Performance Tweaks
Shopware is a platform for many different projects. It needs to handle a broad range of load characteristics and environments. That means that the default configuration is optimized for the best out-of-the-box experience. But there are many opportunities to increase the performance by fitting the configuration to your needs.
HTTP cache
To ensure a high RPS (Requests Per Second), Shopware offers an integrated HTTP cache with a possible reverse proxy configuration. Any system that handles high user numbers should always use HTTP caching to reduce server resources.
To enable this, set SHOPWARE_HTTP_CACHE_ENABLED=1
in the .env
Reverse proxy cache
When you have many app servers, you should consider using a reverse proxy cache like Varnish. Shopware offers a default configuration for Varnish out-of-the-box.
Logged-in / cart-filled
By default, Shopware can no longer deliver complete pages from a cache for a logged-in customer or if products are in the shopping cart. As soon as this happens, the user sessions differ, and the context rules could be different depending on the user. This results in different content for each customer. A good example is the Dynamic Access plugin.
However, if the project does not require such functionality, pages can also be cached by the HTTP cache/reverse proxy. To disable cache invalidation in these cases:
shopware:
cache:
invalidation:
http_cache: []
Delayed invalidation
A delay for the cache invalidation can be activated for systems with a high update frequency for the inventory (products, categories). Once the instruction to delete the cache entries for a specific product or category occurs, they are not deleted instantly but processed by a background task afterwards. Thus, if two processes invalidate the cache in quick succession, the timer for the invalidation of this cache entry will only reset.
shopware:
cache:
invalidation:
delay: 0
count: 150
MySQL instead of MariaDB
INFO
If you use Elasticsearch/Opensearch as a search engine, you can ignore this section. All filtering, sorting, and aggregations are done in Elasticsearch/Opensearch.
In some places in the code, we use JSON fields. As soon as it comes to filtering, sorting, or aggregating JSON fields, MySQL is ahead of the MariaDB fork. Therefore, we strongly recommend the use of MySQL.
MySQL configuration
Shopware sets some MySQL configuration variables on each request to ensure it works in any environment. You can disable this behavior if you have correctly configured your MySQL server.
- Make sure that
group_concat_max_len
is by default higher or equal to320000
- Make sure that
sql_mode
doesn't containONLY_FULL_GROUP_BY
and then you can setSQL_SET_DEFAULT_SESSION_VARIABLES=0
to your.env
file
SQL is faster than DAL
We designed the DAL (Data Abstraction Layer) to provide developers a flexible and extensible data management. However, features in such a system come at the cost of performance. Therefore, using DBAL (plain SQL) is much faster than using the DAL in many scenarios, especially when it comes to internal processes, where often only one ID of an entity is needed.
Refer to this article to know more on when to use plain SQL and DAL.
Elasticsearch/Opensearch
Elasticsearch/Opensearch is a great tool to reduce the load of the MySQL server. Especially for systems with large product assortments, this is a must-have since MySQL simply does not cope well above a certain assortment size.
When using Elasticsearch, it is important to set the SHOPWARE_ES_THROW_EXCEPTION=1
.env
variable. This ensures that there is no fallback to the MySQL server if an error occurs when querying the data via Elasticsearch. In large projects, the failure of Elasticsearch leads to the MySQL server being completely overloaded otherwise.
Read more on Elasticsearch setup
Prevent mail data updates
INFO
Prevent mail updates feature is available starting with Shopware 6.4.11.0.
To provide auto-completion for different mail templates in the Administration UI, Shopware has a mechanism that writes an example mail into the database when sending the mail.
With the shopware.mail.update_mail_variables_on_send
configuration, you can disable this source of database load:
shopware:
mail:
update_mail_variables_on_send: false
Increment storage
The Increment storage is used to store the state and display it in the Administration. This storage increments or decrements a given key in a transaction-safe way, which causes locks upon the storage. Therefore, we recommend moving this source of server load to a separate Redis:
shopware:
increment:
user_activity:
type: 'redis'
config:
url: 'redis://host:port/dbindex'
message_queue:
type: 'redis'
config:
url: 'redis://host:port/dbindex'
If you don't need such functionality, it is highly recommended to disable this behavior by using array
as a type.
Lock storage
Shopware uses Symfony's Lock component to implement locking functionality. By default, Symfony will use a local file-based lock store, which breaks into multi-machine (cluster) setups. This is avoided using one of the supported remote stores.
framework:
lock: 'redis://host:port'
Number ranges
Number Ranges provide a consistent way to generate a consecutive number sequence that is used for order numbers, invoice numbers, etc. The generation of the number ranges is an atomic operation, which guarantees that the sequence is consecutive and no number is generated twice.
By default, the number range states are stored in the database. In scenarios where high throughput is required (e.g., thousands of orders per minute), the database can become a performance bottleneck because of the requirement for atomicity. Redis offers better support for atomic increments than the database. Therefore the number ranges should be stored in Redis in such scenarios.
shopware:
number_range:
increment_storage: "Redis"
redis_url: 'redis://host:port/dbindex'
Sending mails with the Queue
Shopware sends the mails by default synchronously. This process can take a while when the remote SMTP server is struggling. For this purpose, it is possible to handle the mails in the message queue. To enable this, add the following config to your config:
framework:
mailer:
message_bus: 'messenger.default_bus'
PHP Config tweaks
# don't evaluate assert()
assert.active=0
# cache file_exists,is_file
opcache.enable_file_override=1
# increase opcache string buffer as shopware has many files
opcache.interned_strings_buffer=20
# disable check for BOM
zend.detect_unicode=0
# increase default realpath cache
realpath_cache_ttl=3600
INFO
The web updater is not compatible with opcache, as updates require an opcache clear.
Also, PHP PCRE Jit Target should be enabled. This can be checked using php -i | grep 'PCRE JIT Target'
or looking into the phpinfo page.
For an additional 2-5% performance improvement, it is possible to provide a preload file to opcache. Preload also brings a lot of drawbacks:
- Each cache clear requires a PHP-FPM restart
- Each file change requires a PHP-FPM restart
- Extension Manager does not work
The PHP configuration would look like:
opcache.preload=/var/www/html/var/cache/opcache-preload.php
opcache.preload_user=nginx
Cache ID
The Shopware cache has a global cache id to clear the cache faster and work in a cluster setup. This cache id is saved in the database and will only be changed when the cache is cleared. This ensures that the new cache is used and the message queue can clean the old folder. If this functionality is not used, this cache id can also be hardcoded SHOPWARE_CACHE_ID=foo
in the .env
to save one SQL query on each request.
.env.local.php
Symfony recommends that a .env.local.php
file is used in Production instead of a .env
file to skip parsing of the .env file on every request. If you are using a containerized environment, all those variables can also be set directly in the environment variables instead of dumping them into a file.
Since Shopware 6.4.15.0, you can dump the content of the .env
file to a .env.local.php
file by running bin/console system:setup --dump-env
or bin/console dotenv:dump {APP_ENV}
.
Benchmarks
In addition to the benchmarks that Shopware regularly performs with the software, we strongly recommend integrating your benchmark tools and pipelines for larger systems. A generic benchmark of a product can rarely be adapted to individual, highly customized projects. Tools such as locust or k6 can be used for this purpose.
Logging
Set the log level of the monolog to error
to reduce the amount of logged events. Also, limiting the buffer_size
of monolog prevents memory overflows for long-lived jobs:
monolog:
handlers:
main:
level: error
buffer_size: 30
business_event_handler_buffer:
level: error
The business_event_handler_buffer
handler logs flow. Setting it to error
will disable the logging of flow activities that succeed.