Plus
subscription, and that catalog membership is managed in Shopify’s
admin rather than in the ERP.&lt;/p&gt;
&lt;p&gt;The ownership rule that follows is the same on both: the ERP owns
what a given customer has been contracted to pay, and the platform
owns which catalog that customer is attached to. Splitting it any
further puts contract terms in a merchandising tool.&lt;/p&gt;
&lt;h2 id=&quot;the-case-for-a-pim&quot;&gt;The case for a PIM&lt;/h2&gt;
&lt;p&gt;A PIM is worth its cost at the point where per-locale, per-channel
values stop fitting in the platform. Both leading options model that
directly, and their licensing differs enough to decide the choice
before the features do.&lt;/p&gt;
&lt;p&gt;Akeneo’s mechanism is two independent flags on each attribute,
localizable and scopable. A product value is a triple of locale,
scope and data, and a channel bundles the locales, currencies,
category tree and conversion units for one publication target. Its
Community Edition is open source under OSL-3.0 and actively
maintained, with release v2026.3 tagged on March 30, 2026, and Akeneo
scopes its end-of-support table to the paid editions rather than to
Community. The caveat is upgrades: Community switched to
calendar versioning and Akeneo disclaims backwards compatibility
between any two releases. Commercially it sells three SaaS tiers, and
its comparison page
renders pricing through client-side scripting rather than as static
text, so we do not quote a figure from it. Third-party price figures
circulate for Akeneo. None is corroborated on an Akeneo page and we
do not repeat them.&lt;/p&gt;
&lt;p&gt;Pimcore, based in Salzburg, changed license in a way that matters
directly to this article’s reader.
&lt;a href=&quot;https://pimcore.com/en/resources/blog/breaking-free-pimcore-says-goodbye-to-gpl-and-enters-a-new-era-with-pocl&quot;&gt;Since Platform Version 2025.1 it is no longer open source&lt;/a&gt;:
the publicly available code is under the Pimcore Open Core Licence,
a source-available license, and its own &lt;code&gt;composer.json&lt;/code&gt; on the
current branch declares the license as proprietary. The
&lt;a href=&quot;https://raw.githubusercontent.com/pimcore/pimcore/2026.x/LICENSE.md&quot;&gt;license text&lt;/a&gt;
grants free production use only to organizations whose total global
revenue does not exceed EUR 5 million annually, self-certified and
subject to audit with retroactive fees, forbids offering Pimcore as
a hosted or managed service, forbids running GPLv3-licensed Pimcore
alongside it, and includes a telemetry clause the licensee agrees
not to interfere with. Its
&lt;a href=&quot;https://pimcore.com/en/pricing&quot;&gt;published list prices&lt;/a&gt; are $9,900 a
year for Professional and $29,900 for Enterprise, both on-premises,
with PaaS quoted. For a store above EUR 5 million in revenue,
Pimcore is a commercial license purchase.&lt;/p&gt;
&lt;h2 id=&quot;the-procedure&quot;&gt;The procedure&lt;/h2&gt;
&lt;p&gt;Write the field list before choosing any tool. For each field, name
the system that creates and maintains the value, the scope at which
the platform can store it, and what happens when the two sides
disagree. Direction follows from the first of those: the system of
record writes and everything else reads, which is why a field with
two writers has no owner. Then check three things that catch most
designs: that the scope you assumed exists (price per store view
does not on Magento), that the owner’s data model can hold the range
you need (five currencies, five tiers, three extra locales), and
that nothing else writes the same field behind your back. Bidirectional
editing of one field in two systems is the design to avoid, and it
is usually reached by omission rather than by choice.&lt;/p&gt;
&lt;p&gt;Then decide the propagation, remembering that the first load and the
steady state are different systems. The initial catalog load is a
bulk job: Shopify’s bulk operations take JSONL and run asynchronously,
and Magento has async bulk REST and the CSV importer. The steady state
is a delta, and the mechanism is dictated by what the owner can emit.
An ERP that offers only a nightly file export cannot push, however the
design is drawn, so the delta becomes a scheduled pull with a
changed-since filter, and the storefront’s freshness is bounded by the
export interval whatever anyone promised. Reconciliation is the third
job, not an optional one: a periodic full comparison that finds what
the delta missed, because a feed nobody audits drifts silently.&lt;/p&gt;
&lt;p&gt;On Shopify one more rule follows from how
&lt;a href=&quot;https://shopify.dev/docs/apps/build/product-merchandising/products-and-collections/sync-data&quot;&gt;&lt;code&gt;productSet&lt;/code&gt; synchronizes&lt;/a&gt;:
within a list field you send, it deletes the entries you leave out, so
a partial variant list removes the variants it omits. Use it only
where the sending system owns the whole of every list it sends, and
use the targeted mutations where ownership is split.&lt;/p&gt;
&lt;p&gt;Mapping that field list, and building the integrations that carry
it, is what our
&lt;a href=&quot;/services/integrations/&quot;&gt;integrations service&lt;/a&gt; covers, and the
custom work above the platform’s data model is
&lt;a href=&quot;/services/custom-development/&quot;&gt;custom development&lt;/a&gt;. Whichever tools
you choose, the field list is what the integration is built from.&lt;/p&gt;</content:encoded><category>integrations</category><category>erp</category><category>pim</category><category>architecture</category></item><item><title>Choosing hosting for Magento 2</title><link>https://zapolu.com/blog/magento-hosting/</link><guid isPermaLink="true">https://zapolu.com/blog/magento-hosting/</guid><description>Adobe&apos;s cloud, managed Magento specialists, and your own AWS or Hetzner: who runs which layer, who keeps up with the platform&apos;s requirements, what the SLAs say, and what is published about pricing, checked against the vendors&apos; own pages.</description><pubDate>Tue, 04 Aug 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Magento hosting is sold as plans and specs, but the decision
underneath is about responsibility. Somebody has to patch the
operating system, upgrade OpenSearch when Adobe’s requirements move,
keep cron and the queue consumers running, and answer when the store
is down. Every hosting tier is a different answer to who that
somebody is.&lt;/p&gt;
&lt;p&gt;This post lays out what Magento demands from infrastructure, gives a
checklist for judging a host, and applies it to the three tiers:
Adobe’s own cloud, the managed Magento specialists (including the
Czech and Polish providers your region offers), and self-managed
infrastructure on AWS or Hetzner. Every vendor fact was checked
against the linked source on August 3, 2026. Where something could
not be verified, we say so.&lt;/p&gt;
&lt;h2 id=&quot;what-magento-demands-from-a-server&quot;&gt;What Magento demands from a server&lt;/h2&gt;
&lt;p&gt;Adobe publishes a
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/installation-guide/system-requirements&quot;&gt;tested-combinations table&lt;/a&gt;
per release line and is explicit about its scope: “Adobe supports
only the system requirement combinations listed in the following
tables.” For Magento 2.4.9 (released May 12, 2026, per the
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/versions&quot;&gt;released-versions page&lt;/a&gt;),
the on-premises table lists OpenSearch 3, MariaDB 12.3, MySQL 8.4,
PHP 8.5, RabbitMQ 4.3, Valkey 9, Varnish 8 and nginx 1.30. The
table for 2.4.8-p5 stays one step behind: PHP 8.4 and 8.3, with
Elasticsearch 8 still listed beside OpenSearch 3, and Valkey 8.1.
Three shifts in those tables decide whether a host is current:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;OpenSearch is required. Elasticsearch is being removed from the
tested table.&lt;/strong&gt; “As of Adobe Commerce 2.4, all installations must
be configured to use Elasticsearch or OpenSearch as the catalog
search solution”
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/installation-guide/prerequisites/search-engine/overview&quot;&gt;search engine prerequisites&lt;/a&gt;),
and the 2.4.9 table lists only OpenSearch 3, with no Elasticsearch
row at all. Elasticsearch 7.17 reached end of support on
January 15, 2026, per Adobe’s system-requirements page.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Valkey replaced Redis.&lt;/strong&gt; The 2.4.9 and 2.4.8-p5 tables carry no
Redis row at all, the latest patches of the 2.4.5 through 2.4.7
lines mark Redis “Not supported” with Valkey listed instead, and
Adobe’s cache documentation states: “Starting with Adobe Commerce
2.4.9, Valkey officially replaced Redis in CLI tooling”
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/configuration-guide/cache/redis/redis-pg-cache&quot;&gt;Redis page cache doc&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;MySQL 8.0 reached end of support on April 30, 2026&lt;/strong&gt;, and
Adobe’s system-requirements page advises the affected release
lines to move to MariaDB.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Beyond the table, two operational requirements rule out generic
shared hosting.
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/configuration-guide/cli/configure-cron-jobs&quot;&gt;Cron is mandatory&lt;/a&gt;:
“Commerce depends on proper cron job configuration for many
important system functions, including indexing. Failure to set it up
properly means that Commerce will not function as expected,” and the
list of cron-dependent functions includes all transactional email.
Queue consumers are worker processes that cron
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/configuration-guide/message-queues/manage-message-queues&quot;&gt;restarts in batches by default&lt;/a&gt;.
Keeping them healthy is the host’s or the team’s job. RabbitMQ is
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/installation-guide/prerequisites/message-brokers/rabbitmq&quot;&gt;optional in general&lt;/a&gt;
and
&lt;a href=&quot;https://developer.adobe.com/commerce/webapi/rest/use-rest/bulk-endpoints/&quot;&gt;the default for the Bulk API&lt;/a&gt;.
A host that offers PHP and MySQL but cannot run OpenSearch, a
Valkey-backed cache, Varnish and several worker processes cannot run
Magento the way Adobe tests it, whatever the plan page says. Mage-OS
publishes
&lt;a href=&quot;https://mage-os.org/get-started/system-requirements&quot;&gt;its own requirements&lt;/a&gt;
with looser minimums (PHP 8.3 and up, OpenSearch 2 and up, with
Elasticsearch and Redis still listed), but the shape of the stack is
the same: a search engine, cron and worker processes stay mandatory
whichever distribution you run.&lt;/p&gt;
&lt;h2 id=&quot;the-checklist-who-runs-what-and-what-is-in-writing&quot;&gt;The checklist: who runs what, and what is in writing&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;Who runs which layer.&lt;/strong&gt; The three tiers split the work
differently. On Adobe’s PaaS, Adobe runs the infrastructure and the
merchant still “manages application code, upgrades, patching,
infrastructure configuration within Adobe’s hosted environment”
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce/cloud-service/migration/overview&quot;&gt;Adobe’s migration guide&lt;/a&gt;).
On a managed specialist, the host runs the stack and the merchant’s
team or agency runs Magento itself. On self-managed infrastructure,
AWS’s
&lt;a href=&quot;https://aws.amazon.com/compliance/shared-responsibility-model/&quot;&gt;shared responsibility model&lt;/a&gt;
is explicit: customers deploying EC2 “are responsible for management
of the guest operating system (including updates and security
patches)”, and everything above it. Application-level work, Magento
upgrades, extensions, custom code, never transfers to any host on
any tier. The question is only where the line below it sits.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Whether the host keeps up with the platform.&lt;/strong&gt; The requirements
above move with every release cycle, and hosts lag them by
documented, checkable amounts. Compare the host’s published
component versions (PHP, OpenSearch, Valkey) against Adobe’s current
table. The profiles below record what each vendor states, and an
absent version statement is a reason to ask before signing. The lag
matters most when you are building new or upgrading to the newest
line. A host matching the tested combinations of the release line
you are on remains supported until that line’s end of support.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;What the SLA document itself says.&lt;/strong&gt; Several vendors advertise
uptime figures their own legal pages do not contain. The mismatches
we found are noted in the profiles. Read what the credit scheme is,
who verifies an outage, and on what deadline a claim must be filed.
The structural limit everywhere: the provider itself reports the
outage, announces its own maintenance, and its own logs decide the
claim. Credits are capped, at a month’s fees or less in the terms we
read. Each cap we found is noted below. A public status page with
visible incident history is still worth reading, and each one we
found is linked.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The operations behind the plan page.&lt;/strong&gt; Four questions separate a
Magento specialist from a generic managed host, and no vendor
profiled below publishes full answers to them. How does the plan
handle a sales peak: resize lead time, autoscaling, and what
peak-season capacity planning is included. How long does a restore
take on a production-sized dataset, and is database recovery
point-in-time or snapshot-only. The backup schedule is the easy
half. What does a deploy do to the live store: maintenance window or
atomic switch, and does staging match production. And on
multi-server plans, which node owns cron and the queue consumers,
and how are they kept away from web traffic. Ask all four. The
answers are rarely public.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Data location and the legal basis for US providers.&lt;/strong&gt; EU-to-US
transfers currently rest on the
&lt;a href=&quot;https://commission.europa.eu/law/law-topic/data-protection/international-dimension-data-protection/eu-us-data-transfers_en&quot;&gt;EU-US Data Privacy Framework&lt;/a&gt;,
adopted July 10, 2023. The EU General Court upheld it against the
Latombe challenge on September 3, 2025
(&lt;a href=&quot;https://iapp.org/news/a/european-general-court-dismisses-latombe-challenge-upholds-eu-us-data-privacy-framework/&quot;&gt;IAPP report&lt;/a&gt;),
and Latombe’s
&lt;a href=&quot;https://digitalpolicyalert.org/event/35459&quot;&gt;further appeal&lt;/a&gt;
to the Court of Justice (Case C-703/25 P, filed October 31, 2025)
was still pending when we checked in August 2026. A US-owned host’s
participation can be verified on the
&lt;a href=&quot;https://www.dataprivacyframework.gov/list&quot;&gt;official DPF list&lt;/a&gt;. The
practical check is simpler: does the vendor document, in writing,
which data center your store runs in? The profiles record who does.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The PCI split.&lt;/strong&gt; A hosting provider is a third-party service
provider under PCI DSS, and the Council’s 2016 third-party assurance
supplement is blunt: “The use of a TPSP, however, does not relieve
the entity of ultimate responsibility for its own PCI DSS
compliance …”
(&lt;a href=&quot;https://listings.pcisecuritystandards.org/documents/ThirdPartySecurityAssurance_March2016_FINAL.pdf&quot;&gt;PCI SSC supplement&lt;/a&gt;).
Ask the host for its service-provider Attestation of Compliance and
keep the matrix of which requirements it covers. The merchant side
of that split stays with you.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;How each tier prices.&lt;/strong&gt; Managed specialists sell fixed plans,
usage-billed configurations (JetRails), or, at MGT-Commerce, a
management fee plus your own cloud bill. Self-managed infrastructure
is metered. Adobe is sales-only. All figures below are net of VAT,
which EU businesses reverse-charge either way. The USD-billed
vendors add exchange costs and currency exposure that the
EUR-billed ones do not. The profiles quote published figures
verbatim. Where a vendor publishes none, that is stated, and we do
not estimate totals for metered setups.&lt;/p&gt;
&lt;h2 id=&quot;adobes-cloud-the-first-party-tier&quot;&gt;Adobe’s cloud: the first-party tier&lt;/h2&gt;
&lt;p&gt;Adobe Commerce on cloud infrastructure is the platform vendor’s
PaaS, running on AWS or, for Pro, Azure, per its
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-on-cloud/user-guide/architecture/pro-architecture&quot;&gt;Pro architecture doc&lt;/a&gt;.
Its
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-on-cloud/user-guide/architecture/security&quot;&gt;security guide&lt;/a&gt;
documents the bundled Fastly CDN with DDoS protection and WAF,
hourly production backups, and that production instances can run in
most AWS regions, with the customer specifying where production
resides. Adobe’s unified
&lt;a href=&quot;https://www.adobe.com/cc-shared/assets/pdf/legal/terms/enterprise/pdfs/unified-sla-actionabilityaddendum2025oct12.pdf&quot;&gt;SLA addendum&lt;/a&gt;
(effective October 12, 2025) defines the commitment as “99.99%
for the Adobe-hosted infrastructure of the Production
Environment”. Two constraints bound the tier.
Licensing: it is an Adobe Commerce product, and Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce/user-guides/product-solutions&quot;&gt;product-solutions page&lt;/a&gt;
classes open-source deployments as customer-managed and self-hosted.
Mage-OS, which Adobe’s documentation does not mention, has no path
onto it either. Pricing: not public anywhere. Figures circulating
online are third-party estimates and we do not repeat them.&lt;/p&gt;
&lt;p&gt;Adobe is also changing what this tier is. It introduced Adobe
Commerce as a Cloud Service (SaaS) in 2025 and publishes a
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce/cloud-service/migration/overview&quot;&gt;migration guide&lt;/a&gt;:
the SaaS locks the core application code and moves customization to
APIs and App Builder, and its promoted storefront runs on Edge
Delivery Services, mandatory for Luma stores while existing PWA and
headless storefronts can be kept. For a Luma store the move is a
re-architecture rather than a straight migration. Adobe’s own blog
&lt;a href=&quot;https://business.adobe.com/blog/continued-investment-support-adobe-commerce-on-cloud&quot;&gt;states the PaaS has no end-of-support plans&lt;/a&gt;
(the page would not load for us, so the statement comes from its
indexed text). The hard date sits elsewhere: from June 1, 2027,
Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/planning/lifecycle-policy&quot;&gt;lifecycle policy&lt;/a&gt;
lets it enforce version upgrades on cloud environments and
decommission instances on unsupported releases. Merchants who chose
the PaaS for stability should weigh that policy before renewing.&lt;/p&gt;
&lt;h2 id=&quot;the-managed-specialists&quot;&gt;The managed specialists&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;Hypernode&lt;/strong&gt; (Amsterdam, part of team.blue) is the most current
managed platform we checked. Its docs list PHP 8.5 as fully
supported
(&lt;a href=&quot;https://docs.hypernode.com/hypernode-platform/php/supported-php-versions-and-how-to-change-them-on-hypernode.html&quot;&gt;PHP versions doc&lt;/a&gt;),
and its changelog dates
&lt;a href=&quot;https://changelog.hypernode.com/release-10641-opensearch-3-3-available-on-hypernode/&quot;&gt;OpenSearch 3.3 to December 2025&lt;/a&gt;
and
&lt;a href=&quot;https://changelog.hypernode.com/release-10564-valkey-8-now-available-at-hypernode/&quot;&gt;Valkey 8 to October 2025&lt;/a&gt;,
with no Valkey 9 statement yet. Magento tooling is native:
&lt;code&gt;hypernode-deploy&lt;/code&gt; with a Magento 2 template, staging via Brancher
on the Falcon plans. Published
&lt;a href=&quot;https://www.hypernode.com/en/plans-and-prices/&quot;&gt;pricing&lt;/a&gt; runs from
EUR 129 per month for the smallest Falcon production node to
EUR 3,462 for the largest, excluding VAT. The AWS-based Eagle line
runs from EUR 672 to EUR 33,409. The caveats are operational. No
public SLA document with an uptime percentage was found. The figures
that exist sit in a
&lt;a href=&quot;https://docs.hypernode.com/about-hypernode/about-hypernode/which-cloud-providers-do-we-use.html&quot;&gt;docs page&lt;/a&gt;
attributed to the infrastructure providers, 99.9% for the Falcon
line and 99.95% for the AWS infrastructure behind Eagle. Free
support runs office hours (9:00 to 18:00 CET/CEST). Per its
&lt;a href=&quot;https://docs.hypernode.com/about-hypernode/support/emergency-support-outside-office-hours.html&quot;&gt;emergency-support doc&lt;/a&gt;,
out-of-hours help is free when the cause is a Hypernode outage and
otherwise billed EUR 100 to 200 by SLA level for the first hour,
then hourly at the same rate. The Falcon line runs in Belgium.
Frankfurt requires the Eagle line. The
&lt;a href=&quot;https://www.hypernode-status.com/&quot;&gt;status page&lt;/a&gt; is public with
visible incident history.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;maxcluster&lt;/strong&gt; (Paderborn, part of
&lt;a href=&quot;https://press.team.blue/228140-maxcluster-becomes-part-of-team-blue/&quot;&gt;team.blue since 2023&lt;/a&gt;)
runs managed failover clusters in German data centers and
&lt;a href=&quot;https://maxcluster.de/magento-hosting&quot;&gt;names Mage-OS and Hyvä support explicitly&lt;/a&gt;.
Its &lt;a href=&quot;https://maxcluster.de/sla&quot;&gt;SLA document&lt;/a&gt; guarantees “99,99%
Verfügbarkeit” of hardware and network as a yearly average (at
most 52 minutes down a year) and commits to a first response
within 30 minutes, around the clock, via its emergency line.
