eZ Platform can be set up to run efficiently on almost any modern configuration. What follows is a list of recommendation that will make your installation perform better.
All the following recommendations are valid for both development and production setups, unless otherwise noted.
If you are in a hurry, the most important recommendations on this page are:
- Use PHP 7.x, and dump optimized Composer autoload classmap
- Use a full web (Nginx/Apache) server with vhost
- Avoid shared filesystems for code (Docker for Mac/Win, VirtualBox/*, Vagrant, etc.), or find ways to optimize or work around the issues.
- For clustering (mainly relevant for production/staging), reduce latency to Redis/Memcached, use Varnish and Solr.
- Always use an up-to-date browser and an up-to-date operating system so you have access to latest browser versions
- If possible, use a fast, stable internet connection, because an unreliable connection will slow down UI
In production setups:
- Always use reverse proxy, and if possible use Varnish.
- Compared to the built-in Symfony Proxy in PHP Varnish is much faster and is able to queue up requests for the same fresh/invalidated resource.
- With ezplatform-http-cache support for xkey and grace Varnish provides more stable performance in read/write scenarios.
- Set up eZ Platform in cluster mode if you need to handle bigger spikes of traffic than a single server can manage.
The following recommendations are ordered from largest to smallest impact they have on performance in general.
- Avoid shared filesystems for code (Docker for Mac/Win, VirtualBox/*, Vagrant, etc.), because they typically slow down the application 10x or more, compared to native Linux filesystem.
- VM in itself also adds 10-30% of overhead. However when it comes to production, e.g. AWS vs barebones, it also comes down to cost and convenience factors.
For Development use, try eZ Launchpad
For a ready solution that allows you to share code between your host and the underlying running VM system without this performance hit, try eZ Launchpad.
- Use Nginx/Apache even for development, as PHP's built-in web server (as exposed via Symfony's
server:*commands) is only able to handle one request at a time (including JS/CSS/* asset loading, etc.).
- Use a recent version of nginx, set up https, and enable http/2 to reduce connection latency on parallel requests.
- Use PHP 7.0 (or, better yet, PHP 7.1 if available).
- Always enable opcache for php-fpm/
- Prefer php-fpm and web server using it over fast-cgi for lower overall memory usage.
- Keep Composer up to date.
- Always dump optimized class map using
composer dump-autoload --optimizeor relevant flags on
- Memcached/Redis is recommended over filesystem cache even with a single server, as it offers better general performance for operations invalidating cache.
- However, pure read performance is slower, especially if the next point is not optimized.
- If you use Redis, make sure to tune it for in-memory cache usage. Its persistence feature is not needed with eZ Platform cache and will severely slow down execution time.
- Use Solr Bundle and Solr to greatly offload your database and get more stable performance on your installation.
Executing long-running console commands¶
Executing long-running console commands can result in running out of memory. Two examples of such commands are a custom import command and the indexing command provided by the Solr Bundle.
Reducing memory usage¶
To avoid quickly running out of memory while executing such commands you should make sure to:
Always run in prod environment using:
For logging using monolog, if you use either the default
bufferhandler, make sure to specify
buffer_sizeto limit how large the buffer grows before it gets flushed:
1 2 3 4 5 6
# config_prod.yml (partial example) monolog: handlers: main: type: fingers_crossed buffer_size: 200
Run PHP without memory limits using:
php -d memory_limit=-1 bin/console <command>
xdebug(PHP extension to debug/profile php use) when running the command, this will cause php to use much more memory.
Memory will still grow
Even when everything is configured like described above, memory will grow for each iteration of indexing/inserting a Content item with at least 1kb per iteration after the initial first 100 rounds. This is expected behavior; to be able to handle more iterations you will have to do one or several of the following:
- Change the import/index script in question to use process forking to avoid the issue.
- Upgrade PHP: newer versions of PHP are typically more memory-efficient.
- Run the console command on a machine with more memory (RAM).
Process forking with Symfony¶
The recommended way to completely avoid "memory leaks" in PHP in the first place is to use processes. For console scripts this is typically done using process forking which is quite easily achievable with Symfony.
The things you will need to do:
Change your command so it supports taking slice parameters, like for instance a batch size and a child-offset parameter.
- If defined, child-offset parameter denotes if a process is a child, this can be accomplished using two commands as well.
- If not defined, it is the master process which will execute the processes until nothing is left to process.
Change the command so that the master process takes care of forking child processes in slices.