##// END OF EJS Templates
release: Finish preparation for 4.10.0
release: Finish preparation for 4.10.0

File last commit:

r2205:dd780472 default
r2218:4ba4d749 v4.10.0 stable
Show More
tuning-scale-horizontally.rst
58 lines | 2.2 KiB | text/x-rst | RstLexer
/ docs / admin / tuning-scale-horizontally.rst

Scale Horizontally

|RCE| is built in a way it support horizontal scaling across multiple machines. There are two main pre-requisites for that:

  • Shared storage that each machine can access.
  • Shared DB connection across machines.

Horizontal scaling means adding more machines or workers into your pool of resources. Horizontally scaling |RCE| gives a huge performance increase, especially under large traffic scenarios with a high number of requests. This is very beneficial when |RCE| is serving many users simultaneously, or if continuous integration servers are automatically pulling and pushing code.

If you scale across different machines, each |RCM| instance needs to store its data on a shared disk, preferably together with your |repos|. This data directory contains template caches, a full text search index, and is used for task locking to ensure safety across multiple instances. To do this, set the following properties in the :file:`rhodecode.ini` file to set the shared location across all |RCM| instances.

cache_dir = /shared/path/caches                 # set to shared location
search.location = /shared/path/search_index     # set to shared location

####################################
###         BEAKER CACHE        ####
####################################
beaker.cache.data_dir = /shared/path/data       # set to shared location
beaker.cache.lock_dir = /shared/path/lock       # set to shared location

Note

If you use custom caches such as beaker.cache.auth_plugins. it's recommended to set it to the memcached/redis or database backend so it can be shared across machines.

It is recommended to create another dedicated |RCE| instance to handle traffic from build farms or continuous integration servers.

Note

You should configure your load balancing accordingly. We recommend writing load balancing rules that will separate regular user traffic from automated process traffic like continuous servers or build bots.

Note

If Celery is used on each instance then you should run separate Celery instances, but the message broker should be the same for all of them.