Routine support
and services work run office hours. Its
&lt;a href=&quot;https://maxcluster.de/en/roadmap&quot;&gt;public roadmap&lt;/a&gt; dates PHP 8.5 to
December 2025 and Valkey to November 2025, without a version number.
An explicit OpenSearch 3 statement was not found. Pricing is public
but lives in a
&lt;a href=&quot;https://maxcluster.de/preisliste&quot;&gt;German-language PDF price list&lt;/a&gt; dated
January 12, 2026. Clusters start at EUR 19 per month for the
smallest failover pair, though that entry cluster (1 CPU, 2 GB RAM)
is below what a Magento 2.4 stack needs, and a Magento-sized cluster
prices well above it, plus a managed-SLA tier that runs up to
EUR 350 per month (Enterprise). Per the same price list, hands-on
work beyond the SLA is metered (developer support EUR 30
per 15 minutes in office hours, EUR 50 outside them), and the
automatic
malware scan becomes a paid add-on for the lower tiers from
January 1, 2027.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;MGT-Commerce&lt;/strong&gt; (Berlin) sells a different shape: a management fee
on top of AWS infrastructure that runs in your own AWS account,
“billed separately to your own AWS account for full transparency”
(&lt;a href=&quot;https://www.mgt-commerce.com/&quot;&gt;homepage&lt;/a&gt;). That structure means
real cost transparency and portability, and it means two bills. The
stack covers Varnish, Valkey or Redis, OpenSearch or Elasticsearch
and RabbitMQ. The
&lt;a href=&quot;https://www.mgt-commerce.com/magento-hosting/&quot;&gt;hosting page&lt;/a&gt; states
PHP 8.1 to 8.4, with no PHP 8.5 platform statement yet even though
its own blog covers 2.4.9’s requirements. Management fees are
published (from EUR 149 per month for a single server, EUR 1,499 for
auto-scaling), though tier names and figures differ between the
homepage and the plan sub-pages. On availability, the homepage
advertises 99.99% and 99.95% at once, and the
&lt;a href=&quot;https://www.mgt-commerce.com/service-level-agreements&quot;&gt;SLA overview page&lt;/a&gt;
states a “99.95% network uptime guarantee” with no credit scheme,
exclusions or claim procedure. No public status page was found. The
“15 Min Response Time” line is the vendor’s marketing average. The
SLA page commits to under 24 hours for general guidance and
emergency responses from under 8 hours on the entry tier to under
30 minutes on the top auto-scaling tier.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;JetRails&lt;/strong&gt; (Illinois) is Magento-centric and provisions
single-tenant AWS environments through its AutoPilot portal, with
root access and hourly billing. Its December 2025
&lt;a href=&quot;https://jetrails.com/december-2025-autopilot-changelog/&quot;&gt;changelog&lt;/a&gt;
records OpenSearch 3.2 and default Magento provisioning at
“v2.4.8.2”. No PHP support matrix exists on its marketing pages or
changelogs. An article on its
&lt;a href=&quot;https://learn.jetrails.com/&quot;&gt;support portal&lt;/a&gt; documents switchable
PHP-FPM versions to 8.3. JetRails’ marketing claims 99.99% uptime
(in its pages’ search snippets), a figure its
&lt;a href=&quot;https://jetrails.com/terms-of-service/&quot;&gt;terms of service&lt;/a&gt; never
state: the terms promise “commercially reasonable efforts”, with
credits capped at two weeks per month, under Illinois law. Public
pricing is a range (“Our regular configurations range between
$100-$2,500 per month” on the
&lt;a href=&quot;https://jetrails.com/magento-hosting/&quot;&gt;Magento page&lt;/a&gt;). Exact quotes
require an account.
&lt;a href=&quot;https://jetrails.com/dedicated-servers/&quot;&gt;Dedicated servers&lt;/a&gt; are
stated as US data centers, and no marketing page documents
selectable EU regions (a JetRails Cloud FAQ says multi-region setups
are supported, without naming regions), which makes EU data
residency a question to settle in writing before signing. The
&lt;a href=&quot;https://jetrails-status.com/&quot;&gt;status page&lt;/a&gt; is public.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Nexcess&lt;/strong&gt;, one of the longest-standing Magento hosts, now sells
through Liquid Web: the
&lt;a href=&quot;https://www.liquidweb.com/magento-hosting/&quot;&gt;Magento plans&lt;/a&gt; run on
Nexcess infrastructure, from $74 per month at list for the smallest
plan (promotional pricing shown first), USD only. Every plan carries
Varnish, Redis and a dedicated OpenSearch container. The platform
provides staging, SSH and Git. The gap is in how current the stack
is and how clearly it is stated: the vendor’s own
&lt;a href=&quot;https://www.liquidweb.com/magento/latest-magento-version/&quot;&gt;reference page&lt;/a&gt;
still called 2.4.8 the latest Magento version when we checked, the
one reachable
&lt;a href=&quot;https://docs.nexcess.com/hosting/control-panel/whm/php-management/using-the-cloudlinux-php-selector-in-cpanel-and-whm/&quot;&gt;docs page&lt;/a&gt;
that enumerates PHP versions (for its cPanel line) stops at 8.3, and
no PHP, OpenSearch or Valkey statement was found for the Magento
plans themselves. The marketing page advertises “99.99% uptime
guaranteed” while the
&lt;a href=&quot;https://www.liquidweb.com/policies/sla/&quot;&gt;SLA page&lt;/a&gt; contains no such
figure. Its Nexcess section, which the plans’ infrastructure points
to (confirm at signing which scheme applies), promises 100%
uninterrupted transit, with credits of 5% of monthly fees for every
15 minutes of verified outage beyond the first 15 minutes in a
month, capped at a month’s fees, claims within seven days, verified
against the vendor’s own logs. EU-mainland presence is one location,
Amsterdam, per the &lt;a href=&quot;https://status.nexcess.net/&quot;&gt;status page&lt;/a&gt;.
Nearly two decades of Magento history and the brand churn since the
Liquid Web merger are both visible in its pages.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Cloudways&lt;/strong&gt; (part of DigitalOcean) is the budget managed option
smaller stores choose, and the checkable facts warrant caution. It
manages the server, and its
&lt;a href=&quot;https://www.cloudways.com/en/support.php&quot;&gt;scope of support&lt;/a&gt;
excludes the application: debugging custom code, application-level
security and module upgrades are out of scope. Its
&lt;a href=&quot;https://www.cloudways.com/en/changelog.php&quot;&gt;changelog&lt;/a&gt; dates
Magento 2.4.8 support to July 2025 and PHP 8.4 to September 2025,
and lists available search-engine versions topping out at
OpenSearch 2.19, while Adobe’s tested combinations for 2.4.8-p5 and
2.4.9 list only OpenSearch 3. There is no infrastructure SLA at all.
The only published SLA covers the support console and states it does
not apply to the cloud providers’ availability
(&lt;a href=&quot;https://www.cloudways.com/en/support-sla.php&quot;&gt;support SLA&lt;/a&gt;).
Data-center location is picked on the underlying provider. The
&lt;a href=&quot;https://www.cloudways.com/en/magento-hosting.php&quot;&gt;Magento hosting page&lt;/a&gt;
lists Amsterdam, Frankfurt and London among DigitalOcean’s
locations.
&lt;a href=&quot;https://www.cloudways.com/en/pricing.php&quot;&gt;Pricing&lt;/a&gt; is published and
genuinely low, from $11 per month on DigitalOcean. For a store that
fits inside one server and a team that owns Magento itself, it can
work on the release line you are on. Until Cloudways ships
OpenSearch 3, an upgrade to 2.4.8-p5 or 2.4.9 leaves Adobe’s tested
combinations. The specialization does not extend past provisioning.&lt;/p&gt;
&lt;h2 id=&quot;the-czech-and-polish-options&quot;&gt;The Czech and Polish options&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;vshosting~&lt;/strong&gt; (Prague, part of the Contabo Group per its
&lt;a href=&quot;https://vshosting.eu/about&quot;&gt;about page&lt;/a&gt;) is, by its own account,
the managed-infrastructure provider Czech and Slovak e-commerce uses
most. That self-description
(&lt;a href=&quot;https://vshosting.cz/blog/vshosting-nejvetsi-e-commerce-infrastructure-provider&quot;&gt;on its own site&lt;/a&gt;)
and its client lists are the vendor’s claims, with no independent
measurement we could find. What is documented: a
&lt;a href=&quot;https://vshosting.eu/references/takoy&quot;&gt;published Magento reference&lt;/a&gt;
for Takoy (“a hosting solution tailored to Magento”), its own Prague
data center plus local German data centers it uses, and
&lt;a href=&quot;https://vshosting.eu/services/managed-servers&quot;&gt;managed-server pricing&lt;/a&gt;
from EUR 217 per server per month, with the fine print that the
starting prices assume ten or more servers on 36-month contracts.
“Administrator Response Within 60 Seconds” by phone and 24/7
monitoring are the vendor’s wording. No per-plan Magento package
pricing is published, and the pages we checked state no SLA
percentage or stack versions.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Centuria&lt;/strong&gt; (Poznań) is the clearest Polish managed-Magento
specialist we found: a dedicated
&lt;a href=&quot;https://centuria.pl/en/magento-hosting/&quot;&gt;Magento hosting page&lt;/a&gt; with
24/7 service desk, named data centers (Equinix Frankfurt, Beyond and
Talex in Poznań) and testimonials from Castorama and CD PROJEKT
RED’s store (vendor-published). Its published stack names Varnish,
Redis and Elasticsearch. Elasticsearch 8 is still within Adobe’s
tested combinations for 2.4.8-p5, Redis is not, so the page’s
component list is worth a direct question. Pricing is indicative
only: monthly maintenance from about PLN 1,500 to 2,000 per its own
page.&lt;/p&gt;
&lt;p&gt;The search also settled which providers to rule out. WEDOS, whose
hosting now trades as VEDOS, answered in its
&lt;a href=&quot;https://help.wedos.cz/otazka/hosting-magento-2/2697/&quot;&gt;knowledge base&lt;/a&gt;
in 2019 that it recommends a VPS for Magento and calls running it on
their webhosting problematic, and we found no newer Magento guidance
there. Websupport
&lt;a href=&quot;https://www.websupport.sk/podpora/kb/instalacia-cms-magento-cez-webovu-konzolu/&quot;&gt;documents&lt;/a&gt;
only that Magento can be installed on shared hosting despite
limitations, and no Magento hosting page was found at Forpsi, Master
Internet or cyber_Folks. We found no Slovak-domiciled managed-Magento
specialist at all. Slovak stores in this tier buy from the Czech,
Polish, German or international providers here. OVHcloud’s Polish
&lt;a href=&quot;https://www.ovhcloud.com/pl/web-hosting/magento-hosting/&quot;&gt;Magento page&lt;/a&gt;
exists but recommends self-managed VPS or dedicated servers, which
places it with the self-managed options.&lt;/p&gt;
&lt;h2 id=&quot;self-managed-aws-and-hetzner&quot;&gt;Self-managed: AWS and Hetzner&lt;/h2&gt;
&lt;p&gt;Running Magento on raw cloud is a real option with a plain price:
your team becomes the managed host. On &lt;strong&gt;AWS&lt;/strong&gt;, the managed services
map onto part of the stack (Aurora for the database, OpenSearch
Service, ElastiCache, Amazon MQ), and Adobe’s requirements table
carries AWS-specific rows for exactly these, plus S3. Adobe’s
ElastiCache row lists “ElastiCache 7.1 for Redis OSS (enhanced).
Valkey 8 is available.” as the tested combination, and CloudFront
covers the CDN outside the table. Varnish, PHP-FPM, cron and the
queue workers stay on EC2 you administer, and a multi-AZ setup adds
Magento-specific work of its own: shared media storage, shared
sessions and cache, database failover, and a single owner for cron
and the consumers. What AWS does not have is a production-ready
first-party Magento path: its Magento
&lt;a href=&quot;https://docs.aws.amazon.com/whitepapers/latest/migrating-magento-open-source-adobe-commerce-to-aws/migrating-magento-open-source-adobe-commerce-to-aws.html&quot;&gt;whitepaper&lt;/a&gt;
carries the banner “This whitepaper is for historical reference
only”, the
&lt;a href=&quot;https://github.com/aws-ia/terraform-adobe-magento&quot;&gt;Terraform module&lt;/a&gt;
it links is labeled beta and not recommended for production (last
commit August 2025), the original Magento Quick Start page no longer
exists, and the
&lt;a href=&quot;https://github.com/amazon-archives/aws-refarch-magento&quot;&gt;reference architecture repository&lt;/a&gt;
has not been touched since April 2018 and sits archived. SLAs are
per-service and credit-only: 99.99% for EC2 requires a multi-AZ
deployment, and a single instance carries 99.5%
(&lt;a href=&quot;https://aws.amazon.com/compute/sla/&quot;&gt;compute SLA&lt;/a&gt;). Eight
&lt;a href=&quot;https://docs.aws.amazon.com/general/latest/gr/rande.html&quot;&gt;European regions&lt;/a&gt;
are available, six of them in EU member states, including Frankfurt.
The cost is metered across many line items. We quote no totals,
because a realistic total requires modeling your own traffic.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Hetzner&lt;/strong&gt; is the EU cost baseline, and it documents almost nothing
about Magento. There is no product page, one-click app or version matrix.
The closest thing is a
&lt;a href=&quot;https://community.hetzner.com/tutorials/install-magento-on-managed-server/&quot;&gt;community-written tutorial&lt;/a&gt;
for its Managed Server line. Hetzner’s own documentation has no
Magento page (checked on October 12, 2026). The tutorial documents the friction itself, a
self-installed OpenSearch that needs a support-approved Java
process, with no RabbitMQ documented. On the Managed Server line,
Magento remains a self-installed workaround with no product support
behind it. What Hetzner sells is capable, well-priced infrastructure
and nothing above it: cloud instances from EUR 5.99 per month with
an IPv4 address for 2 vCPU and 4 GB
(&lt;a href=&quot;https://www.hetzner.com/cloud/&quot;&gt;cloud page&lt;/a&gt;, where prices render
client-side), dedicated servers beyond, EU data centers in Germany
and Finland (US and Singapore locations exist too), free
&lt;a href=&quot;https://www.hetzner.com/unternehmen/ddos-schutz/&quot;&gt;network-level DDoS protection&lt;/a&gt;,
and a cloud SLA of “commercially reasonable efforts to ensure a
monthly availability of 99.9%”, with credits paid in Cloud Credits,
capped at that server’s monthly invoice and claimed within 14 days
(&lt;a href=&quot;https://www.hetzner.com/legal/cloud-server/&quot;&gt;cloud terms&lt;/a&gt;). On
Hetzner, the entire Magento stack, and the on-call duty, is yours. A
store on one of the managed plans above, from EUR 129 per month at
Hypernode or EUR 149 at MGT-Commerce, is partly paying for the
operational work this section describes.&lt;/p&gt;
&lt;h2 id=&quot;side-by-side&quot;&gt;Side by side&lt;/h2&gt;
&lt;div class=&quot;table-scroll&quot;&gt;




































































































&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;Option&lt;/th&gt;&lt;th&gt;Model&lt;/th&gt;&lt;th&gt;Uptime in the SLA&lt;/th&gt;&lt;th&gt;Current stack, as stated&lt;/th&gt;&lt;th&gt;EU locations&lt;/th&gt;&lt;th&gt;Public pricing&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;&lt;tr&gt;&lt;td&gt;Adobe cloud (PaaS)&lt;/td&gt;&lt;td&gt;First-party PaaS&lt;/td&gt;&lt;td&gt;99.99% (SLA addendum)&lt;/td&gt;&lt;td&gt;First-party&lt;/td&gt;&lt;td&gt;AWS/Azure, customer-chosen&lt;/td&gt;&lt;td&gt;No&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Hypernode&lt;/td&gt;&lt;td&gt;Managed&lt;/td&gt;&lt;td&gt;No SLA doc, 99.9–99.95% in docs&lt;/td&gt;&lt;td&gt;PHP 8.5, OpenSearch 3.3, Valkey 8&lt;/td&gt;&lt;td&gt;BE, plus DE via AWS line&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;maxcluster&lt;/td&gt;&lt;td&gt;Managed&lt;/td&gt;&lt;td&gt;99.99% yearly average&lt;/td&gt;&lt;td&gt;PHP 8.5 (roadmap), Valkey unversioned&lt;/td&gt;&lt;td&gt;DE&lt;/td&gt;&lt;td&gt;Yes (PDF)&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;MGT-Commerce&lt;/td&gt;&lt;td&gt;Fee + your AWS&lt;/td&gt;&lt;td&gt;99.95%, no credit terms&lt;/td&gt;&lt;td&gt;PHP to 8.4 stated&lt;/td&gt;&lt;td&gt;AWS regions, not pinned&lt;/td&gt;&lt;td&gt;Fees yes, AWS metered&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;JetRails&lt;/td&gt;&lt;td&gt;Managed on AWS&lt;/td&gt;&lt;td&gt;No percentage in terms&lt;/td&gt;&lt;td&gt;OpenSearch 3.2, PHP-FPM to 8.3 in support docs&lt;/td&gt;&lt;td&gt;Not documented&lt;/td&gt;&lt;td&gt;Range only&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Nexcess (Liquid Web)&lt;/td&gt;&lt;td&gt;Managed&lt;/td&gt;&lt;td&gt;100% transit, 5% credits per 15 min&lt;/td&gt;&lt;td&gt;Not stated, cPanel docs reach PHP 8.3&lt;/td&gt;&lt;td&gt;Amsterdam&lt;/td&gt;&lt;td&gt;Yes (USD)&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Cloudways&lt;/td&gt;&lt;td&gt;Server-managed&lt;/td&gt;&lt;td&gt;Support-only SLA&lt;/td&gt;&lt;td&gt;PHP 8.4, OpenSearch 2.19&lt;/td&gt;&lt;td&gt;Via chosen IaaS&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;vshosting~&lt;/td&gt;&lt;td&gt;Managed&lt;/td&gt;&lt;td&gt;Not found&lt;/td&gt;&lt;td&gt;Not published&lt;/td&gt;&lt;td&gt;CZ, DE&lt;/td&gt;&lt;td&gt;Entry prices&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Centuria&lt;/td&gt;&lt;td&gt;Managed&lt;/td&gt;&lt;td&gt;Not found&lt;/td&gt;&lt;td&gt;Names Redis, Elasticsearch&lt;/td&gt;&lt;td&gt;PL, DE&lt;/td&gt;&lt;td&gt;Indicative&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;AWS (self-managed)&lt;/td&gt;&lt;td&gt;IaaS&lt;/td&gt;&lt;td&gt;Per-service&lt;/td&gt;&lt;td&gt;Your job&lt;/td&gt;&lt;td&gt;6 EU regions&lt;/td&gt;&lt;td&gt;Metered&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Hetzner (self-managed)&lt;/td&gt;&lt;td&gt;IaaS&lt;/td&gt;&lt;td&gt;99.9% efforts-based&lt;/td&gt;&lt;td&gt;Your job&lt;/td&gt;&lt;td&gt;DE, FI&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;“Not found”, “not stated”, “not published” and “not documented” all
record our own searches of the linked pages. “Your job” means the
merchant installs and tracks the stack. “Current stack” compares
each vendor’s published component versions against Adobe’s
2.4.8-p5/2.4.9 tables (PHP 8.3–8.5 depending on the line, OpenSearch
3 or, on 2.4.8-p5 only, Elasticsearch 8, Valkey). All as of
August 3, 2026.&lt;/p&gt;
&lt;h2 id=&quot;choosing-between-the-tiers&quot;&gt;Choosing between the tiers&lt;/h2&gt;
&lt;p&gt;Start with the license. Magento Open Source and Mage-OS stores
cannot buy Adobe’s cloud, and their choice is between a managed
specialist and their own infrastructure. An Adobe Commerce license
opens all three tiers: the first-party cloud, with its SLA and its
2027 forced-upgrade policy, is one more option beside the
specialists, several of which host Adobe Commerce too.&lt;/p&gt;
&lt;p&gt;Between a specialist and your own cloud, the deciding variable is
your team. A managed specialist is worth its price when nobody on
staff should be tracking OpenSearch versions or holding the on-call
duty. The comparison above then narrows to who is current, who
writes their SLA down, and who documents your data center. On the
published evidence, Hypernode tracks the platform closest, with
maxcluster and JetRails current on parts of the stack and silent on
the rest. On location, maxcluster documents German data centers and
vshosting its own Prague site, while on MGT-Commerce the data sits in whichever
AWS region you run. Self-managed AWS or Hetzner wins when a capable
team already exists and the store needs either scale economics or a
topology no plan offers. It costs part of the team’s attention
permanently, and that cost appears on no invoice.&lt;/p&gt;
&lt;p&gt;Whichever way the comparison points, moving is its own project with
its own bill: syncing media at catalog scale, reindexing search on
the new stack, carrying over cron and queue state, TLS and DNS
cutover, and a freeze window that must miss the sales peak. Budget
the migration before comparing plan prices. A cheaper plan plus a
migration can cost more than the incumbent’s renewal.&lt;/p&gt;
&lt;p&gt;For the Czech, Slovak and Polish market specifically, the local
managed providers are ahead on language and data location, but they
publish less: no fixed Magento plans, indicative or conditional
pricing, no SLA percentages on the pages we checked, and no stack
versions to compare. Ask for the SLA, the price, the contract term
and the notice period in writing. A discount bound to a 36-month
term, as in vshosting’s entry pricing, is part of the price. Among
the internationals, maxcluster and Nexcess publish contractual SLA
terms with their prices, and most publish at least prices. A local
provider can be asked to match that in writing.&lt;/p&gt;
&lt;p&gt;Hosting is one of the three columns this practice’s
&lt;a href=&quot;/services/devops/&quot;&gt;DevOps service&lt;/a&gt; covers, and the same assessment
runs inside our &lt;a href=&quot;/services/magento/&quot;&gt;Magento work&lt;/a&gt;: which tier fits,
whether the current host still matches Adobe’s requirements table,
and what the migration would cost if not. Whoever hosts you,
tracking that table is a task someone must own.&lt;/p&gt;</content:encoded><category>magento</category><category>hosting</category><category>devops</category></item><item><title>Choosing a payment gateway for Magento 2</title><link>https://zapolu.com/blog/magento-payment-gateways/</link><guid isPermaLink="true">https://zapolu.com/blog/magento-payment-gateways/</guid><description>A decision framework for payment gateways on Magento 2, Adobe Commerce and Mage-OS: module maintenance, upgrade behavior, Hyvä and headless support, and the local payment methods that decide the shortlist in each market.</description><pubDate>Sun, 02 Aug 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Choosing a payment provider for a Magento store is usually framed as
a fee comparison. On Magento the fees are the smaller part of the
decision, because almost every provider decision carries a second
one: whose Magento module you will run. The module lives in your
codebase and renders inside your checkout, and a platform upgrade
either carries it along or stops on it. A payment module is the
third-party code closest to the money on every order.&lt;/p&gt;
&lt;p&gt;This post gives a checklist for choosing a gateway, then applies it
to six international providers (Adyen, Stripe, Mollie, PayPal with
Braintree, Klarna, Unzer) and four gateways specific to the Czech,
Slovak and Polish markets (GoPay, Comgate, PayU, Przelewy24). Every
vendor fact was checked against the linked source on August 2, 2026.
