Magento performance checklist: checks you can test yourself
Each row below says what to check, how to check it with a read-only command or a header, and what a healthy result looks like, based on Adobe's own Commerce documentation. Work through it on staging or with read-only commands on production, and change settings only after you have a baseline.

How to use this checklist
Run the checks in order. The first sections decide whether pages come from cache at all, and nothing further down matters much if they do not. Use the read-only commands as written; they show state and change nothing. Run every bin/magento command as the file system owner, as Adobe's CLI documentation requires. Where a row needs a configuration change, make it on staging, measure again with the same URLs, and only then plan it for production.
A note on safety: on a shared cache server use bin/magento cache:clean, not a storage flush. Adobe's documentation explains that cleaning only removes Magento's own enabled cache types, while flushing purges the whole storage and can affect anything else using it.
This page is the do-it-yourself list. For why stores get slow in the first place, read why Magento is slow. For browser metrics in depth, see the Core Web Vitals diagnosis guide. If you would rather have the checks run and the fixes implemented for you, that is the Magento performance optimization service.
1. Baseline before you change anything
| Check | How to check | What good looks like |
|---|---|---|
| Field data for key templates | PageSpeed Insights on one home, category, product and cart URL. Note whether it shows URL data or falls back to origin data. | At the 75th percentile over 28 days: LCP 2.5 s or less, INP 200 ms or less, CLS 0.1 or less, for mobile and desktop. |
| Server time, cache hit vs miss | A first request to a live URL is usually already a hit. Add a unique query parameter that is not a marketing parameter (Magento's VCL strips utm_*, gclid and similar), request it once for the miss and again for the hit, capturing header and time to first byte from the same request (commands below). | The first request shows MISS, the repeat shows HIT and is clearly faster. Both numbers are written down for later comparison. |
| Lab runs you can repeat | Lighthouse several times per URL, same version, form factor, clean browser profile, logged out. Keep the JSON files. | You compare medians across runs, not one score. |
# Cache header and server time from ONE request. First run with a new value: expect MISS
curl -s -o /dev/null -D - -w 'ttfb %{time_starttransfer}s\n' 'https://www.example.com/some-category.html?cachecheck=run1' | grep -i -e x-magento-cache-debug -e ttfb
# Same command again with the same value: expect HIT
curl -s -o /dev/null -D - -w 'ttfb %{time_starttransfer}s\n' 'https://www.example.com/some-category.html?cachecheck=run1' | grep -i -e x-magento-cache-debug -e ttfb
If you are deciding whether a different theme would help, the Magento frontend comparison uses the same measuring method.
2. Application mode and caches
| Check | How to check | What good looks like |
|---|---|---|
| Production mode | bin/magento deploy:mode:show as the file system owner. | production. Adobe describes production mode as the one for live systems, with static files served from pub/static and code compiled. Default mode is not optimised for production, and developer mode is for development work. |
| Cache types enabled | bin/magento cache:status | Every cache type shows 1. A 0 needs a reason someone can explain. |
3. Full page cache
| Check | How to check | What good looks like |
|---|---|---|
| Varnish is the caching application | bin/magento config:show system/full_page_cache/caching_application | 2 (Varnish). 1 is the built-in cache, which Adobe describes as much slower than Varnish and does not recommend for production. |
| Cacheable pages are hits | Run the two commands from section 1 on a home, category and product URL and read X-Magento-Cache-Debug. | HIT on the second request. UNCACHEABLE on a catalog page means Magento sent it as not cacheable, often because a block in its layout is marked cacheable="false". Magento's Varnish VCL sends this header in every mode; the built-in cache only sends it in developer mode, so a missing header there is not a miss. A CDN or custom VCL can remove it; then use the CDN's cache status header. |
| Varnish is really in front | Combine the caching_application value above with the response headers. On a production-mode store the built-in cache does not send X-Magento-Cache-Debug, so a HIT or MISS value comes from Varnish. Magento's generated VCL removes X-Varnish and Via, so their absence proves nothing; a custom VCL may keep them. | Value 2 and X-Magento-Cache-Debug on cacheable pages. An empty var/page_cache is not proof on its own: with Redis or Valkey as the page cache backend it stays empty without Varnish. |
For what Varnish does and why the full page cache matters, see what Varnish cache is in Magento.
