Magento Open Source Adobe Commerce Storefront design & front-end Independent · not affiliated with Adobe
Magento Themes Store Storefront Lab
Published 13 Nov 2026 10 min read

Magento hosting in 2026: the stack 2.4.9 actually requires

Hosting decisions turn on a short support matrix, a changed search and cache stack, and workload peaks that ordinary traffic estimates miss.

The supported stack for Magento 2.4.9

Magento Open Source and Adobe Commerce 2.4.9 have a deliberately narrow list of supported platform combinations. For an on-premises installation, Adobe lists PHP 8.5, MariaDB 12.3 or MySQL 8.4, OpenSearch 3, Valkey 9, RabbitMQ 4.3 and Composer 2.10.

Those are not broad suggestions. They are the combinations Adobe identifies as supported for 2.4.9. A hosting provider can run other software, configure it differently or support a newer release separately, but that arrangement should not be presented as the supported 2.4.9 matrix.

The headline change for stores moving from Magento 2.4.6 or 2.4.7 is that Elasticsearch and Redis are no longer listed as supported for 2.4.9. OpenSearch 3 replaces Elasticsearch in the supported search position, while Valkey 9 occupies the cache and session position.

This does not mean an Elasticsearch or Redis installation will instantly fail. It means Adobe does not list either as a supported part of the 2.4.9 combination. A host offering those services as its default Magento stack is therefore asking the merchant to accept unsupported infrastructure choices.

Component Supported version Magento responsibility
PHP 8.5 Runs Magento application code and the command-line tools used for administration and deployment.
Database MariaDB 12.3 or MySQL 8.4 Stores catalogue, customer, order, checkout and administrative data.
OpenSearch 3 Indexes catalogue data for storefront search and relevant administrative search operations.
Valkey 9 Provides the supported cache and session layer.
RabbitMQ 4.3 Transfers work to consumers that perform asynchronous Magento tasks.
Composer 2.10 Manages PHP dependencies during installation, deployment and updates.

Why Magento needs a dedicated search engine

The database is Magento’s system of record, but it is not the best tool for every search request. A storefront search can involve product text, attribute filters, price ranges, category constraints, visibility rules, stock considerations and sorting. An index is intended to hold a representation designed for those retrieval operations.

OpenSearch 3 therefore has a different job from MariaDB or MySQL. The database remains authoritative. OpenSearch receives indexed catalogue data and answers the queries that need to find matching products quickly.

The distinction matters when planning infrastructure. Search needs its own memory, disk and CPU allocation. It also introduces operational work: indexes must be kept current, mappings need to remain compatible with the platform, and the search service must be monitored for failures and degraded responses.

A Magento search index is not simply a copy that can be ignored after initial creation. Catalogue changes, price updates, stock changes and other data updates can require index work. A host that provisions OpenSearch but leaves indexing to an improvised cron job has not completed the deployment.

Search also affects admin work. Administrators searching the catalogue rely on indexed data, so a weak search service can slow routine merchandising and troubleshooting as well as customer-facing requests. Its capacity should be considered separately from the database’s capacity.

What Valkey holds

Valkey 9 is the supported cache and session component in the 2.4.9 on-premises matrix. Magento uses the cache to avoid repeating expensive work and the session layer to keep state associated with a visitor or authenticated user.

Cached data can include rendered or prepared blocks, configuration values and other results that are expensive to recreate. Session data supports continuity through a shopping journey, including basket and checkout state. Losing either can appear as poor performance, broken journeys or both.

Cache and session storage should not be treated as disposable scratch space. Persistence, eviction behaviour, memory limits, backups and recovery procedures all affect what happens under pressure. The relevant question is not whether Valkey is installed, but whether the host protects the data Magento expects it to retain.

Removing Redis from Adobe’s supported list is not a cosmetic edit to a compatibility table. The architecture now needs to be assessed around Valkey rather than assuming that an existing Redis configuration carries across without validation.

Nor should the cache be allowed to grow until the host is forced to evict it. A sensible managed Magento hosting service defines limits and monitoring, then tests whether critical data remains available when memory pressure occurs.

What RabbitMQ is doing

RabbitMQ 4.3 is the supported message broker in the 2.4.9 matrix. It allows Magento to hand work to consumers instead of making an HTTP request wait for every related task to finish.

Magento’s asynchronous work includes operations such as product and index updates, email processing and cache-related processing. The exact queue arrangement depends on the installation, but the principle remains: the web request should not carry out all of that work in the foreground.

A queue is not proof that the work has been completed. Messages can accumulate, consumers can fail and jobs can remain blocked by a later task. Queue depth, consumer status and processing failures need to be visible and actionable.

RabbitMQ also changes failure diagnosis. An empty page, stale catalogue data or delayed email may originate in a consumer rather than in PHP or the database. A host should be able to show where a queued job has reached, why it has stopped and whether it can be retried safely.

Indexing can be implemented through Magento command-line work, cron-driven commands or queue consumers, depending on the operating model. That choice affects where failures appear and how quickly a backlog can clear. It should be documented rather than left for the merchant to infer from the server layout.

Why cron is not optional

Magento relies on scheduled execution for recurring maintenance. Cron starts the scheduled tasks defined by the application, including processes connected to catalogue indexing and other periodic jobs.

Running the cron command does not guarantee that every scheduled task succeeds. PHP errors, permissions, execution time limits, queue failures and overlapping jobs can all interrupt the schedule. The distinction is important: a host may show a green cron endpoint while important jobs remain stale.

A useful check is to inspect recent cron runs, the tasks due, their outcomes and the time taken to complete them. Administrators also need to know whether a missed run is retried, whether jobs overlap and whether long-running indexing blocks the next schedule.