Where something could not be verified, we say so.&lt;/p&gt;
&lt;h2 id=&quot;what-to-check-before-you-sign&quot;&gt;What to check before you sign&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;Who maintains the module, and can you read the code?&lt;/strong&gt; Adyen,
Mollie, Unzer and PayU maintain their Magento modules in public
repositories. Stripe publishes only release archives, with the issue
tracker disabled. Klarna and Przelewy24 publish no source repository
at all, despite open-source license labels on their packages. GoPay
and Comgate have no module of their own. The modules on offer are
paid and closed, from third-party vendors. The bundled Braintree
extension’s current line has no public source either. Gene’s
repository stops at its legacy 4.0.x releases. A public repository
is not a quality guarantee, but it lets you read the diff before an
upgrade instead of relying on the changelog.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;How it has behaved across platform upgrades.&lt;/strong&gt; The next section
covers the three failure patterns worth searching for. Where the
vendor has no public tracker, and seven of the ten here do not,
search Magento’s own tracker for the module name: both incidents
below were recorded there, not in any vendor’s tracker.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The operational flows behind the checkout.&lt;/strong&gt; A redirect module
depends on its asynchronous confirmation: the webhook endpoint, its
signature checks and retries, and what happens to an order whose
customer never returns from the payment page. Refunds are the next
check. Some modules issue them from a Magento credit memo, others
only in the gateway’s back office, and the same split runs through
capture on shipment and orders created in the admin. Payment tokens
deserve their own question, because Magento-side subscription
extensions expect
&lt;a href=&quot;https://developer.adobe.com/commerce/php/development/payments-integrations/vault/&quot;&gt;Vault&lt;/a&gt;
while the local gateways below generally document their own
server-side tokenized recurring. Find out which of the two the
module exposes. In the
Czech Republic, Slovakia and Poland, a payment captured before
shipment can create an advance-invoice VAT obligation, one reason
merchants there configure capture on shipment. Confirm the workflow
with your accountant. The profiles below do not carry these facts.
Most vendors publish them only in module documentation and the
closed modules not at all. Run this check on the modules that make
your shortlist.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Hyvä support, as the public trackers record it.&lt;/strong&gt; Hyvä maintains a
public
&lt;a href=&quot;https://gitlab.hyva.io/hyva-public/module-tracker&quot;&gt;compatibility module tracker&lt;/a&gt;
for the theme and a separate
&lt;a href=&quot;https://gitlab.hyva.io/hyva-public/checkout-integration-tracker&quot;&gt;integration tracker&lt;/a&gt;
for Hyvä Checkout, and its documentation
&lt;a href=&quot;https://docs.hyva.io/hyva-checkout/integrations/available-payment-methods.html&quot;&gt;lists the primary tracked payment integrations&lt;/a&gt;.
Each ticket on the checkout integration tracker is labeled Vendor or
Community to show who is responsible for the implementation and
first-line support. Who that is matters as much as whether the
integration exists, because it decides who fixes it when the gateway
ships a breaking release.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Headless coverage.&lt;/strong&gt; Magento’s
&lt;a href=&quot;https://developer.adobe.com/commerce/webapi/graphql/schema/cart/mutations/set-payment-method&quot;&gt;&lt;code&gt;setPaymentMethodOnCart&lt;/code&gt; mutation&lt;/a&gt;
carries a method code plus a method-specific input object, which the
reference requires for every online payment method. A module without
its own GraphQL layer cannot receive tokenized payment data or
return redirect handling over GraphQL. A headless build then covers
the difference with custom work, usually over REST.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;PCI DSS scope.&lt;/strong&gt; Per the PCI Security Standards Council’s
&lt;a href=&quot;https://listings.pcisecuritystandards.org/documents/Understanding_SAQs_PCI_DSS_v3.pdf&quot;&gt;SAQ guidance&lt;/a&gt;
(a v3-era explainer whose redirect-versus-iframe distinction the
current v4 questionnaires keep), a full redirect or a
provider-hosted iframe can keep a merchant on SAQ A, the short
questionnaire. Payment forms or payment JavaScript served from your
own page mean SAQ A-EP. Under the v4.0.1 eligibility criteria,
explained in
&lt;a href=&quot;https://www.pcisecuritystandards.org/faqs/1588/&quot;&gt;FAQ 1588&lt;/a&gt;
(February 2025), even an iframe integration requires the merchant
page to be protected against script attacks, or the provider’s
confirmation that its solution protects it. A plain redirect stays
the smallest scope.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;The payment methods your market actually uses.&lt;/strong&gt; Every gateway
here processes cards. The deciding methods are invoice purchase in
Germany, BLIK depth in Poland, and in the Czech Republic whether the
transfer and cash-on-delivery share justifies a local gateway at
all. The market section below has the numbers and their sources.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Published pricing, in the currencies you settle.&lt;/strong&gt; Adyen, Stripe,
Mollie, PayPal, GoPay, Comgate, PayU and Przelewy24 publish price
lists. Unzer publishes entry-level rates and negotiates volume
pricing. Klarna publishes no European price list, and its total cost
cannot be calculated from public information. Check also whether the
gateway settles in each currency you sell in or converts at its own
rate. The conversion sits outside the headline fee. Much of this is
public:
&lt;a href=&quot;https://docs.adyen.com/account/supported-currencies&quot;&gt;Adyen lists 26 settlement currencies&lt;/a&gt;
for EU merchants including CZK and PLN,
&lt;a href=&quot;https://help.mollie.com/hc/en-us/articles/5546871220370-What-are-multi-currency-payouts&quot;&gt;Mollie lists 13 payout currencies&lt;/a&gt;,
&lt;a href=&quot;https://developers.payu.com/europe/docs/get-started/integration-overview/references/&quot;&gt;PayU settles 1:1 in 11&lt;/a&gt;,
and
&lt;a href=&quot;https://help.gopay.com/cs/tema/mam-platebni-branu/chci-pouzivat-gopay-obchodni-ucet/chci-prijimat-platby-v-dalsich-menach/v-jakych-menach-mohu-prijimat-platby-u-ceskych-a-slovenskych-bank&quot;&gt;GoPay documents payout currencies per bank&lt;/a&gt;.
&lt;a href=&quot;https://help.comgate.eu/docs/vyplaceni-penez-na-ucet&quot;&gt;Comgate publishes per-currency payout terms&lt;/a&gt;,
and
&lt;a href=&quot;https://developer.paypal.com/braintree/docs/reference/general/currencies&quot;&gt;Braintree publishes its scheme-currency list&lt;/a&gt;,
with each merchant’s set agreed at onboarding and extended by
&lt;a href=&quot;https://developer.paypal.com/braintree/articles/get-started/currencies&quot;&gt;requesting additional merchant accounts&lt;/a&gt;.
&lt;a href=&quot;https://docs.klarna.com/acquirer/klarna/web-payments/additional-resources/use-cases/consumer-fx/&quot;&gt;Klarna pays out in the transaction currency&lt;/a&gt;.
The rest publish less: Stripe renders its settlement table
client-side, Przelewy24 states payouts in PLN or EUR, and Unzer
publishes no list.&lt;/p&gt;
&lt;h2 id=&quot;what-upgrades-do-to-payment-modules&quot;&gt;What upgrades do to payment modules&lt;/h2&gt;
&lt;p&gt;Three failure patterns are documented in public trackers and release
notes, and all three are worth checking against any module you
shortlist.&lt;/p&gt;
&lt;p&gt;The first is unbundling. Magento’s 2.4.4 release notes state:
“With the exception of Braintree, all vendor-bundled extensions have
been removed from the Magento Open Source 2.4.4 code base”
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/notes/magento-open-source/2-4-4&quot;&gt;release notes&lt;/a&gt;).
Since then, payment modules version independently of the platform,
and a core upgrade no longer implies a compatible payment module.&lt;/p&gt;
&lt;p&gt;The second is dependency drift.
&lt;a href=&quot;https://github.com/magento/magento2/issues/39989&quot;&gt;Issue #39989&lt;/a&gt;
against Magento documents the security patch 2.4.8-p1 downgrading
six &lt;code&gt;paypal/module-braintree&lt;/code&gt; packages from 4.7.0 to 4.6.1-p5 on
affected installs, through a metapackage constraint that left
composer free to pick either version. In the other direction,
&lt;a href=&quot;https://github.com/magento/magento2/issues/39803&quot;&gt;issue #39803&lt;/a&gt;
records Klarna’s module blocking the 2.4.8 upgrade outright through
a &lt;code&gt;monolog&lt;/code&gt; version conflict. Adobe closed the issue in eleven days
by pointing stores to Klarna as the extension owner. The fix shipped
in module release 3.3.1: archived captures of the Marketplace
listing still show the pre-fix 3.3.0 on
&lt;a href=&quot;http://web.archive.org/web/20250520080628/https://commercemarketplace.adobe.com/klarna-m2-klarna.html&quot;&gt;May 20, 2025&lt;/a&gt;
and first show the fixed line on
&lt;a href=&quot;http://web.archive.org/web/20250613064417/https://commercemarketplace.adobe.com/klarna-m2-klarna.html&quot;&gt;June 13&lt;/a&gt;,
so stores with the module were blocked for six to nine weeks after
the April 8 GA.&lt;/p&gt;
&lt;p&gt;The third is frontend coupling. The
&lt;a href=&quot;https://mage-os.org/releases/2026-03-18-mage-os-2-2-1-release/&quot;&gt;Mage-OS 2.2.1 release notes&lt;/a&gt;
revert a reCAPTCHA deferred-loading optimization from 2.2.0 because
it caused reCAPTCHA failures on checkout pages using hosted and
iframe payment forms. Payment modules depend on checkout load order
and script timing, so an unrelated frontend change can break the
payment step at checkout. This is the same coupling that makes Hyvä
compatibility a separate deliverable.&lt;/p&gt;
&lt;p&gt;On Mage-OS specifically: its distribution ships the core PayPal
module but not Braintree (verified against the
&lt;code&gt;mage-os/product-community-edition&lt;/code&gt; metapackage), and none of the
ten providers below states Mage-OS support anywhere we could find.
Mage-OS itself
&lt;a href=&quot;https://mage-os.org/faq/&quot;&gt;claims full compatibility&lt;/a&gt; with Magento
Open Source extensions, so a module that runs on Open Source is
expected to run there. That expectation rests on Mage-OS’s claim,
with no payment vendor’s commitment behind it. It is the same for
all ten gateways and belongs to the platform decision, not to the
gateway choice.&lt;/p&gt;
&lt;h2 id=&quot;hyvä-changes-the-shortlist&quot;&gt;Hyvä changes the shortlist&lt;/h2&gt;
&lt;p&gt;A &lt;a href=&quot;/blog/hyva-vs-luma/&quot;&gt;Hyvä&lt;/a&gt; storefront changes the payment
question in one way. With the Hyvä theme alone, checkout can stay on
the stock Luma checkout through the documented
&lt;a href=&quot;https://docs.hyva.io/hyva-themes/building-your-theme/luma-theme-fallback.html&quot;&gt;theme fallback&lt;/a&gt;,
where existing payment modules keep working. The fallback is
configured per route and means maintaining a Luma-based checkout
theme alongside Hyvä. And because it applies only to the configured
routes, the express-payment buttons (Apple Pay, Google Pay, PayPal)
that modules render on product, cart and minicart pages remain on
the Hyvä theme and still need Hyvä compatibility from the module.
&lt;a href=&quot;https://docs.hyva.io/hyva-checkout/devdocs/payments/payment-in-hyva-checkout.html&quot;&gt;Hyvä Checkout&lt;/a&gt;
is different: order placement moves from the payment method to a
central service, so a payment module’s Luma frontend does not run
there. Per Hyvä’s docs the module’s backend can usually be reused,
and the checkout-facing layer must be built anew. Each provider
needs a dedicated integration. Short of building one, the documented
way around a missing integration is the Luma fallback.&lt;/p&gt;
&lt;p&gt;The maintenance arrangements behind those integrations are not
uniform, and Adyen shows why it matters. Adyen wrote its own Hyvä
Checkout module, then ended its maintenance. Its README now opens
with a notice that the integration is no longer officially supported
and will not get new updates
(&lt;a href=&quot;https://github.com/adyen-examples/adyen-magento2-hyva&quot;&gt;adyen-magento2-hyva&lt;/a&gt;).
The Hyvä team took the integration over and released a rewritten
package, &lt;code&gt;hyva-themes/magento2-hyva-checkout-adyen-payment-v2&lt;/code&gt;,
against Adyen’s v10 plugin in March 2026
(&lt;a href=&quot;https://github.com/adyen-examples/adyen-magento2-hyva/issues/140&quot;&gt;handover issue&lt;/a&gt;).
Merchants who chose Adyen partly for its first-party Hyvä support
now depend on the Hyvä-maintained package, on a plugin major version
that requires Magento 2.4.8 or newer.&lt;/p&gt;
&lt;h2 id=&quot;the-six-international-providers&quot;&gt;The six international providers&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;Adyen&lt;/strong&gt; maintains
&lt;a href=&quot;https://github.com/Adyen/adyen-magento2&quot;&gt;its Magento plugin&lt;/a&gt; in the
open under an MIT license and released it forty times across the
parallel v9 and v10 lines in the trailing twelve months, counted
from the repository’s releases page. The cost of that pace is a
strict support schedule:
&lt;a href=&quot;https://docs.adyen.com/plugins/adobe-commerce&quot;&gt;Adyen’s own table&lt;/a&gt;
gives each plugin major a fixed window (two years for v10), and
stores on 2.4.4 to 2.4.7 remain on the v9 line, in
security-updates-only mode since May 2026 and ending December 2026.
The v10 line added Magento 2.4.9 and PHP 8.5 support in
&lt;a href=&quot;https://github.com/Adyen/adyen-magento2/releases/tag/v10.10.0&quot;&gt;v10.10.0&lt;/a&gt;
(March 31, 2026), although the support table and the README still
said 2.4.8 only when we checked. The release notes are the current
record. Headless is covered through documented
&lt;a href=&quot;https://docs.adyen.com/plugins/adobe-commerce/headless-integration/&quot;&gt;GraphQL support&lt;/a&gt;,
and the EU method coverage
&lt;a href=&quot;https://docs.adyen.com/plugins/adobe-commerce/supported-payment-methods/&quot;&gt;includes iDEAL, SEPA Direct Debit, BLIK, Twint and Bancontact&lt;/a&gt;,
with cards rendered as embedded components inside the checkout.
Pricing is &lt;a href=&quot;https://www.adyen.com/pricing&quot;&gt;published as interchange++&lt;/a&gt;
with per-method fees. Plan for a plugin migration every one to two
years. Adyen’s README recommends exactly that.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Stripe&lt;/strong&gt; ships a capable module on a closed process. The
&lt;a href=&quot;https://github.com/stripe/stripe-magento2-releases&quot;&gt;official module&lt;/a&gt;
is distributed as release archives under Stripe’s own proprietary
license, with GitHub issues disabled: there is no public bug
tracker to consult before an upgrade. Releases arrive roughly
monthly, support covers Magento 2.3.7 to 2.4.x with fixes only from
module 4.5.x up
(&lt;a href=&quot;https://docs.stripe.com/use-stripe-apps/adobe-commerce/payments/install&quot;&gt;lifecycle table&lt;/a&gt;),
and both checkout flows, embedded Payment Element and redirect, are
documented as SAQ A eligible in Stripe’s docs, which also cover
&lt;a href=&quot;https://docs.stripe.com/use-stripe-apps/adobe-commerce/payments/custom-storefront&quot;&gt;REST and GraphQL for custom storefronts&lt;/a&gt;.
We unpacked the 4.6.4 release archive: the GraphQL schema and its
resolvers are in the source, and composer.json requires only PHP 7.4
or later.
Read the
&lt;a href=&quot;https://github.com/stripe/stripe-magento2-releases/blob/master/CHANGELOG.md&quot;&gt;changelog&lt;/a&gt;
before every upgrade: version 4.6.0 shipped breaking changes for
headless builds, and the 4.6.2 notes state that the stronger
authentication requirements introduced in 4.6 “were causing
decreased checkout conversions” before being relaxed. The Hyvä
Checkout integration is published under Hyvä’s own namespace.
Stripe’s changelog has mentioned Hyvä once. EU methods include SEPA
Direct Debit, iDEAL, Przelewy24 and Klarna. BLIK is absent from
&lt;a href=&quot;https://docs.stripe.com/use-stripe-apps/adobe-commerce/payments&quot;&gt;the module’s documented method list&lt;/a&gt;
yet present in the source, which maps the method and ships its icon.
Methods render through Stripe’s Payment Element and are switched on
from the Stripe Dashboard, so verify in a Polish test mode.