4. Cache and session storage
| Check | How to check | What good looks like |
|---|---|---|
| Default cache and sessions use Redis or Valkey | Read the cache and session sections of app/etc/env.php on the server. Do not paste this file anywhere: it holds credentials. | The default cache backend and session save handler point to Redis or Valkey, not the file system, on any multi-node setup. |
| Separate storage per use | Compare the host, port and database number for default cache, page cache and sessions. | Adobe's best result is a dedicated instance each. If one instance is shared, each use has its own database number (Adobe's example is 0, 1 and 2). |
| Service answers | redis-cli ping, or valkey-cli ping on Valkey. | PONG. Adobe notes that from Adobe Commerce 2.4.9, Valkey replaced Redis in the CLI tooling. |
| Supported version | Compare the server version with the Valkey or Redis row for your exact release in Adobe's system requirements. | A version listed for your release and patch level. |
5. Search engine
| Check | How to check | What good looks like |
|---|---|---|
| Engine configured | bin/magento config:show catalog/search/engine | opensearch, or the engine your release supports. Adobe states that production instances need Elasticsearch or OpenSearch. |
| Version supported | Compare the cluster version with the search engine row for your release in Adobe's system requirements. | A listed version. Recent 2.4.8 and 2.4.9 rows list OpenSearch. |
| Cluster health | curl -s http://127.0.0.1:9200/_cluster/health from the application server (adjust host and port, and pass credentials if the OpenSearch security plugin is enabled). | green. yellow is normal on a single node with replica shards configured; red means some data is unavailable. |
6. Indexers and cron
| Check | How to check | What good looks like |
|---|---|---|
| Indexer mode | bin/magento indexer:show-mode | Update by Schedule, which Adobe recommends for large sites and sites with frequent updates and heavy traffic. Before 2.4.8 the Customer Grid indexer supports Update on Save only. |
| Indexer state and backlog | bin/magento indexer:status | Ready, with a backlog that drains between checks instead of growing. |
| Cron installed | crontab -l as the file system owner. | A block between #~ MAGENTO START and #~ MAGENTO END that runs cron:run every minute. |
| Cron actually runs | Recent rows in the cron_schedule table and entries in var/log/cron.log. | Recent successful jobs, no growing pile of missed or error rows. Adobe lists indexing, emails, sitemaps and price rules among the features that depend on cron. |
Adobe advises enabling maintenance mode and disabling cron before switching indexer modes, to avoid database locks. Plan that change for a quiet window.
7. PHP runtime
| Check | How to check | What good looks like |
|---|---|---|
| OPcache settings | Read the values PHP-FPM loads as well as the CLI ones: for example php-fpm -i | grep -i opcache with your FPM binary name. CLI and FPM often read different ini files, and -i does not show per-pool php_value or php_admin_value overrides, so also read the pool configuration files. | Adobe's recommended values: opcache.memory_consumption=512, opcache.max_accelerated_files=60000, opcache.consistency_checks=0, opcache.validate_timestamps=0, opcache.enable_cli=1. |
| Deploys reset OPcache | Read your deploy script. | With validate_timestamps=0, PHP does not notice changed files on its own. The deploy reloads PHP-FPM after new code is in place. Calling opcache_reset() from the CLI does not clear PHP-FPM's cache, because the CLI has its own. |
| Realpath cache | Same ini output, realpath_cache lines. | realpath_cache_size=10M and realpath_cache_ttl=7200, as Adobe recommends. |
| APCu and memory | Same ini output, apc and memory_limit lines. | APCu loaded with apc.enabled=1; memory_limit=1G per Adobe's recommendation. |
| Only needed extensions | php-fpm -m | No profiling or debugging extension such as Xdebug loaded on production. Adobe notes that an idle Xdebug module still slows responses. |
8. Database
| Check | How to check | What good looks like |
|---|---|---|
| Buffer pool size | SHOW VARIABLES LIKE 'innodb_buffer_pool%'; | Not the 128 MB default on a store of any size. Adobe advises that each buffer pool instance is at least 1 GB. |
| Connections match PHP workers | SHOW VARIABLES LIKE 'max_connections'; compared with your PHP-FPM worker count. | Adobe says max_connections should correlate with the total number of PHP threads; it suggests 300 for a small environment and 1,000 for a medium one. |
| Slow queries are visible | SHOW VARIABLES LIKE 'slow_query_log%'; On staging, or in an agreed window, enable the slow query log in the server configuration and summarise it with mysqldumpslow. | You can name the slowest query patterns and which page or cron job runs them. |