Cron must also work when no browser request is arriving. Some hosts schedule work from web traffic or run cron only while the application receives traffic. That model fails during quiet periods and precisely when a backlog needs to be cleared.

A Magento 1 store cannot remain a long-term hosting solution: Magento 1 reached end of life on 30 June 2020. Migration planning should include the destination’s queue, search, cache, cron and deployment processes, rather than treating the move as a file transfer.

Sizing from work, not traffic alone

Magento hosting decisions often begin with visitor numbers. That is useful for web capacity, but it is an incomplete description of the workload. The questions that usually matter are how large the catalogue is, how often it changes, how many administrators work at once and which jobs run during the busiest period.

Catalogue size affects database storage, search indexing, import work and cache behaviour. A catalogue with many simple products does not create the same load as one with extensive variants, complex attributes and frequent updates. The relevant test is the shape of the data and the rate at which it changes.

Concurrent administrative users need their own capacity consideration. Store administrators can trigger imports, catalogue updates, index operations and reports while customers browse. Heavy administrative work can therefore compete with storefront traffic even when the number of shoppers is modest.

Concurrent front-end requests tell a different part of the story. Search queries, cache misses and checkout operations place different demands on PHP, the database, OpenSearch and Valkey. A server that handles ordinary browsing may still struggle when many requests miss the cache or arrive as search-intensive traffic.

The busiest moment is often not the largest sale. A sale adds a relatively small amount of persistent data compared with a full reindex, which may rewrite substantial search data and trigger related work. A large import, a promotional catalogue update or a postponed index can therefore create a sudden resource demand unrelated to the order volume.

Indexing is not just CPU work. It can involve reading from the database, writing to OpenSearch, updating cache data and passing messages through RabbitMQ. Storage throughput, memory and queue processing can constrain the operation at different points.

Ask the host how it separates interactive work from maintenance. A capable provider can identify long-running index and import jobs, avoid uncontrolled overlap and preserve capacity for checkout and browsing while those jobs run.

The practical test is to reproduce the merchant’s actual catalogue and update cycle. A staging environment should be able to process a representative index build while controlled storefront and admin activity continues. The result should be observed across PHP, database, search, cache and queue metrics.

Watch queue depth and index freshness as well as response times. A dashboard showing only requests per second can look healthy while jobs are piling up or search results remain stale.

Questions to ask a managed Magento host

  • Which exact platform versions are included? The answer should identify PHP 8.5, MariaDB 12.3 or MySQL 8.4, OpenSearch 3, Valkey 9, RabbitMQ 4.3 and Composer 2.10 for a 2.4.9 on-premises stack, with any deviation stated plainly.
  • Is Elasticsearch or Redis part of the default deployment? The desired answer is that neither is used as a substitute for the supported OpenSearch 3 and Valkey 9 services in the 2.4.9 matrix.
  • How is search indexing operated? The answer should explain the commands, schedules or consumers involved, how overlap is prevented and how index freshness is monitored.
  • How are failed queue messages detected and handled? The desired answer includes visible queue depth, failed-job alerts, consumer health and a documented retry or recovery process.
  • How is cron monitored? The answer should show whether missed runs, failed tasks and long-running schedules are reported, rather than merely confirming that a cron endpoint exists.
  • What happens during a full reindex? The answer should describe resource isolation, expected operational impact and how heavy indexing is prevented from degrading checkout or normal catalogue editing.
  • How are catalogue size and concurrency used in capacity planning? The answer should ask about products, variants, attributes, update frequency, administrative users and peak requests, not only monthly visits.
  • What is the backup and restore position? The desired answer covers the database, search indexes where appropriate, configuration, deployment artefacts and session or cache recovery, including a tested restore procedure.
  • How are updates and rollbacks handled? The answer should identify the deployment sequence for application code, Composer dependencies, configuration, search compatibility and any required index rebuild.
  • Which monitoring and alerting is included? The answer should cover PHP errors and saturation, database pressure, OpenSearch health, Valkey memory and persistence, RabbitMQ backlog, cron failures and certificate or network issues.
  • Can the host provide evidence from a comparable workload? The desired evidence is relevant operational detail, not a sales claim: how the workload was measured, which components were stressed and what the resulting behaviour showed.

The stack must be supported as a whole

Magento 2.4.9 hosting is not a matter of renting a server with PHP and a database on it. The supported combination includes a search engine, cache and session layer, message broker, Composer toolchain and reliable scheduled execution.

The change from Elasticsearch and Redis to OpenSearch and Valkey is the most visible compatibility issue for an upgrading store. The less visible risk is operating those components as an unexplained collection of services without monitoring, ownership or a clear recovery process.

Capacity planning must then include the work Magento performs between requests. A quiet storefront can still be generating index work, queue messages or overdue cron tasks. Managed hosting should account for that background load and make its handling visible before it becomes an incident.

Questions this page answers

What software stack does Magento 2.4.9 support for on-premises installations?
PHP 8.5, MariaDB 12.3 or MySQL 8.4, OpenSearch 3, Valkey 9, RabbitMQ 4.3 and Composer 2.10.
Are Elasticsearch and Redis supported in Magento 2.4.9?
No. Adobe does not list either as supported in the 2.4.9 matrix; OpenSearch 3 replaces Elasticsearch and Valkey 9 occupies the cache and session position.
Why does Magento need OpenSearch?
The database is Magentou2019s system of record, but it is not the best tool for every search request. OpenSearch receives indexed catalogue data and answers queries that need to find matching products quickly.
Why is cron important for Magento hosting?
Magento relies on scheduled execution for recurring maintenance, including catalogue indexing and other periodic jobs. Cron must also work when no browser request is arriving.

More in Hosting & operations

Editorial & corrections: [email protected]