Per-country pricing is published, including a
&lt;a href=&quot;https://stripe.com/en-cz/pricing&quot;&gt;Czech price list&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Mollie&lt;/strong&gt; is the strongest combination of openness and Hyvä
alignment. The &lt;a href=&quot;https://github.com/mollie/magento2&quot;&gt;module&lt;/a&gt; is
public, vendor-maintained, and released roughly monthly. GraphQL
support sits in the source and covers the full headless placement
flow: it extends Magento’s own payment mutations and hands back the
Mollie redirect URL. Mollie maintains its own
&lt;a href=&quot;https://github.com/mollie/magento2-hyva-compatibility&quot;&gt;Hyvä theme&lt;/a&gt;
and
&lt;a href=&quot;https://github.com/mollie/magento2-hyva-checkout&quot;&gt;Hyvä Checkout&lt;/a&gt;
compatibility modules, announced a
&lt;a href=&quot;https://www.hyva.io/blog/news/hyva-commerce-mollie-partnership.html&quot;&gt;strategic partnership&lt;/a&gt;
with Hyvä in October 2024, and contributes financially to Mage-OS on
&lt;a href=&quot;https://opencollective.com/mage-os&quot;&gt;Open Collective&lt;/a&gt;, though it
states no technical Mage-OS support. The
&lt;a href=&quot;https://github.com/mollie/magento2/releases/tag/v3.0.0&quot;&gt;v3.0.0 release&lt;/a&gt;
in June 2026 was a breaking upgrade (Magento 2.4.5 and PHP 8.1
minimums, Orders API removed). Stores on older platforms stay on
the 2.x line with no stated end-of-support date. Method coverage is
Western and Central European (iDEAL, SEPA, BLIK, Przelewy24, EPS,
Twint), mostly as redirects to Mollie’s hosted page. We found no
Czech or Slovak bank-button method in its list. Pricing is
&lt;a href=&quot;https://www.mollie.com/pricing&quot;&gt;published&lt;/a&gt; per method, per
transaction, with no minimum costs.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;PayPal and Braintree&lt;/strong&gt; are two different products that ship
together. The core PayPal module is part of the platform in every
distribution including Mage-OS, with its source public inside
Magento itself. Braintree is the one extension still bundled with
Magento, developed by Gene Commerce for PayPal. The currently
bundled 4.x line has no public source or issue tracker
(&lt;a href=&quot;https://github.com/genecommerce/module-braintree-magento2&quot;&gt;Gene’s public repository&lt;/a&gt;
stops at the legacy 4.0.x line), so bugs are filed against Magento
itself. The platform resolves the Braintree version through an
extensions metapackage whose loose constraint is how the downgrade
in &lt;a href=&quot;https://github.com/magento/magento2/issues/39989&quot;&gt;issue #39989&lt;/a&gt;
happened. Newer versions ship on Adobe’s composer
repository between platform releases and can be required explicitly.
Magento 2.4.9 added BLIK to the Braintree extension, along with invoice
purchase for Germany via Ratepay
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/notes/magento-open-source/2-4-9&quot;&gt;2.4.9 release notes&lt;/a&gt;),
while
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-admin/stores-sales/payments/braintree&quot;&gt;Adobe’s Braintree page&lt;/a&gt;
lists SEPA/ELV Direct Debit as not yet supported. GraphQL support
ships as the separate &lt;code&gt;paypal/module-braintree-graph-ql&lt;/code&gt; package,
visible in the composer output of the same issue, and Adobe documents
the
&lt;a href=&quot;https://developer.adobe.com/commerce/webapi/graphql/payment-methods/braintree-vault&quot;&gt;GraphQL flows&lt;/a&gt;.
PayPal publishes Braintree pricing per country: the
&lt;a href=&quot;https://www.paypal.com/de/enterprise/paypal-braintree-fees&quot;&gt;German fee page&lt;/a&gt;
lists 1.9% plus EUR 0.30 for cards and third-party wallets at
standard-merchant rates, the same figure as its
&lt;a href=&quot;https://www.paypal.com/cz/enterprise/paypal-braintree-fees&quot;&gt;Czech&lt;/a&gt;
and
&lt;a href=&quot;https://www.paypal.com/pl/enterprise/paypal-braintree-fees&quot;&gt;Polish&lt;/a&gt;
pages, though with no rate for the Ratepay invoice method. The
legacy Payflow methods in the core module are
&lt;a href=&quot;https://developer.paypal.com/api/nvp-soap/&quot;&gt;marked legacy by PayPal&lt;/a&gt;
and available only in the United States, Canada, Australia and New
Zealand per
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-admin/stores-sales/payments/paypal/paypal-payflow-pro&quot;&gt;Adobe’s Payflow Pro page&lt;/a&gt;,
which also flags Payflow Pro as needing a third-party plugin for
PSD2. Treat them as absent in Europe. The Hyvä Checkout integrations
for PayPal and for Braintree are both Hyvä-namespace packages.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Klarna&lt;/strong&gt; sells its own methods (invoice, installments, pay now)
rather than a full acquiring stack, and in DACH that makes it hard
to avoid. Its
&lt;a href=&quot;https://docs.klarna.com/acquirer/klarna/get-started/data-requirements/puchase-countries-currencies-locales/&quot;&gt;purchase-country list&lt;/a&gt;
includes Germany, Austria, the Czech Republic, Slovakia and Poland.
The module record is mixed. Distribution is Marketplace-only with an
Apache 2.0 label but no public source or issue tracker, and the
&lt;a href=&quot;https://commercemarketplace.adobe.com/klarna-m2-klarna.html&quot;&gt;listing&lt;/a&gt;
contradicts its own release notes: the description still advertises
Klarna Checkout and Magento 2.4.4 support, while the release notes
record Klarna Checkout removed in 4.0.0 and support for 2.4.4 and
2.4.5 dropped in the 4.x line. The &lt;code&gt;monolog&lt;/code&gt; conflict above made the
module an upgrade blocker for six to nine weeks.
Klarna’s
&lt;a href=&quot;https://docs.klarna.com/platform/adobe-commerce/payments/klarna-payments-module/&quot;&gt;own docs&lt;/a&gt;
state GraphQL and headless support. There is no European price
list. The Marketplace listing says additional fees apply and points
its contact link at a signup page, not at published rates. The Hyvä
Checkout integration is another hyva-themes package, and its
&lt;a href=&quot;https://gitlab.hyva.io/hyva-public/checkout-integration-tracker/-/issues/12&quot;&gt;tracker ticket&lt;/a&gt;
labels it community-maintained. The 2026 cadence has been fast:
4.1.1 in January
(&lt;a href=&quot;http://web.archive.org/web/20260215102634/https://commercemarketplace.adobe.com/klarna-m2-klarna.html&quot;&gt;archived listing&lt;/a&gt;)
to 4.11.1 in July, on the live listing above.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Unzer&lt;/strong&gt;, the former heidelpay, carries the DACH method set under a
single acquiring contract:
&lt;a href=&quot;https://docs.unzer.com/payment-methods/&quot;&gt;its own secured invoice purchase and installments&lt;/a&gt;
for Germany, Austria and Switzerland, plus EPS, Twint and SEPA
Direct Debit. Cards render as Unzer-hosted embedded components,
several methods as redirects. The engineering signals are weaker.
The &lt;a href=&quot;https://github.com/unzerdev/magento2&quot;&gt;repository&lt;/a&gt; is public
under Apache 2.0 but has issues disabled. The module contains no
GraphQL support at all (we checked the full file tree), and the
vendor’s documentation contradicts itself on supported Magento
versions (2.4.5 and later on the
&lt;a href=&quot;https://docs.unzer.com/plugins/magento-2/magento2-install-plugin/&quot;&gt;installation page&lt;/a&gt;,
2.4.6 to 2.4.8 on the
&lt;a href=&quot;https://docs.unzer.com/plugins/magento-2/&quot;&gt;plugin overview&lt;/a&gt;). Its
three Hyvä repositories
(&lt;a href=&quot;https://github.com/unzerdev/magento2-hyva-checkout&quot;&gt;magento2-hyva-checkout&lt;/a&gt;
and siblings) carry no tags or releases and were last pushed in
December 2025, although the
&lt;a href=&quot;https://docs.unzer.com/plugins/magento-2-hyva-checkout/&quot;&gt;docs&lt;/a&gt;
state a Hyvä Checkout plugin 1.0.0 released February 2026. Unzer
&lt;a href=&quot;https://www.unzer.com/en/pricing-information/&quot;&gt;publishes entry-level rates&lt;/a&gt;
(from 1.50% plus EUR 0.20 per transaction, a monthly fee from EUR
19, and EUR 149 setup) and negotiates volume pricing individually.&lt;/p&gt;
&lt;h2 id=&quot;the-local-gateways-cz-sk-pl&quot;&gt;The local gateways: CZ, SK, PL&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;GoPay&lt;/strong&gt;, by
&lt;a href=&quot;https://www.gopay.com/cs/platebni-brana/&quot;&gt;its own figures&lt;/a&gt; one of
the largest Czech and Slovak gateways, does not maintain a Magento
module itself. Its
&lt;a href=&quot;https://www.gopay.com/en/integration/&quot;&gt;integration page&lt;/a&gt; lists
first-party modules for Shoptet, WooCommerce, Shopify, OpenCart and
PrestaShop. For Magento, its
&lt;a href=&quot;https://help.gopay.com/en/knowledge-base/integration-of-payment-gateway/essential-guide-to-integration&quot;&gt;help center&lt;/a&gt;
points to five third-party suppliers. Of the supplier pages we could
reach, every module is paid and closed-source, with no public
repository and no mention of Hyvä.
&lt;a href=&quot;https://maghos.com/gopay-payment-for-magento-2.html&quot;&gt;One&lt;/a&gt; supports
Magento only up to 2.4.6, and another (Artio, reachable from the
supplier list above) states compatibility only with Magento 2.2.x on
PHP 7.0. The fifth supplier’s store was unreachable when we checked,
and the
&lt;a href=&quot;https://web.archive.org/web/20211203/https://connect20.aveo-trade.cz/details=1/extension=atconnect/magento-two-gopay&quot;&gt;last archived version&lt;/a&gt;
of its module dates to February 2019.
Neither Hyvä tracker has a GoPay entry, no reachable supplier page
mentions GraphQL, and a search on the Adobe Commerce Marketplace
currently displays no GoPay extension. The gateway itself covers
what a Czech checkout needs
(&lt;a href=&quot;https://www.gopay.com/en/payment-methods/&quot;&gt;bank buttons, QR payments, wallets, deferred payment&lt;/a&gt;)
at &lt;a href=&quot;https://www.gopay.com/en/pricing/&quot;&gt;published pricing&lt;/a&gt;, through a
redirect or an embedded overlay depending on the supplier’s module.
Budget for the integration as custom work with a third-party vendor,
and review the module before it ships, starting with its webhook
handling.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Comgate&lt;/strong&gt; follows the same pattern: its own docs delegate the
Magento module to
&lt;a href=&quot;https://help.comgate.eu/docs/open-source-reseni&quot;&gt;Platiti.cz&lt;/a&gt;, the
modules on offer are paid and closed, the only public repository is
an &lt;a href=&quot;https://github.com/ppexxi/comgate_magento2&quot;&gt;unofficial one&lt;/a&gt;
inactive since 2021, neither Hyvä tracker has a Comgate entry, and
neither supplier page mentions GraphQL or headless support. The
gateway’s standard flow, a redirect to Comgate’s hosted page, keeps
the store in the smallest PCI scope. Comgate also offers embedded
modes, and at least one of the closed modules advertises them, so
confirm which flow the module you buy actually uses. The method list is broad for the region: cards,
wallets, BLIK, deferred payment via Twisto and Skip Pay, and what
&lt;a href=&quot;https://www.comgate.eu/cs/platebni-brana&quot;&gt;Comgate’s site&lt;/a&gt; describes
as a full range of banks in the Czech Republic, Slovakia and
Poland. On transfers and QR payments, the
gateway’s value over the self-generated QR code many Czech stores
already print on invoices is instant confirmation and automatic
order matching. Pricing is
&lt;a href=&quot;https://www.comgate.eu/cs/ceniky-platebni-brany&quot;&gt;published&lt;/a&gt; down to
the chargeback fee.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;PayU&lt;/strong&gt; is the one CEE gateway that maintains its Magento module
the way the international vendors do:
&lt;a href=&quot;https://github.com/PayU-EMEA/plugin_magento_24&quot;&gt;public repository&lt;/a&gt;,
Apache 2.0, vendor-maintained, Magento 2.4 only. The cadence is
uneven, with a nine-month gap between releases in 2025 and 2026,
then four releases in 2026. Hyvä Checkout support exists but is
third-party (Snowdog) and marked incomplete on the
&lt;a href=&quot;https://gitlab.hyva.io/hyva-public/checkout-integration-tracker/-/work_items/553&quot;&gt;official tracker&lt;/a&gt;,
with its last tagged version (2.0.0) from April 2024. The module has
no GraphQL support. Its feature list has no BLIK-specific method, so
BLIK runs inside the generic redirect, which
&lt;a href=&quot;https://developers.payu.com/europe/docs/payment-solutions/blik/&quot;&gt;PayU’s docs&lt;/a&gt;
describe as limited to standard pay-by-link. The in-checkout code
entry and one-click variants belong to its direct API. Card payment
is an on-site form. The remaining methods redirect. Polish pricing
is &lt;a href=&quot;https://poland.payu.com/commercial-offer/&quot;&gt;published&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Przelewy24&lt;/strong&gt; maintains its module itself, which sets it apart from
the Czech gateways, though without a public repository the
maintenance cannot be watched. The current module for Magento 2.4.4
and later is
&lt;a href=&quot;https://commercemarketplace.adobe.com/przelewy24-magento2-przelewy24.html&quot;&gt;distributed through the Marketplace&lt;/a&gt;
under a GPL-3.0 label with no public repository or issue tracker and
no per-version release dates. Older builds, a separate
GraphQL/PWA/VUE variant, and the vendor’s own Hyvä Checkout module
(v1.0.0, which the page dates March 24, 2026) are
&lt;a href=&quot;https://www.przelewy24.pl/en/download&quot;&gt;downloadable from its own site&lt;/a&gt;.
We unpacked both extras. The Hyvä module carries Magewire checkout
components for BLIK, cards, Google Pay and Apple Pay on top of
&lt;code&gt;hyva-themes/magento2-hyva-checkout&lt;/code&gt;. The GraphQL variant (its
composer manifest says 1.3.0, and the page dates it February 9, 2026) is
a server-side GraphQL module wired into Magento’s payment mutation,
with no PWA or Vue frontend code despite the label. Both extras are
absent from the Hyvä trackers, so tracker visibility and first-line
support sit with the vendor alone. Its Polish competitors TPay and
PayPo have Hyvä Checkout tracker entries
(&lt;a href=&quot;https://gitlab.hyva.io/hyva-public/checkout-integration-tracker/-/issues/549&quot;&gt;TPay&lt;/a&gt;,
&lt;a href=&quot;https://gitlab.hyva.io/hyva-public/checkout-integration-tracker/-/issues/550&quot;&gt;PayPo&lt;/a&gt;),
PayPo’s module Snowdog-maintained and TPay’s the vendor’s own. The
default flow is a redirect, with optional in-checkout card and BLIK
entry. Its documented BLIK support is the most complete among these
modules: the redirect, in-checkout code entry, and one-click BLIK,
all per the vendor’s Marketplace listing and release notes.
Standard pricing is
&lt;a href=&quot;https://www.przelewy24.pl/bramka-platnicza&quot;&gt;published&lt;/a&gt;.&lt;/p&gt;
&lt;h2 id=&quot;side-by-side&quot;&gt;Side by side&lt;/h2&gt;
&lt;div class=&quot;table-scroll&quot;&gt;




























































































&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;Gateway&lt;/th&gt;&lt;th&gt;Module maintainer&lt;/th&gt;&lt;th&gt;Public source&lt;/th&gt;&lt;th&gt;Hyvä Checkout&lt;/th&gt;&lt;th&gt;GraphQL&lt;/th&gt;&lt;th&gt;Public pricing&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;&lt;tr&gt;&lt;td&gt;Adyen&lt;/td&gt;&lt;td&gt;Vendor&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;td&gt;Hyvä namespace&lt;/td&gt;&lt;td&gt;Documented&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Stripe&lt;/td&gt;&lt;td&gt;Vendor&lt;/td&gt;&lt;td&gt;Archives only&lt;/td&gt;&lt;td&gt;Hyvä namespace&lt;/td&gt;&lt;td&gt;In source&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Mollie&lt;/td&gt;&lt;td&gt;Vendor&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;td&gt;Vendor&lt;/td&gt;&lt;td&gt;In source&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Braintree (bundled)&lt;/td&gt;&lt;td&gt;Gene for PayPal&lt;/td&gt;&lt;td&gt;No&lt;/td&gt;&lt;td&gt;Hyvä namespace&lt;/td&gt;&lt;td&gt;Platform package&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Klarna&lt;/td&gt;&lt;td&gt;Vendor&lt;/td&gt;&lt;td&gt;No&lt;/td&gt;&lt;td&gt;Hyvä namespace&lt;/td&gt;&lt;td&gt;Documented&lt;/td&gt;&lt;td&gt;No&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Unzer&lt;/td&gt;&lt;td&gt;Vendor&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;td&gt;Vendor, unversioned&lt;/td&gt;&lt;td&gt;No&lt;/td&gt;&lt;td&gt;Entry rates&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;GoPay&lt;/td&gt;&lt;td&gt;Third parties&lt;/td&gt;&lt;td&gt;No&lt;/td&gt;&lt;td&gt;None found&lt;/td&gt;&lt;td&gt;None stated&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Comgate&lt;/td&gt;&lt;td&gt;Third parties&lt;/td&gt;&lt;td&gt;No&lt;/td&gt;&lt;td&gt;None found&lt;/td&gt;&lt;td&gt;None stated&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;PayU&lt;/td&gt;&lt;td&gt;Vendor&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;td&gt;Third party, partial&lt;/td&gt;&lt;td&gt;No&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;Przelewy24&lt;/td&gt;&lt;td&gt;Vendor&lt;/td&gt;&lt;td&gt;No repository&lt;/td&gt;&lt;td&gt;Vendor, untracked&lt;/td&gt;&lt;td&gt;Separate build&lt;/td&gt;&lt;td&gt;Yes&lt;/td&gt;&lt;/tr&gt;&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;“Hyvä namespace” means the integration package is published under
&lt;code&gt;hyva-themes&lt;/code&gt;, with first-line responsibility per the checkout
tracker’s Vendor and Community labels. In the GraphQL column,
“Documented” means the vendor states support and we did not read the
code. “In source” means we read it in public code. “Platform
package” means it ships as a separate composer package alongside the
platform, with the flows documented by Adobe. Przelewy24’s GraphQL
support ships as a separate downloadable module. We unpacked it and
found a server-side GraphQL API with no frontend code despite its
PWA label. The Braintree row covers the bundled extension. The core
PayPal module’s source is public inside Magento itself. Cells
follow from the sources linked above. “None stated” and “None
found” record our own searches. All as of August 2, 2026.&lt;/p&gt;
&lt;h2 id=&quot;start-from-the-market-then-disqualify&quot;&gt;Start from the market, then disqualify&lt;/h2&gt;
&lt;p&gt;In Germany, the EHI Retail Institute’s payment study puts PayPal
at 28.5 percent of online retail revenue in 2024, invoice purchase
at 25.8 percent, direct debit at 17.3 percent and cards at 12.3
percent
(&lt;a href=&quot;https://www.ehi.org/presse/paypal-festigt-spitzenposition/&quot;&gt;EHI&lt;/a&gt;).
A German checkout therefore needs PayPal, an invoice product and
SEPA Direct Debit before it needs anything else. Braintree covers
PayPal and, on Magento 2.4.9, invoice purchase via Ratepay. Klarna
and Unzer both sell the invoice product. For direct debit the
options here are Unzer, Mollie, Adyen or Stripe, since Adobe’s
Braintree documentation lists SEPA Direct Debit as not yet
supported. The Ratepay route requires Magento 2.4.9, which Adyen’s
plugin has supported since v10.10.0. The pairing therefore works on
a current platform, and on 2.4.8 and below the invoice options are
Klarna and Unzer. None of the three carriers publishes a rate for
the invoice product itself: Braintree’s German fee page prices
cards and wallets with no Ratepay row, Klarna publishes no list at
all, and Unzer’s entry rates are generic “from” figures. For the
market’s second-largest method, the fee comparison runs on quotes.
Cards carried about an eighth of that revenue. A cards-only
checkout therefore gives up conversion against competitors that
offer the preferred methods, even where buyers could pay by card.&lt;/p&gt;
&lt;p&gt;In Poland, BLIK’s operator commissioned an EY report that puts BLIK
at about half of Polish online sales value in 2023, measured across
BLIK, cards and transfers
(&lt;a href=&quot;https://www.blik.com/media/2025_03_12_EY_BLIK_payments_and_economy_report.pdf&quot;&gt;EY report for BLIK&lt;/a&gt;,
a figure of the operator’s own, built on central bank data). A
Polish checkout without BLIK gives up the market’s leading online
payment method. PayU, Przelewy24 and Comgate carry it natively.