9. Frontend assets
| Check | How to check | What good looks like |
|---|---|---|
| Minification | bin/magento config:show dev/js/minify_files and dev/css/minify_files, at default and store view scope. | Luma store views: 1. Hyvä store views: Hyvä does not switch these off for you; its documentation recommends setting Magento's minify, merge and bundle options (dev/js/*, dev/css/*) to 0 for every Hyvä store view. CSS minification can only be set at global scope, so on an install that mixes Hyvä and Luma views one CSS setting applies to all of them; Hyvä's FAQ covers that case. |
| JavaScript merging | bin/magento config:show dev/js/merge_files | 0. Adobe does not recommend Merge JavaScript Files: it can break bundling and RequireJS logic and gives no benefit with HTTP/2. |
| JavaScript bundling | bin/magento config:show dev/js/enable_js_bundling, then check the protocol with curl -sI --http2 https://www.example.com/ | A decision backed by a measurement. Adobe advises against merging or bundling on HTTP/2 and notes that built-in bundling makes the browser load all bundles rather than just the ones a page needs. |
| LCP image | The LCP element in the Lighthouse report for each template. | The LCP image is not lazy-loaded, is sized for its slot, and has width and height attributes. Images below the first screen use loading="lazy". |
| Third-party scripts | Lighthouse with --blocked-url-patterns for one tag at a time, compared with a normal run. | You know what each tag costs in Total Blocking Time, and the owner of each tag has agreed it is needed. |
10. Static files, CDN and HTTP
| Check | How to check | What good looks like |
|---|---|---|
| Long cache on static files | curl -sI on one file under /static/ and one under /media/. | A long Cache-Control lifetime. Magento's nginx.conf.sample sets public, max-age=31536000, immutable on static file types such as JS, CSS, images and fonts under /static/, and a one-year expiry on the same media file types. |
| CDN in use, if you have one | Stores > Configuration > General > Web > Base URLs: Base URL for Static View Files and Base URL for User Media Files, then the asset URLs in page source. | Static and media URLs point to the CDN host, and the CDN reports cache hits for them. |
| Compression | A GET request that prints headers only: curl -s -o /dev/null -D - -H 'Accept-Encoding: gzip, br' https://www.example.com/ | A Content-Encoding header on HTML, CSS and JavaScript responses. The sample nginx config turns gzip on. |
Reading the results
Most failed rows point in one of three directions. If cache-miss server time is high, look at sections 6 to 8, and at extensions doing work on every request. If pages are cache hits but LCP or INP still fail in field data, the problem is in the browser: images, third-party tags and theme JavaScript, which the Core Web Vitals guide and the frontend comparison cover. If pages are rarely cache hits, go back to section 3 before anything else.
After any change, repeat the baseline in section 1 with the same URLs and compare. Field data needs a full 28-day window after the change before it reflects it.
Frequently asked questions
Do these checks apply to Magento 2.4.9?
The checks apply across Magento 2.4.x, but supported versions of PHP, Valkey or Redis, Varnish and OpenSearch change by release and patch. Compare your servers with Adobe's system requirements table for your exact version rather than relying on any version list on this page.
Are the commands safe to run on production?
The show, status and SHOW VARIABLES commands only read state. Changing indexer modes, PHP settings, database settings or cache backends should happen on staging first. On a shared Redis or Valkey server, use cache:clean rather than a storage flush.
Will passing every row make my store pass Core Web Vitals?
Not on its own. The backend rows make pages fast to serve, but LCP, INP and CLS also depend on images, layout and the JavaScript the theme and third-party tags run. Check field data again after each change.
I use Hyvä. Do I still need this checklist?
Yes. Hyvä changes the theme layer, so the minification and bundling rows work differently, but caching, Valkey or Redis, search, indexers, cron, PHP and the database are the same Magento backend.
Does this work on Adobe Commerce on cloud infrastructure?
Many rows still apply, but some services are managed for you and the platform always runs in production mode. Adobe's documentation for cloud projects sets frontend options through the command line and config.php rather than the Admin, so follow the cloud-specific pages for those changes.
Want this checklist run on your store? Written audit + prioritised fix list, with a fixed written quote after a short brief.
Book the audit