Among the internationals, Adyen and Mollie list BLIK, Braintree
added it in Magento 2.4.9, and Stripe’s module documentation omits
it. Most of the field carries BLIK in some form. What separates the
options in Poland is BLIK depth (redirect, in-checkout entry, or
one-click) and module quality. Przelewy24 documents the most
complete implementation, with the caveat from its profile: its Hyvä
Checkout module ships from its own site and appears in neither Hyvä
tracker, so first-line support sits with the vendor alone.&lt;/p&gt;
&lt;p&gt;In the Czech Republic, the 2024 wave of APEK’s consumer survey, as
published in Shoptet’s Stav české e-commerce, has cards as the
method shoppers use most, at 42 percent, with Google and Apple Pay
at 23, cash on delivery at 13 and bank transfer at 12
(&lt;a href=&quot;https://www.ceska-ecommerce.cz/&quot;&gt;live page&lt;/a&gt;,
&lt;a href=&quot;http://web.archive.org/web/20250401235543/https://www.ceska-ecommerce.cz/&quot;&gt;archived&lt;/a&gt;
since the live page is overwritten annually, and the shares match
&lt;a href=&quot;https://data.apek.cz/&quot;&gt;APEK’s own 2024 series&lt;/a&gt;). Any international
gateway covers the cards and wallets. The bank buttons and QR
payments behind part of that transfer share are carried by GoPay
and Comgate, whose Magento
modules are the weakest of the ten. So the Czech question is whether
the transfer and cash-on-delivery share justifies a second gateway
at all: an international provider for cards plus a local gateway for
transfers costs a second module on every upgrade, a second
reconciliation and fee schedule, and per-method deduplication in
checkout, since both sides offer cards and wallets. The pairing
makes sense only if the card-fee savings exceed that overhead.
(The Czech and Slovak acquirer-side gateways, GP webpay, ČSOB,
TrustPay, are outside this comparison, and the bank-button coverage
runs through GoPay and Comgate.) In Slovakia, cash on delivery is still
at 25 to 30 percent of orders per an analysis by Upgates, an
e-commerce platform operator, reported by the TASR press agency in
&lt;a href=&quot;https://www.teraz.sk/ekonomika/dobierky-predstavuju-pre-e-shopy-najna/973629-clanok.html&quot;&gt;June 2026&lt;/a&gt;.
That is mostly a fulfillment decision, but it still
touches the payment setup through carrier reconciliation and the
offline payment method, and the deferred-payment products above are
the gateways’ answer to it. Slovakia settles in euro, which removes
the currency argument for a local gateway there.&lt;/p&gt;
&lt;p&gt;So the procedure is: list the methods your market requires, keep the
gateways, or minimal combinations of them, that jointly carry those
methods, disqualify on the module and operational criteria above,
and only then compare fees, settlement currency included, among the
survivors.&lt;/p&gt;
&lt;p&gt;If you are choosing a gateway as part of a broader build, or
inheriting one that blocks an upgrade, that assessment is part of
our &lt;a href=&quot;/services/magento/&quot;&gt;Magento work&lt;/a&gt;, and connecting a gateway to
an ERP or subscription flow is what our
&lt;a href=&quot;/services/integrations/&quot;&gt;integrations service&lt;/a&gt; covers.&lt;/p&gt;</content:encoded><category>magento</category><category>payments</category><category>checkout</category><category>hyvä</category></item><item><title>Your store has been in scope of the Accessibility Act since June 2025</title><link>https://zapolu.com/blog/accessibility-act-ecommerce/</link><guid isPermaLink="true">https://zapolu.com/blog/accessibility-act-ecommerce/</guid><description>The European Accessibility Act covers e-commerce services sold to consumers. The microenterprise exemption is narrower than it looks, the 2030 date is not an extension, and the payment step is named explicitly.</description><pubDate>Fri, 31 Jul 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;The European Accessibility Act has applied to e-commerce services
since June 28, 2025. A year on, the common assumption is still that
accessibility is a public sector obligation, or a deadline somewhere
ahead. Article 2(2)(f) lists e-commerce services among the services
covered, and Article 3(30) defines them as “services provided at a
distance, through websites and mobile device-based services by
electronic means and at the individual request of a consumer with a
view to concluding a consumer contract”.&lt;/p&gt;
&lt;p&gt;That is a description of an online store selling to consumers. The
definition turns on a consumer contract, and Article 2(2) applies the
directive to services “provided to consumers”, so a storefront that
sells only to business customers falls outside it. A mixed store is
inside it for the consumer side.&lt;/p&gt;
&lt;p&gt;The directive is &lt;a href=&quot;https://eur-lex.europa.eu/eli/dir/2019/882/oj/eng&quot;&gt;Directive (EU)
2019/882&lt;/a&gt;,
OJ L 151, 7.6.2019. Every article, recital and annex cited below was
read in that text on July 30, 2026, because most of what circulates
about this law is a summary of a summary. We read it as engineers
who have to implement it rather than as lawyers, and the questions
of scope are worth raising with your own counsel, citing the
article numbers.&lt;/p&gt;
&lt;h2 id=&quot;the-exemption-is-narrower-than-small-business&quot;&gt;The exemption is narrower than “small business”&lt;/h2&gt;
&lt;p&gt;Article 4(5) exempts microenterprises providing services from the
accessibility requirements “and any obligations relating to the
compliance with those requirements”. Article 3(23) then defines a
microenterprise as one “which employs fewer than 10 persons and which
has an annual turnover not exceeding EUR 2 million or an annual
balance sheet total not exceeding EUR 2 million”.&lt;/p&gt;
&lt;p&gt;The question hangs on the conjunctions, and the two tests are not
symmetrical. Fewer than ten persons is a hard ceiling: ten people
puts a store in scope whatever its finances. The money test is
satisfied by either limb, so a six-person store turning over
EUR 5 million is still a microenterprise if its balance sheet total
stays under EUR 2 million. Recital 53 adds that a microenterprise
must “genuinely fulfil the requirements of Commission Recommendation
2003/361/EC, and the relevant case law, aimed at preventing the
circumvention of its rules”, which is where the rules for counting
staff and financial thresholds actually live. Read those before
concluding you are exempt.&lt;/p&gt;
&lt;h2 id=&quot;june-2030-is-not-an-extension&quot;&gt;June 2030 is not an extension&lt;/h2&gt;
&lt;p&gt;Article 32(1) has Member States provide “a transitional period ending
on June 28, 2030, during which service providers may continue to
provide their services using products which were lawfully used by
them to provide similar services before that date”, and allows
service contracts agreed before June 28, 2025 to “continue without
alteration until they expire, but no longer than five years from that
date”.&lt;/p&gt;
&lt;p&gt;This is the most misread sentence in the directive. It covers
&lt;em&gt;products&lt;/em&gt; used in providing a service, and Article 3(2) defines a
product as a good produced through a manufacturing process, while
Article 2(1) lists which products are actually in scope: payment
terminals, ticketing and check-in machines, ATMs, interactive
information terminals, e-readers, and consumer computer hardware with
its operating system. It also covers contracts already signed.
Recital 101 settles the reading: the relief runs to products placed
on the market before the application date, and it lapses if the
provider replaces them during the period.&lt;/p&gt;
&lt;p&gt;The transitional period does exist, then, and it applies to the
product layer. If you run collection kiosks or in-store payment
terminals, those have until 2030, and Article 32(2) gives
self-service terminals up to twenty years. A website is not a product
under this directive, and an order placed by a customer today is not
a contract agreed before June 2025. The storefront itself gets
nothing from Article 32.&lt;/p&gt;
&lt;h2 id=&quot;what-the-directive-asks-of-a-storefront&quot;&gt;What the directive asks of a storefront&lt;/h2&gt;
&lt;p&gt;Annex I, Section III(c) requires making websites, related online
applications and mobile services accessible “in a consistent and
adequate way by making them perceivable, operable, understandable and
robust”. Those four adjectives are the four principles of WCAG, which
recital 47 carries in from the Web Accessibility Directive and
declares “also relevant for this Directive”.&lt;/p&gt;
&lt;p&gt;Section IV(g) then adds three requirements specific to e-commerce.
The first is information about the accessibility of the products and services
being sold, where the responsible economic operator provides it.
The second is accessibility of the functionality for identification, security and
payment “when delivered as part of a service instead of a product”.
The third is identification methods, electronic signatures and payment
services that are themselves perceivable, operable, understandable
and robust.&lt;/p&gt;
&lt;p&gt;That last one closes a gap worth knowing about. Article 2(4)(d)
excludes, from the content of websites and mobile applications,
“third-party content that is neither funded, developed by, or under
the control of, the economic operator concerned”. A hosted payment
form does not fall under that exclusion, because the exclusion
requires all three to be absent, and a merchant who pays the provider
and configures the integration has both funded it and kept control of
it. Recital 20 says the same thing directly: “Even if a service, or
part of a service, is subcontracted to a third party, the
accessibility of that service should not be compromised and the
service providers should comply with the obligations of this
Directive.” In practice the two components most likely to fail an
audit are the two nobody at the company wrote: the hosted card fields
served by the payment provider, and the consent banner. Choosing a
payment provider is now partly an accessibility decision.&lt;/p&gt;
&lt;h2 id=&quot;the-standard-a-supplier-cannot-yet-sell-you&quot;&gt;The standard a supplier cannot yet sell you&lt;/h2&gt;
&lt;p&gt;Article 15(1) grants a presumption of conformity to standards “the
references of which have been published in the Official Journal of
the European Union”, and Article 15(3) lets the Commission establish
technical specifications that carry the same presumption. EN 301 549
V3.2.1 (March 2021) is published in the Journal under the Web
Accessibility Directive, by Commission Implementing Decision (EU)
2021/1339. Neither route has been used for this directive: the
Commission’s index of harmonised standards carries no entry for
Directive (EU) 2019/882, checked on July 30, 2026.&lt;/p&gt;
&lt;p&gt;The revision meant to cover the Accessibility Act came out of
standardisation request M/587, reached final draft as V4.1.0 in June
2026, and is in ETSI’s approval procedure. So a supplier offering
conformance with EN 301 549 is offering good engineering, not a legal
presumption under this directive. Follow the standard regardless: it
is the technical specification the directive’s machinery points at,
and its web chapter is WCAG 2.1 levels A and AA. V4.1.0 retargets
that chapter to WCAG 2.2, so the version you implement to will change
once it is cited.&lt;/p&gt;
&lt;h2 id=&quot;the-statement-most-stores-have-not-written&quot;&gt;The statement most stores have not written&lt;/h2&gt;
&lt;p&gt;Article 13(2), with Annex V, requires the service provider to explain
how the service meets the accessibility requirements, in the general
terms and conditions or an equivalent document. Annex V asks for
three things: a general description of the service in accessible
formats, the explanations needed to understand how it operates, and a
description of how the Annex I requirements are met. Article 13(2)
adds that this has to be public in written and oral format, itself
accessible, and kept for as long as the service is in operation.&lt;/p&gt;
&lt;p&gt;It is the one obligation that can be checked from outside without
touching the code, so it is where an inspection starts. For services
that inspection does not come from a market surveillance authority,
which Article 19 confines to products: Article 23 has each Member
State designate a separate authority responsible for checking the
compliance of services. Article 13(4) adds the last step. On finding
the service non-compliant, the provider takes corrective measures and
immediately informs the competent national authorities.&lt;/p&gt;
&lt;h2 id=&quot;requirements-of-a-disproportionate-burden-claim&quot;&gt;Requirements of a disproportionate-burden claim&lt;/h2&gt;
&lt;p&gt;Article 14(1) applies the requirements only so far as compliance
avoids a fundamental alteration of the service and a disproportionate
burden on the operator. Annex VI sets three criteria. The first is the ratio of
net compliance costs to overall operating and capital costs. The second
weighs the estimated costs and benefits for the operator against the
estimated benefit to persons with disabilities, given how much the
service is used. The third is the ratio of net compliance costs to net
turnover. The second sets one estimate against another, so the
assessment involves judgment, but it is judgment you have to write
down and defend.&lt;/p&gt;
&lt;p&gt;Article 14(3) requires the assessment documented, the results kept
five years, and a copy handed to the authority on request.
Article 14(5) requires it renewed whenever the service is altered,
whenever the authority asks, and at least every five years.
Article 14(6) bars any operator that took outside funding for
accessibility work from relying on disproportionate burden at all,
though the fundamental-alteration limb stays open, and Article 14(8)
requires telling the authority you are relying on it. Before choosing
that route, compare the cost of the file and its five-year upkeep
with the cost of the remediation it is meant to avoid.&lt;/p&gt;
&lt;h2 id=&quot;the-usual-failures&quot;&gt;The usual failures&lt;/h2&gt;
&lt;p&gt;The recurring ones in e-commerce, roughly in the order they stop a
purchase:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Keyboard alone, end to end.&lt;/strong&gt; Category page, product page, cart,
checkout, payment, confirmation, without touching a mouse. Cart,
checkout and payment are where this most often stops.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Labels and errors.&lt;/strong&gt; Fields with placeholder text instead of a
programmatically associated label, and validation errors signaled
by a red border rather than tied to the field and announced.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Focus in overlays.&lt;/strong&gt; Drawers, modals, and cookie dialogs where
focus never enters, escapes early, or is lost on close.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Custom widgets.&lt;/strong&gt; Variant swatches, quantity steppers, filter
facets, and selects rebuilt from generic elements for design
reasons.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Color carrying meaning alone.&lt;/strong&gt; Stock state, discount, validity.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Contrast on what gets lightened last.&lt;/strong&gt; Sale prices, hint text
under form fields, secondary labels in the order summary. Disabled
controls are exempt under WCAG 1.4.3, so leave them out of the
count.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Silent state changes.&lt;/strong&gt; Add to cart, cart totals, filter results,
and asynchronous price recalculation that updates the page without
announcing anything.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Automated scanners are worth running and will not find most of this.
The test that matters most is a person completing a purchase with a
keyboard, then with a screen reader. It will not settle everything:
reflow at 400 percent, text resizing, and target size need their own
checks.&lt;/p&gt;
&lt;p&gt;Enforcement is national. Article 29 requires member states to let
consumers act before the courts or the competent administrative
bodies, and lets organizations with a legitimate interest act on
their behalf. Article 30 leaves the penalties to each member state.
The transpositions covering the markets we work in all apply from
June 28, 2025: Germany’s Barrierefreiheitsstärkungsgesetz, Czechia’s
zákon č. 424/2023 Sb., Slovakia’s zákon č. 351/2022 Z. z., and
Poland’s ustawa of April 26, 2024 (Dz. U. z 2024 r. poz. 731).&lt;/p&gt;
&lt;p&gt;Two things are worth doing, in this order. Write the Annex V
statement, because it needs no code and it can be checked without
your help. Then walk the purchase path with a keyboard and fix what
stops you. If you would rather have someone walk it with you,
&lt;a href=&quot;/contact/&quot;&gt;tell us what you sell and where&lt;/a&gt;.&lt;/p&gt;</content:encoded><category>accessibility</category><category>compliance</category><category>checkout</category></item><item><title>How Magento security patches arrive now, and what skipping one costs</title><link>https://zapolu.com/blog/magento-security-patch-cadence/</link><guid isPermaLink="true">https://zapolu.com/blog/magento-security-patch-cadence/</guid><description>Adobe ships one full Magento release a year, and the most urgent security fixes now arrive as standalone files outside Composer. The current patch channels and support dates, and what the SessionReaper timeline says a delayed patch costs.</description><pubDate>Sat, 18 Jul 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;If your Magento patching routine was designed around the old
quarterly release train, it is now aimed at a schedule Adobe no
longer follows. There is one full release a year, security-only
releases a few times a year, and, since this year, the most urgent
fixes ship as standalone patch files that never appear in Composer
at all. A process that waits for &lt;code&gt;composer update&lt;/code&gt; to surface
security work misses exactly the fixes with a deadline measured in
days.&lt;/p&gt;
&lt;p&gt;Each vendor fact below links to the vendor’s own page and was
checked on July 17, 2026.&lt;/p&gt;
&lt;h2 id=&quot;the-channels-a-fix-can-arrive-through&quot;&gt;The channels a fix can arrive through&lt;/h2&gt;
&lt;p&gt;Adobe’s published
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/planning/schedule&quot;&gt;release schedule&lt;/a&gt;
commits to one full patch release for the 2.4.x line each year, in
May, and to security patches “at minimum annually” for every
supported line. Practice runs well ahead of that minimum, and the
recent security releases have landed on the second Tuesday of a
month: the emergency SessionReaper fix on September 9, 2025, its
scheduled follow-up on October 14, 2025, the 2.4.9 release with a
security batch for the older lines on May 12, 2026, and the current
bulletin,
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/experience-cloud-kcs/kbarticles/ka-37421&quot;&gt;APSB26-73&lt;/a&gt;,
on July 14, 2026.&lt;/p&gt;
&lt;p&gt;Four kinds of artifact carry the fixes
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/planning/versioning-policy&quot;&gt;release policy&lt;/a&gt;):&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Full patch releases&lt;/strong&gt;, such as 2.4.9: one a year, carrying
security work alongside quality fixes and platform compatibility.
Adobe’s release notes for 2.4.9 count
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/notes/magento-open-source/2-4-9&quot;&gt;580 fixed core issues&lt;/a&gt;,
which is the honest measure of how much testing one demands.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Security patch releases&lt;/strong&gt;, the &lt;code&gt;-pN&lt;/code&gt; versions: cumulative,
security-only updates on top of your line’s last full release, so
a store can take the security fixes without the feature payload
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/notes/security-patches/overview&quot;&gt;security patch overview&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Isolated patches&lt;/strong&gt;: standalone, non-cumulative files fixing one
or a few vulnerabilities. Two properties matter operationally.
They apply only on top of your line’s latest &lt;code&gt;-pN&lt;/code&gt; release, and
they are not Composer packages: the July 2026 fixes shipped as
downloadable patch files only, with no Composer release alongside
them
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/experience-cloud-kcs/kbarticles/ka-37421&quot;&gt;Adobe’s APSB26-73 notice&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Hotfixes&lt;/strong&gt; for critical issues, outside any schedule.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Isolated patches are the piece existing processes miss. A pipeline
that patches by raising the Composer version constraint sees full
releases and &lt;code&gt;-pN&lt;/code&gt; releases and nothing else, and Adobe now routes
the most time-critical fixes around it.&lt;/p&gt;
&lt;h2 id=&quot;the-support-clock-per-line&quot;&gt;The support clock per line&lt;/h2&gt;
&lt;p&gt;Every line receives security fixes for three years from its release
date, per Adobe’s
&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/planning/lifecycle-policy&quot;&gt;lifecycle policy&lt;/a&gt;.
The dates, as that page states them today:&lt;/p&gt;
&lt;div class=&quot;table-scroll&quot;&gt;





























&lt;table&gt;&lt;thead&gt;&lt;tr&gt;&lt;th&gt;Line&lt;/th&gt;&lt;th&gt;Released&lt;/th&gt;&lt;th&gt;Security fixes end&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;&lt;tbody&gt;&lt;tr&gt;&lt;td&gt;2.4.9&lt;/td&gt;&lt;td&gt;May 12, 2026&lt;/td&gt;&lt;td&gt;May 31, 2029&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;2.4.8&lt;/td&gt;&lt;td&gt;April 8, 2025&lt;/td&gt;&lt;td&gt;May 31, 2028&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;2.4.7&lt;/td&gt;&lt;td&gt;April 9, 2024&lt;/td&gt;&lt;td&gt;May 31, 2027&lt;/td&gt;&lt;/tr&gt;&lt;tr&gt;&lt;td&gt;2.4.6&lt;/td&gt;&lt;td&gt;March 14, 2023&lt;/td&gt;&lt;td&gt;August 11, 2026&lt;/td&gt;&lt;/tr&gt;&lt;/tbody&gt;&lt;/table&gt;&lt;/div&gt;
&lt;p&gt;Adobe does advertise extended support beyond these dates, and the
fine print matters: extended support security patches “are available
to Adobe Commerce customers only. They are not available for the
Magento Open Source code base”
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/planning/versioning-policy&quot;&gt;release policy&lt;/a&gt;).
For a Magento Open Source store the table above is the whole story.
The 2.4.6 line stopped receiving security fixes on August 11, 2026, so
a store on 2.4.6 or older now receives nothing.&lt;/p&gt;
&lt;h2 id=&quot;what-a-delay-costs-with-dates&quot;&gt;What a delay costs, with dates&lt;/h2&gt;
&lt;p&gt;The current reference case is SessionReaper, CVE-2025-54236: an
unauthenticated vulnerability rated CVSS 9.1 and affecting every
supported line
(&lt;a href=&quot;https://helpx.adobe.com/security/products/magento/apsb25-88.html&quot;&gt;APSB25-88&lt;/a&gt;).
Adobe published an emergency fix on September 9, 2025, as an isolated
hotfix outside the schedule, and deployed WAF rules in front of
Commerce Cloud stores
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/experience-cloud-kcs/kbarticles/ka-27397&quot;&gt;Adobe’s advisory&lt;/a&gt;).&lt;/p&gt;
&lt;p&gt;Mass exploitation began on October 22, 2025, six weeks later. That
day,
&lt;a href=&quot;https://sansec.io/research/sessionreaper-exploitation&quot;&gt;Sansec’s scans&lt;/a&gt;
still found 62% of Magento stores unprotected. Two days on, Adobe
raised the bulletin’s priority to its highest rating, the one that
means patch within 72 hours. The previous year ran the same shape
faster: merchants who applied the June 2024 CosmicSting patch more
than two weeks late had their cryptographic keys stolen, which
&lt;a href=&quot;https://sansec.io/guides/magento-security&quot;&gt;Sansec confirmed on more than 4,200 stores&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;Sansec’s explanation for the urgency is that attackers
reverse-engineer each patch on the day it ships: the fix documents
the vulnerability. SessionReaper’s grace period was six weeks.
CosmicSting’s was two.&lt;/p&gt;
&lt;h2 id=&quot;the-routine-that-matches-the-calendar&quot;&gt;The routine that matches the calendar&lt;/h2&gt;
&lt;p&gt;Kept up with, the cadence above is small, scheduled work.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Watch the security bulletins rather than the package feed.&lt;/strong&gt;
Adobe announces isolated patches there. They do not appear in
Composer.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Stay on your line’s latest &lt;code&gt;-pN&lt;/code&gt;.&lt;/strong&gt; Isolated patches require
it, so falling one security release behind also delays the next
emergency fix until you catch up.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Budget the May full release as a yearly upgrade.&lt;/strong&gt; It moves PHP
and dependency versions and touches extensions. The &lt;code&gt;-pN&lt;/code&gt; and
isolated patches between Mays are hours of work on a well-kept
codebase.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Plan the line upgrade against the three-year clock&lt;/strong&gt; rather
than waiting for a feature reason. The table above already names
the quarter by which your next upgrade must land.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Running &lt;a href=&quot;/blog/magento-vs-mageos-vs-adobe-commerce/&quot;&gt;Mage-OS&lt;/a&gt;
changes who packages the fixes, and keeps the routine. The
association supports its latest release branch only and typically
publishes within days of each Adobe release. Its 3.2.0 security
release shipped on July 14, 2026, the same day as Adobe’s bulletin,
porting the isolated patch into the distribution’s source
(&lt;a href=&quot;https://mage-os.org/product/releases/&quot;&gt;Mage-OS releases&lt;/a&gt;).&lt;/p&gt;
&lt;p&gt;This routine is part of the long-term operational support described
on our &lt;a href=&quot;/services/magento/&quot;&gt;Magento service page&lt;/a&gt;. If nobody at your
company can currently say which &lt;code&gt;-pN&lt;/code&gt; your store runs, or the date
its line stops receiving fixes, that is worth
&lt;a href=&quot;/contact/&quot;&gt;a short conversation&lt;/a&gt;.&lt;/p&gt;</content:encoded><category>magento</category><category>security</category><category>operations</category></item><item><title>On August 26, Shopify stops running your additional scripts</title><link>https://zapolu.com/blog/shopify-additional-scripts-sunset/</link><guid isPermaLink="true">https://zapolu.com/blog/shopify-additional-scripts-sunset/</guid><description>Non-Plus stores lose the additional scripts box on the Thank you and Order status pages. What stops firing, what replaces it, and how to migrate without silently losing conversion tracking.</description><pubDate>Mon, 13 Jul 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;&lt;strong&gt;Update, October 2026:&lt;/strong&gt; the deadline of August 26, 2026 has passed.
Shopify’s &lt;a href=&quot;https://help.shopify.com/en/manual/checkout-settings/customize-checkout-configurations/upgrade-thank-you-order-status&quot;&gt;upgrade page&lt;/a&gt;,
checked on October 12, 2026, says that stores not upgraded by that
date were upgraded automatically. The post below is written as of
July 2026. Its inventory, triage and verification steps still apply to a
store whose conversion tracking stopped on that date. The step to
upgrade yourself, and the 30-day revert that came with it, no longer
apply.&lt;/p&gt;
&lt;p&gt;On August 26, 2026, Shopify removes additional scripts from the
Thank you and Order status pages on every non-Plus store. Whatever
JavaScript sits in that box stops firing that day. In most stores that
box holds the Google Ads conversion tag, the Meta pixel, and a layer
of custom snippets added over the years.&lt;/p&gt;
&lt;p&gt;If you are on Plus, this already happened to you. Your deadline was
August 28, 2025, and Shopify began auto-upgrading the stores that
ignored it in January 2026. Worth checking what you lost.&lt;/p&gt;
&lt;p&gt;Both dates are Shopify’s own, checked against its documentation on
July 13, 2026: the &lt;a href=&quot;https://help.shopify.com/en/manual/checkout-settings/customize-checkout-configurations/upgrade-thank-you-order-status&quot;&gt;non-Plus upgrade deadline&lt;/a&gt;
of August 26, 2026, and the &lt;a href=&quot;https://shopify.dev/docs/storefronts/themes/architecture/layouts/checkout-liquid&quot;&gt;sunset of &lt;code&gt;checkout.liquid&lt;/code&gt;, additional
scripts and script tags&lt;/a&gt;
for Plus on August 28, 2025.&lt;/p&gt;
&lt;p&gt;One clarification, because this is widely misreported: if you are not
on Plus you never had &lt;code&gt;checkout.liquid&lt;/code&gt;. Shopify documents it as a
Plus-only layout. You cannot be broken by its removal, and any agency
telling you to migrate your &lt;code&gt;checkout.liquid&lt;/code&gt; is quoting you for work
that does not exist. What you have is &lt;em&gt;Settings &gt; Checkout &gt;
Additional scripts&lt;/em&gt;, and that is what is going away.&lt;/p&gt;
&lt;h2 id=&quot;what-actually-stops-firing&quot;&gt;What actually stops firing&lt;/h2&gt;
&lt;p&gt;Everything in the additional scripts box, which in practice means:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Conversion tracking.&lt;/strong&gt; Google Ads, Meta, TikTok, affiliate
postbacks. The revenue keeps arriving. The attribution stops.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Analytics fired at the point of purchase.&lt;/strong&gt; Custom GA4 events,
server-side tagging bridges, anything pasted into the box instead of
built as an integration.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Content on the two pages.&lt;/strong&gt; Delivery estimates, review prompts,
referral offers, surveys left over from finished campaigns.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Apps that install through script tags.&lt;/strong&gt; These break the same
way, and the app’s dashboard will not warn you.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The failure is quiet. Nothing goes down. The store keeps taking
orders, the checkout keeps working, and the only symptom is that your
ad platforms slowly stop being able to tell you which spend produced
which revenue. Nobody opens an incident for that. It surfaces a month
later as an inexplicable drop in reported ROAS, and by then the data
is gone.&lt;/p&gt;
&lt;h2 id=&quot;what-replaces-it&quot;&gt;What replaces it&lt;/h2&gt;
&lt;p&gt;Shopify’s replacement is checkout extensibility. It offers three
structured routes, and none of them runs unrestricted JavaScript in
the checkout page itself:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Custom pixels&lt;/strong&gt; for tracking, under &lt;em&gt;Settings &gt; Customer events&lt;/em&gt;.
Each script becomes a custom pixel, or is dropped in favor of the
official app integration, which is usually the better answer for
Google and Meta.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Blocks&lt;/strong&gt; for anything the page should display. Design and
functional customizations get rebuilt as blocks in the checkout
editor.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;App pixels&lt;/strong&gt; for apps, provided the vendor shipped one. This is
the part you cannot fix yourself.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id=&quot;the-six-weeks&quot;&gt;The six weeks&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Get the actual list, not your memory of it.&lt;/strong&gt; &lt;em&gt;Settings &gt;
Checkout&lt;/em&gt;, then the upgrade notice, then Review customizations.
Shopify sorts what you have into compatible apps, incompatible apps,
and additional scripts.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Triage the scripts by owner.&lt;/strong&gt; Some of them are dead: campaigns
that ended, vendors that churned, tags nobody can attribute.
Deleting those is the cheapest part of the migration.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Move tracking first.&lt;/strong&gt; It is the only category with real money
attached.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Check every incompatible app for a vendor upgrade.&lt;/strong&gt; If the
vendor never shipped an extension, that app is now a decision
rather than a migration, and you want to make it deliberately
rather than discover it on August 27.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Upgrade yourself, on a day you are watching.&lt;/strong&gt; If you do nothing,
Shopify upgrades the pages for you and the customizations stop
working. Upgrading yourself also keeps a way back that the automatic
upgrade does not: Shopify allows a revert within 30 days of a manual
upgrade, for stores created before January 6, 2025 that customized
these pages with additional scripts or script-tag apps.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Then verify, because a migrated pixel that fires wrong looks exactly
like one that fires right. For one week after cutover, reconcile the
conversions your ad platforms report against the orders in your
Shopify admin. If those two numbers agree, the migration worked.
Without that reconciliation, a broken pixel goes unnoticed until the
quarter’s reporting is already wrong.&lt;/p&gt;
&lt;p&gt;The work is well understood and the date is fixed. Moving the tags
themselves is marketing-stack work, which is
&lt;a href=&quot;/what-we-dont-do/&quot;&gt;outside our scope&lt;/a&gt;, and we can refer you to a
specialist. Checkout blocks and incompatible apps are engineering work,
which our &lt;a href=&quot;/services/shopify/&quot;&gt;Shopify page&lt;/a&gt; covers. If nobody currently at
the company can say what is in the additional scripts box, that is
what the &lt;a href=&quot;/services/technical-audit/&quot;&gt;initial technical audit&lt;/a&gt; is for.&lt;/p&gt;</content:encoded><category>shopify</category><category>checkout</category><category>migration</category></item><item><title>Only two payment processors on Shopify can charge in the customer&apos;s currency</title><link>https://zapolu.com/blog/shopify-local-currency-processor/</link><guid isPermaLink="true">https://zapolu.com/blog/shopify-local-currency-processor/</guid><description>Showing a price in euros and charging euros are two different Shopify features, and both belong to your payment processor. What breaks when the processor changes, and the four ways out.</description><pubDate>Mon, 13 Jul 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Showing a market’s prices in another currency and charging in that
currency are two separate Shopify features. Neither belongs to the
store. Both belong to the payment processor.&lt;/p&gt;
&lt;p&gt;For charging, exactly two processors have it. Shopify’s documentation
on local currencies, checked on July 13, 2026, is explicit: &lt;a href=&quot;https://help.shopify.com/en/manual/markets/customizations/local-currencies&quot;&gt;“To
process payments in multiple currencies, your store must use Shopify
Payments or Adyen as its payment processor. If you use a different
provider, then customers in all markets pay in your store’s default
currency.”&lt;/a&gt;
The &lt;a href=&quot;https://help.shopify.com/en/manual/international/pricing/limitations&quot;&gt;guidelines for selling in local
currencies&lt;/a&gt;
state the same rule from the other side: orders placed through other
third-party payment providers are processed in the store’s default
currency, and reports are available only in the store currency.&lt;/p&gt;
&lt;p&gt;The same rule covers showing, which is the part this post originally
got wrong. The same guidelines page, checked on August 13,
2026, says that “showing prices in local currencies and setting manual
exchange rates do require that you use Shopify Payments.” The admin
enforces it: on a store without a completed Shopify Payments account,
the currency setting in Markets is disabled outright, behind a notice
that asks you to complete the Shopify Payments setup first. What stays
available on every plan is market-specific prices and percentage
adjustments, in the store’s own currency.&lt;/p&gt;
&lt;p&gt;So without Shopify Payments a customer in Bratislava never sees a euro
through Markets at all: the product page, the cart and the checkout
all stay in the store’s default currency. A store can still show euros
through a currency app or through prices it computes itself, and that
route is described below, but the checkout still charges the default
currency, and the customer’s bank may add its own conversion fee on
top of that charge.&lt;/p&gt;
&lt;p&gt;You can put any number on a page. You can only charge the one your
payment processor supports.&lt;/p&gt;
&lt;h2 id=&quot;why-it-surprises-people&quot;&gt;Why it surprises people&lt;/h2&gt;
&lt;p&gt;Nobody plans multi-currency as a payment feature. On a store that
runs Shopify Payments you open Markets, add a market, watch the
prices convert, and it reads as a storefront setting. The dependency
runs the other way, and it surfaces at two different moments.&lt;/p&gt;
&lt;p&gt;On a store that never completed Shopify Payments it surfaces at once:
the currency setting is locked, and the notice in Markets says to
finish the Shopify Payments setup before customizing currency. The
admin at least names the dependency there.&lt;/p&gt;
&lt;p&gt;On a store that had the feature, it surfaces on the day the payment
processor changes: a new legal entity behind the store, a move to a
local gateway, a Shopify Payments account that has to be verified
again. Local-currency pricing leaves with it. Nothing in the admin
connects the two events, so what the store sees is that the euro
prices stopped working on a day when nobody touched the euro prices.&lt;/p&gt;
&lt;h2 id=&quot;the-four-routes&quot;&gt;The four routes&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;Set up Shopify Payments.&lt;/strong&gt; This unlocks the feature exactly, and in
engineering terms it is the cheapest of the four, because there is no
engineering in it. The cost is paperwork, and the paperwork has its own
section below.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Adyen as the primary gateway.&lt;/strong&gt; The only other processor Shopify
documents for this, and its requirements page rules it out for most
stores. &lt;a href=&quot;https://help.shopify.com/en/manual/payments/third-party-providers/adyen-gateway&quot;&gt;Shopify’s requirements for the Adyen
integration&lt;/a&gt;,
checked on July 21, 2026, ask for an existing Adyen account, Shopify
Checkout, and the &lt;a href=&quot;https://www.shopify.com/enterprise&quot;&gt;Enterprise plan&lt;/a&gt;,
the custom-priced tier above Shopify Plus that Shopify sells to its
largest merchants. The same page states that access is granted by
approval only. For a mid-size store this route is unavailable in
practice.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;A second store with the euro as its default currency.&lt;/strong&gt; This works,
with no tricks in it, and it is the honest answer when the euro market
is really a different business: different assortment, different
pricing, sometimes a different legal entity. What people cost is the
second subscription, which is the small number. The large number is the
second copy of the catalog, the theme, the apps and every integration,
plus an inventory sync between the two stores that runs for as long as
both exist.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Compute the prices yourself and display them.&lt;/strong&gt; Currency apps do
this, and so does storing your own converted prices in metafields.
Without Shopify Payments it is the only way a localized price reaches
the storefront at all. It is the route that gets proposed once the
other three are unwelcome, and it needs the most care, so it gets its
own section too.&lt;/p&gt;
&lt;h2 id=&quot;what-a-self-computed-price-can-reach-and-what-it-cannot&quot;&gt;What a self-computed price can reach, and what it cannot&lt;/h2&gt;
&lt;p&gt;Prices you compute yourself can reach every surface you control. The
storefront, a draft order, an internal tool, a PDF quote: each of these
renders numbers you supply, and each will render euros correctly.&lt;/p&gt;
&lt;p&gt;The checkout is not a surface you control. With a third-party provider
Shopify processes the order in the store’s default currency, whatever
the product page said, and the reports come out in the store currency
too. The euro price is therefore correct everywhere except at the point
where the money is taken.&lt;/p&gt;
&lt;p&gt;That makes this route conditional on one question about the business:
does any customer ever pay you in euros through Shopify’s checkout? If
a customer can put a euro price in the cart and pay by card, the price
on the product page is not the price that gets charged. If the euro
path instead ends in a draft order, an invoice and a bank transfer,
then the checkout never handles a euro, no one is charged in the wrong
currency, and computing the prices yourself is not a workaround at all.
It is where the pricing has to live.&lt;/p&gt;
&lt;p&gt;Before building any of it, check one thing. Shopify &lt;a href=&quot;https://help.shopify.com/en/manual/fulfillment/managing-orders/create-orders/international-drafts&quot;&gt;documents a
currency override on draft
orders&lt;/a&gt;
once Markets is set up, with the exchange rate fixed at the moment the
draft order is created. Collecting payment on such an order is
restricted: Shopify’s stated limitation is that with deferred payment
terms in a non-default currency you can collect only by credit card or
by marking the order as paid. That may cover the quoting case without
any metafields, or it may fail to survive contact with your invoicing.
Test it on the actual store before writing code that assumes either
answer, and not on a development store. &lt;a href=&quot;https://help.shopify.com/en/partners/dashboard/managing-stores/test-orders-in-dev-stores&quot;&gt;Shopify’s page on test orders
in development
stores&lt;/a&gt;,
checked on August 13, 2026, rules out draft orders and manual payment
methods such as cash on delivery for checkout testing, and warns that
processing an order through one of them can lock the store’s ability
to process orders. The supported way to test a dev-store checkout is
the test payment provider with test card numbers.&lt;/p&gt;
&lt;h2 id=&quot;the-paperwork-and-why-the-owner-cannot-argue-with-it&quot;&gt;The paperwork, and why the owner cannot argue with it&lt;/h2&gt;
&lt;p&gt;The objection to Shopify Payments is rarely technical. Setting it up
means identifying the people behind the company: a photo ID, a date of
birth, the company’s beneficial owners, the registration number, the VAT
number and a SEPA bank account. Shopify publishes the exact list per
country, so it varies by a document here and there, and on the Polish
page a beneficial owner is anyone holding 25 percent of the business or
more. Owners resist all of this, and they resist hardest
when they read it as Shopify prying into a business that is none of its
concern.&lt;/p&gt;
&lt;p&gt;Every one of those country pages gives the same reason, in the same
sentence, with only the country’s name changed. Here is the page for
Czechia, checked on July 13, 2026: &lt;a href=&quot;https://help.shopify.com/en/manual/payments/shopify-payments/supported-countries/czechia/requirements&quot;&gt;“Shopify and its banking partners
are required to collect and verify information about you and your
business, in order to comply with regulations in Czechia that are
designed to prevent financial
crimes.”&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;Every card processor collects the same information, because every card
processor answers to the same anti-money-laundering rules. An owner who
declines to identify the company’s beneficial owners is declining to
accept card payments anywhere, and the next provider asks for the same
documents.&lt;/p&gt;
&lt;h2 id=&quot;if-the-company-is-slovak-the-first-route-does-not-exist&quot;&gt;If the company is Slovak, the first route does not exist&lt;/h2&gt;
&lt;p&gt;Shopify Payments runs in a fixed list of countries. The
&lt;a href=&quot;https://help.shopify.com/en/manual/payments/shopify-payments/supported-countries&quot;&gt;supported-countries
list&lt;/a&gt;,
checked on July 13, 2026, includes Czechia, Germany and Poland. It does
not include Slovakia.&lt;/p&gt;
&lt;p&gt;For a Slovak entity, then, the first route is unavailable rather than
merely difficult, and no amount of persuading the owner to complete the
verification will change that. Adyen requires the Enterprise plan, so
the choice narrows to a second store or self-computed prices. Establish
this early, before somebody spends a week preparing an application that
cannot be submitted.&lt;/p&gt;
&lt;h2 id=&quot;the-question-that-decides-it&quot;&gt;The question that decides it&lt;/h2&gt;
&lt;p&gt;The decision turns on who pays you in euros. If customers pay in euros,
the currency belongs to the payment processor, and for most stores that
means Shopify Payments. If customers are quoted in euros and
settle against an invoice, the checkout never handles the currency, and
you are free to put the price wherever it is most useful.&lt;/p&gt;
&lt;p&gt;Answer that first. Every option above is cheap or expensive depending
only on the answer, and it is a question for the business rather than
for the developer. The work that follows either answer is on our
&lt;a href=&quot;/services/shopify/&quot;&gt;Shopify page&lt;/a&gt;.&lt;/p&gt;</content:encoded><category>shopify</category><category>payments</category><category>markets</category></item><item><title>The online store without a CTO: seven decisions that go wrong</title><link>https://zapolu.com/blog/eshop-without-cto/</link><guid isPermaLink="true">https://zapolu.com/blog/eshop-without-cto/</guid><description>Vendor choice, hiring, hosting, ownership: the technical decisions that quietly go wrong when nobody senior owns them, and what a CTO would ask instead.</description><pubDate>Wed, 08 Jul 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Between roughly ten and a hundred people, an e-commerce company is in
a difficult position: too much technology to manage informally, too
little to justify a full-time CTO. The gap doesn’t stay empty. It gets filled by
whoever is closest (a founder, an agency, the most confident
developer), and a series of decisions gets made that nobody is
qualified to veto. These are the seven we see go wrong most often,
and the question a senior engineer would have asked.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;1. Choosing the platform by the demo.&lt;/strong&gt; Platforms are bought on the
polish of the admin and the promises of the sales call, then lived
with for a decade of edge cases. &lt;em&gt;A CTO would ask for our three most unusual requirements, and for a demonstration of them working on this platform.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;2. Choosing the agency by the pitch.&lt;/strong&gt; The cheapest bid and the
most polished presentation are both selection biases, not evidence. The failure
surfaces two years later as lock-in: code in the agency’s private
repos, infrastructure on their accounts, and the knowledge nowhere
but their heads. &lt;em&gt;A CTO would ask what exactly we walk away with if we part ways in twelve months.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;3. Hiring a senior developer when the problem is architecture&lt;/strong&gt;,
or an architect when the problem is throughput. Without someone who
has seen both, the company overcorrects for its last bad
experience. &lt;em&gt;A CTO would ask whether our bottleneck is deciding what to build or building it.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;4. Letting the plugin count become the architecture.&lt;/strong&gt; Every plugin
was a quick win. Fifty quick wins later, upgrades take a quarter and
nobody can say which plugin owns the price calculation. &lt;em&gt;A CTO would ask, for each of them, what it does and whether we would install it again today.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;5. Inheriting infrastructure by default.&lt;/strong&gt; The store runs wherever
the first developer put it, sized by guesswork, with customer
complaints standing in for monitoring. It works until the first big campaign, which is precisely
when it mustn’t fail. When &lt;a href=&quot;/blog/checkout-collapse-under-load/&quot;&gt;checkout collapses under load that
the servers should handle&lt;/a&gt;,
that default becomes expensive. &lt;em&gt;A CTO would ask what happens at five times today’s traffic, and how we know.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;6. Not owning the accounts.&lt;/strong&gt; Domain at the agency’s registrar,
analytics under a freelancer’s Google account, deploy keys on a
laptop that left with an employee. Nothing appears wrong while the
relationship holds. When it ends, every one of those dependencies
becomes a problem. &lt;em&gt;A CTO would ask for a list of every account the business depends on, with the person who controls recovery for each.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;7. Treating performance as a launch task.&lt;/strong&gt; The site was fast at
launch. It’s been getting slower by one plugin and one banner at a
time ever since, and nobody owns the number. Speed is an operations
metric, like uptime. It needs an owner and a trend line. &lt;em&gt;A CTO would ask what our LCP was last month, and who noticed the change.&lt;/em&gt;&lt;/p&gt;
&lt;h2 id=&quot;the-pattern-behind-all-seven&quot;&gt;The pattern behind all seven&lt;/h2&gt;
&lt;p&gt;All seven are ownership failures: each decision was locally
reasonable and nobody was responsible for the sum. The fix isn’t necessarily a hire: at this company size the
CTO job is real but part-time, which is exactly why
&lt;a href=&quot;/services/fractional-cto/&quot;&gt;fractional CTO engagements&lt;/a&gt; exist. They
give you, on a monthly retainer with a scope agreed in writing, someone
whose job is to ask the questions above while the decisions are still
cheap to change.&lt;/p&gt;
&lt;p&gt;If several of the seven apply to your company,
&lt;a href=&quot;/contact/&quot;&gt;tell us which ones&lt;/a&gt;.&lt;/p&gt;</content:encoded><category>fractional-cto</category><category>leadership</category></item><item><title>WooCommerce to Shopify: what survives the move</title><link>https://zapolu.com/blog/woocommerce-to-shopify/</link><guid isPermaLink="true">https://zapolu.com/blog/woocommerce-to-shopify/</guid><description>What migrates cleanly, what needs rebuilding, and what is silently lost: a field checklist for leaving WooCommerce without losing your rankings.</description><pubDate>Wed, 01 Jul 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;WooCommerce to Shopify is one of the most common e-commerce
migrations, and the pattern behind it is consistent: the store outgrew
WordPress. Plugin conflicts consume a day of work each week, every
update gets postponed because it might break the site, and the
hosting bill keeps climbing because a bigger server is compensating
for the architecture. Shopify’s proposition, that the vendor operates
the platform, is most persuasive at exactly this moment.&lt;/p&gt;
&lt;p&gt;The migration itself is routinely undersold. “There’s an importer”
is accurate but incomplete. This is what survives the move, what
needs rebuilding, and what disappears silently unless someone
verifies it.&lt;/p&gt;
&lt;p&gt;Every Shopify limit below is Shopify’s own and links to its
documentation. We checked all of them on July 13, 2026.&lt;/p&gt;
&lt;h2 id=&quot;what-migrates-cleanly&quot;&gt;What migrates cleanly&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Products&lt;/strong&gt;, mostly. Simple and variable products map to Shopify
products and variants. Note the limits: Shopify raised the old
100-variant cap and since October 2025 allows
&lt;a href=&quot;https://shopify.dev/changelog/posts/the-product-variant-limit-is-now-2048-for-all-merchants&quot;&gt;2,048 variants per product for every merchant&lt;/a&gt;,
but a product still gets
&lt;a href=&quot;https://help.shopify.com/en/manual/products/variants/add-variants&quot;&gt;at most three options&lt;/a&gt;,
and three options is exactly the limit complex WooCommerce catalogs
exceed.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Customers&lt;/strong&gt;: names, emails, addresses. This part is a CSV export
and import, once you’ve handled marketing-consent state.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Historical orders&lt;/strong&gt;, via import apps or the API, with enough
caveats (statuses, partial refunds, tax rounding) that we treat
imported orders as reference data, not accounting truth.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;passwords-do-not-migrate&quot;&gt;Passwords do not migrate&lt;/h2&gt;
&lt;p&gt;Customer passwords are stored as one-way hashes, and Shopify does not
take them. Its
&lt;a href=&quot;https://help.shopify.com/en/manual/customers/import-export-customers&quot;&gt;customer import documentation&lt;/a&gt;
says passwords cannot be migrated from another store and that
imported customers have to be invited to create new ones. Every
customer therefore resets or re-registers. For a store that depends on
repeat purchases, plan that invitation as a re-activation campaign,
scheduled before cutover rather than improvised after it.&lt;/p&gt;
&lt;h2 id=&quot;parts-rebuilt-rather-than-copied&quot;&gt;Parts rebuilt rather than copied&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;The plugin stack.&lt;/strong&gt; Each WooCommerce plugin needs a decision: a
Shopify app, a Shopify Function, a theme feature, or the
conclusion that its purpose is no longer known. This inventory is the
real scope of the project.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Checkout customizations.&lt;/strong&gt; Shopify’s checkout is customizable
only within defined boundaries: Basic and above have the checkout
and accounts editor, while
&lt;a href=&quot;https://help.shopify.com/en/manual/checkout-settings/checkout-extensibility&quot;&gt;apps that customize the information, shipping and payment pages, and the Checkout Branding API, require Shopify Plus&lt;/a&gt;.
Whatever you built into WooCommerce’s checkout beyond those
boundaries is redesigned, not ported.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Multilingual setups.&lt;/strong&gt; WPML or Polylang constructions map onto
Shopify Markets plus translation apps, a different model rather
than a format conversion.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The content site around the store.&lt;/strong&gt; WordPress is a better CMS
than Shopify. Blogs and landing pages need a destination: Shopify’s
blog, a headless setup, or a WordPress that stays for content only.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;the-rankings&quot;&gt;The rankings&lt;/h2&gt;
&lt;p&gt;SEO fails first in low-budget migrations, because the damage shows up
weeks later where nobody connects cause and effect. The URL
structure changes by design: &lt;code&gt;/product-category/shoes/&lt;/code&gt; becomes
&lt;code&gt;/collections/shoes/&lt;/code&gt;, every product path changes, and Shopify’s
&lt;a href=&quot;https://help.shopify.com/en/manual/online-store/os/domains/domains-terminology&quot;&gt;subdirectory names cannot be edited&lt;/a&gt;,
so the old paths cannot be reproduced. The rankings survive
only with mechanical discipline:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;Export a &lt;strong&gt;complete URL inventory&lt;/strong&gt; from the live store. Crawl it
rather than trusting the sitemap alone.&lt;/li&gt;
&lt;li&gt;Map &lt;strong&gt;every URL that has traffic or backlinks&lt;/strong&gt; to its exact new
home. Category pages especially: they usually hold the rankings.&lt;/li&gt;
&lt;li&gt;Ship the &lt;strong&gt;301 redirects at cutover&lt;/strong&gt;. Rankings decline for as
long as the redirects are missing.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Benchmark organic traffic before&lt;/strong&gt;, so that “did the migration
hurt us” can be answered from a chart.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id=&quot;cases-against-shopify&quot;&gt;Cases against Shopify&lt;/h2&gt;
&lt;p&gt;If what’s driving you out of WooCommerce is catalog complexity or
B2B (contract pricing, approval chains, variant structures that
don’t fit three options),
you may hit the same limits on Shopify, and
&lt;a href=&quot;/services/magento/&quot;&gt;Magento’s model&lt;/a&gt; might fit better. An
initial technical audit can settle that decision when its order names
the &lt;a href=&quot;/services/platform-migration/&quot;&gt;migration&lt;/a&gt; as a further area, and the &lt;a href=&quot;/blog/replatform-or-fix/&quot;&gt;replatform-or-fix question&lt;/a&gt;
is worth asking honestly before either.&lt;/p&gt;
&lt;p&gt;With the URL inventory done first, the redirect map shipped at
cutover, and passwords handled as a campaign, this migration is
routine. Skip those steps and it turns into a quarter of lost
rankings and a support queue full of customers who can’t log in.
If you’re planning one,
&lt;a href=&quot;/contact/&quot;&gt;we will review the plan with you&lt;/a&gt;.&lt;/p&gt;</content:encoded><category>platform-migration</category><category>shopify</category><category>woocommerce</category></item><item><title>Replatform or fix? How to tell</title><link>https://zapolu.com/blog/replatform-or-fix/</link><guid isPermaLink="true">https://zapolu.com/blog/replatform-or-fix/</guid><description>A migration is a year of work that is often sold as a quick fix. How to tell whether the platform is the problem or the implementation is.</description><pubDate>Wed, 24 Jun 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;When a store has problems everywhere (slow pages, high-risk
upgrades, every change more expensive than it should be),
replatforming starts to look like the solution. A fresh platform, a
clean codebase, all the accumulated technical debt left behind.
Agencies rarely advise against it. A migration is among the largest
projects they can sell.&lt;/p&gt;
&lt;p&gt;The arithmetic is worth stating plainly: a real migration is typically
six to twelve months of work for a mid-size store, and longer for large
or heavily customized ones, a feature freeze while it happens, an SEO
risk at cutover, and reworking the platform side of every integration
coupled to the platform’s APIs. Sometimes
it’s worth all of that. But the decision deserves a diagnosis first.&lt;/p&gt;
&lt;h2 id=&quot;most-problems-originate-in-the-implementation&quot;&gt;Most problems originate in the implementation&lt;/h2&gt;
&lt;p&gt;The first question to settle: &lt;strong&gt;are the problems caused by the
platform, or by what was built on it?&lt;/strong&gt; In our experience the
majority of “we’ve outgrown the platform” cases are actually:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;An accumulated extension stack.&lt;/strong&gt; Dozens of modules, some abandoned, some
conflicting, each added for one campaign years ago. The slowness
comes from the modules, not from the platform.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Unmaintained hosting.&lt;/strong&gt; The store sits on infrastructure sized
and configured once, long ago. Caches misconfigured or missing,
no CDN, a database whose indexes have never been reviewed.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;A frontend problem that gets blamed on the platform.&lt;/strong&gt; On Magento
specifically, &lt;a href=&quot;/blog/hyva-vs-luma/&quot;&gt;Luma’s architecture&lt;/a&gt; makes the
whole store appear slow even when the backend performs well.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;One catastrophic customization.&lt;/strong&gt; A single early fork of core
behavior that made every subsequent upgrade dangerous. Expensive,
but removable for a fraction of a migration’s cost.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;None of these disappear on a new platform. A migration run by the same
process that created them reproduces them, and costs a year getting
there.&lt;/p&gt;
&lt;h2 id=&quot;when-the-platform-genuinely-is-the-constraint&quot;&gt;When the platform genuinely is the constraint&lt;/h2&gt;
&lt;p&gt;Real platform ceilings exist, and pretending otherwise is the
opposite error. The honest signals:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;End of life.&lt;/strong&gt; Magento 1 today, others tomorrow. Once official
security patches stop, the migration is almost certainly coming.
Unofficial third-party patching can buy time, but only delays it.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;An architectural ceiling you’ve actually hit.&lt;/strong&gt; Not “the demo of
the other platform looks faster,” but a named requirement the
platform cannot meet: checkout customization the tier doesn’t
allow, B2B workflows extended beyond their limits, a catalog model
that fundamentally doesn’t fit.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The upgrade path is gone.&lt;/strong&gt; So many skipped versions and custom
patches that getting current costs as much as leaving.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;The economics inverted.&lt;/strong&gt; License plus specialist maintenance
costs more per year than the platform returns in capability.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;One signal deliberately missing here is “the developers prefer the
other one.” That’s a tie-breaker at best.&lt;/p&gt;
&lt;h2 id=&quot;a-measured-answer-before-the-expensive-decision&quot;&gt;A measured answer before the expensive decision&lt;/h2&gt;
&lt;p&gt;You don’t have to decide from opinion. An
&lt;a href=&quot;/services/technical-audit/&quot;&gt;initial technical audit&lt;/a&gt; settles the underlying
question (where the problems originate) with measurements: how
much of the slowness is backend versus frontend, which extensions
cost what, what the hosting configuration fails to deliver. Then price the
two paths side by side: fix-in-place as the report’s findings, by
severity, with the recommended course of action and an indicative cost
estimate, versus &lt;a href=&quot;/services/platform-migration/&quot;&gt;migration&lt;/a&gt;
with data, integrations, and SEO preservation costed in.&lt;/p&gt;
&lt;p&gt;Sometimes the audit says migrate. We run migrations, and we’ll say
so plainly when the platform really is the ceiling. But we’d rather
base that recommendation on measurements than on one bad quarter’s
frustration. If you’re facing this decision right now,
&lt;a href=&quot;/contact/&quot;&gt;get in touch&lt;/a&gt; before you sign a migration contract.&lt;/p&gt;</content:encoded><category>platform-migration</category><category>strategy</category></item><item><title>Webhooks retry. Your integration should assume it.</title><link>https://zapolu.com/blog/webhooks-retry/</link><guid isPermaLink="true">https://zapolu.com/blog/webhooks-retry/</guid><description>Payment and platform webhooks are delivered at least once within the retry window, not exactly once. What double delivery breaks, and how idempotent handlers absorb it.</description><pubDate>Wed, 17 Jun 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Read the webhook documentation of any serious provider and you find
the same warning in one form or another. Stripe tells you to log the
event IDs you have processed, because an endpoint can receive the same
event more than once. Adyen tells you to expect the same event twice
and to handle the duplicates. Shopify gives you an
&lt;code&gt;X-Shopify-Webhook-Id&lt;/code&gt; header to deduplicate on, and says that
delivery is not guaranteed at all. All three checked on July 13, 2026:
&lt;a href=&quot;https://docs.stripe.com/webhooks&quot;&gt;Stripe&lt;/a&gt;,
&lt;a href=&quot;https://docs.adyen.com/development-resources/webhooks/handle-webhook-events&quot;&gt;Adyen&lt;/a&gt;,
&lt;a href=&quot;https://shopify.dev/docs/apps/build/webhooks&quot;&gt;Shopify&lt;/a&gt;.&lt;/p&gt;
&lt;p&gt;Within the retry window the contract is &lt;em&gt;at-least-once&lt;/em&gt;, not
&lt;em&gt;exactly-once&lt;/em&gt;. If your endpoint stays down for the whole window, the
event arrives zero times, which is why a reconciliation job that polls
the provider’s API is the backstop underneath everything else.
Exactly-once delivery over an unreliable network is not achievable, so
providers take the safe option: when they cannot confirm you received
an event, they send it again.&lt;/p&gt;
&lt;p&gt;Most integrations are written as if that clause didn’t exist. They
work in the demo, they work in month one, and then a timeout happens
somewhere between the provider and a 200 response, and the same
&lt;code&gt;order.paid&lt;/code&gt; event arrives twice.&lt;/p&gt;
&lt;h2 id=&quot;what-double-delivery-actually-breaks&quot;&gt;What double delivery actually breaks&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;The customer receives two order confirmations.&lt;/li&gt;
&lt;li&gt;Stock gets decremented twice, and incorrect “out of stock” listings
start appearing.&lt;/li&gt;
&lt;li&gt;An invoice is issued twice, and accounting finds it months later.&lt;/li&gt;
&lt;li&gt;A loyalty balance is credited twice. Nobody notices until the
numbers stop reconciling.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The provider did what it documented. The handler assumed a guarantee
the provider never made.&lt;/p&gt;
&lt;h2 id=&quot;idempotency-is-the-fix&quot;&gt;Idempotency is the fix&lt;/h2&gt;
&lt;p&gt;An idempotent handler produces the same end state whether an event is
processed once or five times. That single property absorbs retries,
manual replays, and a class of race conditions. Achieving it is
mostly discipline:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Key every event.&lt;/strong&gt; Providers send an event ID. If yours doesn’t,
derive one from stable fields. Record processed IDs and skip
duplicates, and enforce that with a unique constraint in the
database rather than in application code: the application check
has a race window. The constraint doesn’t.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Prefer absolute state over relative changes.&lt;/strong&gt; “Set status to
paid” is naturally idempotent. “Decrement stock by 2” is not.
When the provider sends deltas, translate them into upserts
against your own record of the event, not unconditional arithmetic.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Assume disorder, too.&lt;/strong&gt; Retries also arrive out of order:
&lt;code&gt;order.updated&lt;/code&gt; can land before &lt;code&gt;order.created&lt;/code&gt;. Version numbers
or timestamps on the target record (“apply only if newer”) handle
what sequence assumptions can’t.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id=&quot;the-rest-of-the-checklist&quot;&gt;The rest of the checklist&lt;/h2&gt;
&lt;p&gt;Idempotency is the core. Three habits around it cover the rest:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Acknowledge fast, process async.&lt;/strong&gt; Verify the signature, persist
the event, return 200, and do the real work from a queue. Slow
synchronous handlers cause the timeouts that cause the retries you
were worried about.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Authenticate every webhook.&lt;/strong&gt; A webhook endpoint is an
unauthenticated URL on the public internet that mutates your
business state. Most providers sign the payload with an HMAC you
recompute and compare. Some do not send a payload at all: a GoPay
notification is an HTTP GET carrying the payment ID, and the
authoritative state has to be read back from the GoPay API before
you act on it (&lt;a href=&quot;https://help.gopay.com/en/knowledge-base/integration-of-payment-gateway/integration-of-payment-gateway-1/state-of-payment&quot;&gt;GoPay&lt;/a&gt;,
checked on July 13, 2026). Either way, never act on the raw request
as received.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Keep a dead-letter queue with a replay tool.&lt;/strong&gt; Some events will
fail for reasons retry can’t fix: a deleted SKU, a mapping gap.
Route them there, alert on them, and make it possible to replay a
single event without SSH-ing into production. When a provider has
an outage lasting hours, replay lets you catch up afterwards
without declaring an incident.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;One test covers all of the above: take yesterday’s production
events and deliver every one of them to staging twice, in shuffled
order. If staging ends in the same state as production, your handlers
really are idempotent.&lt;/p&gt;
&lt;p&gt;This is the standard we build integrations to. It is most of what our
&lt;a href=&quot;/services/integrations/&quot;&gt;integrations service&lt;/a&gt; page describes, and
the same thinking shows up in our
&lt;a href=&quot;/services/custom-development/&quot;&gt;custom backend work&lt;/a&gt;. And if your
stock numbers already drift, the webhook layer is one of the
&lt;a href=&quot;/blog/erp-eshop-stock-mismatch/&quot;&gt;first places to look&lt;/a&gt;.&lt;/p&gt;</content:encoded><category>integrations</category><category>webhooks</category><category>reliability</category></item><item><title>Why your ERP and your online store disagree about stock</title><link>https://zapolu.com/blog/erp-eshop-stock-mismatch/</link><guid isPermaLink="true">https://zapolu.com/blog/erp-eshop-stock-mismatch/</guid><description>The synchronization failure every store eventually encounters: two systems that both believe they own the stock number, and the patterns that resolve it.</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Sooner or later every store with an ERP encounters the same support
ticket: the storefront says five in stock, the warehouse says zero,
and a customer has just paid for something that doesn’t exist. The
support team treats it as a one-off error, but it traces back to an
architecture decision nobody remembers making.&lt;/p&gt;
&lt;h2 id=&quot;the-root-cause-is-ownership&quot;&gt;The root cause is ownership&lt;/h2&gt;
&lt;p&gt;When the numbers diverge, everyone debugs the synchronization script.
But the script is usually doing exactly what it was told. The problem
is that two systems both believe they own the stock number. The ERP
decrements when it books an order. The store decrements when checkout
completes. Both are right by their own rules, and the sync job has no
rule for deciding which value prevails.&lt;/p&gt;
&lt;p&gt;The fix is one principle: &lt;strong&gt;one system of record per fact.&lt;/strong&gt; Stock
truth lives in exactly one place, almost always the ERP or the WMS,
because that’s where stock is booked and where goods physically move. The store holds a cached
copy for display and a &lt;em&gt;reservation&lt;/em&gt;.
Once that’s explicit, every sync question has an answer: when in
doubt, the system of record wins.&lt;/p&gt;
&lt;h2 id=&quot;the-recurring-failure-modes&quot;&gt;The recurring failure modes&lt;/h2&gt;
&lt;p&gt;With ownership settled, the remaining bugs are mechanical, and they
repeat across platforms and ERPs:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Lost updates.&lt;/strong&gt; The ERP pushes a stock change, the store is
mid-deploy or rate-limits the call, and nobody retries. Weeks later
someone asks why one SKU has been “out of stock” all month.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Duplicate deliveries.&lt;/strong&gt; The same webhook fires twice (webhook
delivery is typically at-least-once, so &lt;a href=&quot;/blog/webhooks-retry/&quot;&gt;the same event can arrive
more than once&lt;/a&gt;), and the
handler decrements stock twice.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Full syncs racing deltas.&lt;/strong&gt; A nightly full import and event-driven
updates run against the same rows. Whichever finishes last wins,
and it’s sometimes the stale one.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Mapping drift.&lt;/strong&gt; A SKU renamed in the ERP silently becomes a new
product in the store. The old one keeps its stale count forever.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Bundles and kits.&lt;/strong&gt; The store sells a set. The ERP counts
components. If the set-to-component conversion lives in a
spreadsheet, expect it to be out of date.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id=&quot;the-patterns-that-resolve-it&quot;&gt;The patterns that resolve it&lt;/h2&gt;
&lt;p&gt;The integration layer that stops producing these tickets looks the
same regardless of which ERP is behind it:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Deltas plus reconciliation.&lt;/strong&gt; Event-driven updates for speed, and
a periodic full comparison that finds what the events missed. The
reconciliation is what makes the fast path trustworthy.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Idempotent consumers.&lt;/strong&gt; Every update carries an identifier the
consumer deduplicates on, so processing it twice changes nothing.
This one property absorbs
retries and replays. Concurrency conflicts are a separate problem:
lost updates and full-syncs-racing-deltas need version stamps or
sequence numbers so a stale write can’t overwrite a newer one.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Correlation IDs end to end.&lt;/strong&gt; When order 18342 never reached the
warehouse, “where did it go” should be answerable from the logs in
a minute or two.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Alert on divergence.&lt;/strong&gt; A sync that reports
success while numbers drift apart is worse than one that fails
outright.
Compare aggregates on a schedule and page someone when the delta
grows.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Reservations with expiry.&lt;/strong&gt; If your setup reserves stock early,
a cart holds it for minutes and an unpaid order for longer. By
default, though, reservation happens at order placement, not
add-to-cart. Either way, everything expires back to the
system of record. Overselling during campaigns usually traces to
reservations that never expire or never existed.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;None of this depends on the vendor. The same architecture works against
NetSuite, Business Central, Odoo, and the Czech and Slovak systems like
Pohoda and Money S5 that international connector vendors tend to ignore.
That long tail is a large part of why
&lt;a href=&quot;/services/integrations/&quot;&gt;our integration work&lt;/a&gt; exists.&lt;/p&gt;
&lt;p&gt;If your store and your ERP currently disagree about stock, the place
to start is making the ownership question explicit. If you want
help, &lt;a href=&quot;/contact/&quot;&gt;contact us&lt;/a&gt;. Deciding where each fact
lives is &lt;a href=&quot;/services/system-design/&quot;&gt;system design&lt;/a&gt;. Once that
decision is made, building the sync is straightforward.&lt;/p&gt;</content:encoded><category>integrations</category><category>erp</category><category>architecture</category></item><item><title>Why your checkout collapses at 10% server load</title><link>https://zapolu.com/blog/checkout-collapse-under-load/</link><guid isPermaLink="true">https://zapolu.com/blog/checkout-collapse-under-load/</guid><description>Servers idle, checkout down. Lock contention, cache stampedes, and synchronous payment calls: the failure modes autoscaling does not address.</description><pubDate>Thu, 04 Jun 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;It’s one of the most confusing incidents an e-commerce team can have. The
campaign goes live, checkout starts timing out, and the dashboards
insist nothing is wrong: CPU at 10%, memory fine, plenty of headroom.
The instinct is to add servers. The servers arrive, idle alongside
the others, and checkout is still down.&lt;/p&gt;
&lt;p&gt;The confusion comes from a wrong mental model. Load has two
components: work and waiting. Your servers are queueing rather than
working, and queues don’t show up on a CPU graph.&lt;/p&gt;
&lt;h2 id=&quot;the-common-causes&quot;&gt;The common causes&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;Database lock contention.&lt;/strong&gt; Checkout is where the store stops
reading and starts writing: reserve stock, create the order, save the
address. Those writes serialize on row locks, and campaign traffic
concentrates them on the same few hot rows: the order-sequence
counters, the order and quote tables, the stock being reserved. Each
transaction holds its locks while it waits on other work. Throughput
collapses while CPU stays low. On Magento it surfaces in the logs as
MySQL lock-wait timeout and deadlock errors.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Cache stampedes.&lt;/strong&gt; A hot cache entry (the category page every ad
points at) expires under peak traffic. Every request that misses now
regenerates the page simultaneously, issuing identical expensive
queries against the database. The database receives one copy of that
work per concurrent request.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Synchronous third-party calls in the checkout path.&lt;/strong&gt; Shipping
rates, payment authorization, tax calculation, an anti-fraud check:
each a network call to someone else’s system, made while your worker
holds a connection and often a database transaction. When one
provider slows from 200 ms to 10 seconds under the same campaign
load, every worker gradually ends up blocked inside that call.
Checkout then stays down for as long as that provider is slow.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Worker and connection exhaustion.&lt;/strong&gt; The three problems above share a
common amplifier: workers that wait still occupy their slot and their
database connection. PHP-FPM processes sized for ordinary weekday
traffic fill with blocked requests, new requests can’t get a worker or
a connection at all, and now &lt;em&gt;everything&lt;/em&gt; that still needs a PHP worker
is down, however little work that page does.&lt;/p&gt;
&lt;p&gt;None of the four is a hardware problem, which is why autoscaling
doesn’t resolve it: new instances join the same queue for the same
row lock behind the same slow vendor. The added capacity waits in
parallel instead of adding throughput.&lt;/p&gt;
&lt;h2 id=&quot;how-to-see-it-coming&quot;&gt;How to see it coming&lt;/h2&gt;
&lt;p&gt;The diagnostic shift is from &lt;em&gt;utilization&lt;/em&gt; to &lt;em&gt;time&lt;/em&gt;: for a checkout
request, where do its milliseconds go? Working, or waiting on locks,
caches, and third-party APIs? Distributed traces answer this
directly. A slow-query log plus the database’s lock-wait counters
answer it cheaply. If you only add one graph before the next
campaign, make it lock waits.&lt;/p&gt;
&lt;p&gt;The fixes follow from the diagnosis, and most are routine:
timeouts on every third-party call plus cached fallbacks wherever a
stale answer is safe (a shipping rate a few minutes old does no harm,
though a payment authorization must still go live), stampede protection so
one process rebuilds a hot key while others serve stale, moving
non-essential writes out of the transaction and into a queue, and
reserving stock as late in the funnel as the business allows rather
than holding it from the cart.
Which of these matters for your store is an empirical question. An
&lt;a href=&quot;/services/technical-audit/&quot;&gt;initial technical audit&lt;/a&gt; answers it from
your own traces, and our &lt;a href=&quot;/services/devops/&quot;&gt;DevOps work&lt;/a&gt; keeps the
fixes in place.&lt;/p&gt;
&lt;p&gt;Most of these incidents turn out to be concurrency problems, so
start with a trace before you buy more servers.&lt;/p&gt;</content:encoded><category>performance</category><category>infrastructure</category><category>reliability</category></item><item><title>Magento Open Source vs. Mage-OS vs. Adobe Commerce</title><link>https://zapolu.com/blog/magento-vs-mageos-vs-adobe-commerce/</link><guid isPermaLink="true">https://zapolu.com/blog/magento-vs-mageos-vs-adobe-commerce/</guid><description>Three live branches of one platform: where they came from, what each buys you, and how to choose without tying the store to someone else&apos;s roadmap.</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;If you run a store on Magento, you are choosing between three options
that share a codebase and a name: Magento Open Source, Mage-OS, and
Adobe Commerce. The differences are less about features than about
governance: whose priorities decide what the platform becomes. That
makes this more a business decision than a technical one.&lt;/p&gt;
&lt;p&gt;Every vendor fact below links to the vendor’s own documentation,
checked on July 13, 2026.&lt;/p&gt;
&lt;h2 id=&quot;how-the-three-branches-emerged&quot;&gt;How the three branches emerged&lt;/h2&gt;
&lt;p&gt;Adobe bought Magento in 2018. The commercial edition became Adobe
Commerce. The free edition stayed available as Magento Open Source.
Adobe now ships one Magento release line a year, 2.4.8 in April 2025
and 2.4.9 in May 2026, with security patches in between
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-operations/release/versions&quot;&gt;Adobe’s released versions&lt;/a&gt;),
and the roadmap follows Adobe’s commercial interests (understandably,
from Adobe’s perspective).&lt;/p&gt;
&lt;p&gt;In 2022, part of the community responded by founding the Mage-OS
Association, which publishes &lt;a href=&quot;https://mage-os.org&quot;&gt;Mage-OS&lt;/a&gt;: a
community-governed distribution of Magento Open Source with its own
release process. The distribution is deliberately compatible: a
Mage-OS store is still a Magento store, extensions still work,
developers do not need retraining. What changed is who decides, and
how fast fixes land.&lt;/p&gt;
&lt;h2 id=&quot;what-each-one-offers&quot;&gt;What each one offers&lt;/h2&gt;
&lt;p&gt;&lt;strong&gt;Magento Open Source&lt;/strong&gt; is the default you get by not deciding. It is
free and it works, at whatever pace Adobe assigns it. The
risk is a slow drift, where the fixes you need sit behind releases
scheduled around a product you did not buy.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Mage-OS&lt;/strong&gt; is the same platform under community governance. The
association’s stated policy is to ship patches once the community has
validated them rather than wait for a corporate release schedule, and
the project’s incentives point at open-source merchants rather than
enterprise licensing (&lt;a href=&quot;https://mage-os.org/faq/&quot;&gt;Mage-OS FAQ&lt;/a&gt;). We
have not measured that against Adobe’s cadence ourselves. The
trade-off is that Mage-OS is younger as an institution: you are
trusting an association instead of a corporation, which carries
different risks rather than fewer.&lt;/p&gt;
&lt;p&gt;&lt;strong&gt;Adobe Commerce&lt;/strong&gt; is a different commercial proposition: managed
cloud hosting with SLAs, enterprise support contracts, and extensions
Adobe ships to nobody else. The largest of those is Adobe Commerce
B2B, which covers company accounts, shared catalogs, negotiable
quotes and requisition lists. Read the licensing carefully: Adobe
lists B2B as an Adobe Commerce extension that requires a separate
license, so it is not something the base Adobe Commerce fee already
includes
(&lt;a href=&quot;https://experienceleague.adobe.com/en/docs/commerce-admin/start/resources/extensions&quot;&gt;Extensions from Adobe&lt;/a&gt;).
If your requirements genuinely use all this (real B2B purchasing,
procurement demanding a vendor SLA), the licenses buy concrete
capabilities. If they do not, you are paying an enterprise license
fee for the open-source platform plus features you will disable.&lt;/p&gt;
&lt;h2 id=&quot;choosing-between-them&quot;&gt;Choosing between them&lt;/h2&gt;
&lt;p&gt;Three questions settle most cases:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Do you need Adobe Commerce B2B or a contractual SLA?&lt;/strong&gt; If yes,
that is Adobe Commerce, and the licenses are the cost of those
requirements.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Are you on Magento Open Source today?&lt;/strong&gt; Moving to Mage-OS is a
small, low-risk step (same extensions, same developers) that hands
the release schedule to a project whose priorities are the same as
yours. This is why our new open-source builds go to Mage-OS by
default.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Are you choosing the platform fresh?&lt;/strong&gt; Then first decide whether
Magento’s model fits at all: complex catalogs, B2B, multi-store.
If your requirements are simpler, &lt;a href=&quot;/services/shopify/&quot;&gt;Shopify&lt;/a&gt; is
usually the cheaper answer, and this question becomes irrelevant.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;One thing we would warn against: staying on an old Magento Open Source
version to defer the decision. Adobe’s Magento security bulletins
include vulnerabilities that need no login and that Adobe has
confirmed were exploited in the wild, CVE-2024-34102 among them
(&lt;a href=&quot;https://helpx.adobe.com/security/products/magento/apsb24-40.html&quot;&gt;APSB24-40&lt;/a&gt;).
An unpatched store stays exposed to those until it is upgraded, and the
upgrade gets more expensive every quarter you wait, the same dynamic
that made &lt;a href=&quot;/services/platform-migration/&quot;&gt;Magento 1 stores&lt;/a&gt; such a risk
after support ended in June 2020.&lt;/p&gt;
&lt;p&gt;We work with both Mage-OS and Adobe Commerce and have no license
revenue tied to your choice. The reasoning above is the same
reasoning we walk clients through on our
&lt;a href=&quot;/services/magento/&quot;&gt;Magento service page&lt;/a&gt;. If you want it applied to
your specific store, extensions and all, that is a conversation we are
happy to have. &lt;a href=&quot;/contact/&quot;&gt;Book an intro call&lt;/a&gt;.&lt;/p&gt;</content:encoded><category>magento</category><category>mageos</category><category>platforms</category></item><item><title>Hyvä vs. Luma, measured</title><link>https://zapolu.com/blog/hyva-vs-luma/</link><guid isPermaLink="true">https://zapolu.com/blog/hyva-vs-luma/</guid><description>What actually makes Luma slow, what Hyvä changes, and how to measure the difference on your own catalog instead of relying on vendor claims.</description><pubDate>Thu, 21 May 2026 00:00:00 GMT</pubDate><content:encoded>&lt;p&gt;Every Magento performance conversation eventually arrives at the same
question: is the frontend the problem, and is Hyvä the answer? The
honest answer to both is usually yes. It is still worth verifying on
your own store rather than taking it on trust from us or from anyone
selling you a rebuild. This post explains what’s structurally
different between the two frontends and how to measure the gap before
you spend money on it.&lt;/p&gt;
&lt;h2 id=&quot;why-luma-is-slow-by-design&quot;&gt;Why Luma is slow by design&lt;/h2&gt;
&lt;p&gt;Luma, the stock Magento frontend, is slow because of decisions baked
into the platform around 2015:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;RequireJS loads JavaScript as a dependency waterfall.&lt;/strong&gt; The browser
discovers what to fetch next only after executing what it already
fetched. On a high-latency mobile connection, that’s round-trip after
round-trip before the page becomes interactive.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Multiple heavy JS layers ship at once.&lt;/strong&gt; jQuery, KnockoutJS, and
Magento’s Knockout-based UI-component layer all arrive on every page,
whether the page uses them or not.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Layout XML renders more than the page needs.&lt;/strong&gt; Blocks nobody sees still get
built, and third-party extensions attach their scripts globally
rather than where they’re needed.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;None of this appears in a profiler as one slow line, but it does appear
as megabytes of JavaScript and a main thread that stays busy for
seconds on the mid-range phones your customers use.&lt;/p&gt;
&lt;h2 id=&quot;what-hyvä-actually-changes&quot;&gt;What Hyvä actually changes&lt;/h2&gt;
&lt;p&gt;&lt;a href=&quot;https://hyva.io&quot;&gt;Hyvä&lt;/a&gt; is a replacement theme. It removes the Luma
frontend stack entirely, without RequireJS, jQuery or Knockout.
Templates are server-rendered PHP with &lt;a href=&quot;https://alpinejs.dev&quot;&gt;Alpine.js&lt;/a&gt;
for interactivity and Tailwind for styles. In our own measurements the
JavaScript payload typically drops from somewhere north of a megabyte
to well under a hundred kilobytes, something you can verify for
yourself in any browser’s network tab on the demo store.&lt;/p&gt;
&lt;p&gt;The trade-off: every extension that touches the frontend needs a
Hyvä-compatible module or a rewrite. The compatibility ecosystem is
mature for popular extensions and often nonexistent for older bespoke
modules. The extension audit is where most of a Hyvä project’s cost
sits.&lt;/p&gt;
&lt;p&gt;Checkout is a separate decision. A stock Hyvä install keeps the Luma
checkout, through Hyvä’s
&lt;a href=&quot;https://docs.hyva.io/hyva-themes/building-your-theme/luma-theme-fallback.html&quot;&gt;Luma theme fallback&lt;/a&gt;.
&lt;a href=&quot;https://docs.hyva.io/hyva-checkout/index.html&quot;&gt;Hyvä Checkout&lt;/a&gt; is a
separate product (Hyvä’s documentation, checked July 22, 2026).
Checkout is often the page with the largest impact on conversion.&lt;/p&gt;
&lt;h2 id=&quot;measuring-it-yourself&quot;&gt;Measuring it yourself&lt;/h2&gt;
&lt;p&gt;Don’t compare your production store against a Hyvä demo. That
conflates theme, catalog, hosting, and third-party scripts. A fair
comparison looks like this:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Get your baseline from field data, not just Lighthouse.&lt;/strong&gt; CrUX
(via PageSpeed Insights) shows what real visitors experienced over
28 days: LCP, INP, CLS. An unthrottled lab run on a fast developer
machine can look far better than what real visitors on mid-range
phones get.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Separate frontend from backend.&lt;/strong&gt; TTFB is dominated by the backend
and network (server, DNS, TLS). If TTFB is 1.5 s, a new theme won’t
fix it. Most of what happens after TTFB is what a frontend swap can
address. A slow backend or network still needs its own fix.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Stage the comparison honestly.&lt;/strong&gt; Same server, same catalog, same
extensions enabled, third-party tags either on in both or off in
both. Run Lighthouse several times and compare medians. Single
runs are noise.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Track the business metric alongside.&lt;/strong&gt; Record conversion by device
class before and after, and judge the project on that rather than on
a Lighthouse score.&lt;/li&gt;
&lt;/ol&gt;
&lt;h2 id=&quot;cases-against-hyvä&quot;&gt;Cases against Hyvä&lt;/h2&gt;
&lt;p&gt;Hyvä is our default recommendation for Luma stores with a performance
problem, but not for every store. A store with 4-second TTFB has a backend
problem. A store that’s already on PWA Studio faces a different
set of trade-offs. And a store about to replatform shouldn’t invest in either.
That’s a &lt;a href=&quot;/blog/replatform-or-fix/&quot;&gt;replatform-or-fix&lt;/a&gt; question first.&lt;/p&gt;
&lt;p&gt;If you want the measurement done for you, our
&lt;a href=&quot;/services/technical-audit/&quot;&gt;initial technical audit&lt;/a&gt; covers it:
performance and extensions are two of its six areas, and the written
report ranks each finding by severity. It’s also the usual first step of
our &lt;a href=&quot;/services/magento/&quot;&gt;Magento work&lt;/a&gt;.&lt;/p&gt;</content:encoded><category>magento</category><category>performance</category><category>hyvä</category></item></channel></rss>