Increase Synapse caches and enable cache-autotuning by default (#3017)

* Modify Synapse Cache Factor to use Auto Tune

Synapse has the ability to as it calls in its config auto tune caches.

This ability lets us set very high cache factors and then instead limit our resource use.

Defaults for this commit are 1/10th of what Element apparently runs for EMS stuff and matrix.org on Cache Factor and upstream documentation defaults for auto tune.

* Add vars to Synapse main.yml to control cache related config

This commit adds various cache related vars to main.yml for Synapse.

Some are auto tune and some are just adding explicit ways to control upstream vars.

* Updated Auto Tune figures

Autotuned figures have been bumped in consultation with other community members as to a reasonable level. Please note these defaults are more on the one of each workers side than they are on the monolith Side.

* Fix YML Error

The playbook is not happy with the previous state of this patch so this commit hopefully fixes it

* Add to_json to various Synapse tuning related configs

* Fix incorrect indication in homeserver.yaml.j2

* Minor cleanups

* Synapse Cache Autotuning Documentation

* Upgrade Synapse Cache Autotune to auto configure memory use

* Update Synapse Tuning docs to reflect automatic memory use configuration

* Fix Linting errors in synapses main.yml

* Rename variables for consistency (matrix_synapse_caches_autotuning_* -> matrix_synapse_cache_autotuning_*)

* Remove FIX ME comment about Synapse's `cache_autotuning`

`docs/maintenance-synapse.md` and `roles/custom/matrix-synapse/defaults/main.yml`
already contains documentation about these variables and the default values we set.

* Improve "Tuning caches and cache autotuning" documentation for Synapse

* Announce larger Synapse caches and cache auto-tuning

---------

Co-authored-by: Slavi Pantaleev <slavi@devture.com>

CHANGELOG.md

@@ -1,3 +1,23 @@
+# 2024-02-14
+
+## Much larger Synapse caches and cache auto-tuning enabled by default
+
+Thanks to [FSG-Cat](https://github.com/FSG-Cat), the playbook now uses much larger caches and enables Synapse's [cache auto-tuning functionality](https://matrix-org.github.io/synapse/latest/usage/configuration/config_documentation.html#caches-and-associated-values).
+This work and the default values used by the playbook are inspired by [Tom Foster](https://github.com/tcpipuk)'s [Synapse homeserver guide](https://tcpipuk.github.io/synapse/deployment/synapse.html).
+
+The playbook has always used a very conservative cache factor (`matrix_synapse_caches_global_factor`) value of `0.5`, which may be OK for small and underactive deployments, but is not ideal for larger servers. Paradoxically, a small global cache factor value [does not necessarily decrease RAM usage as a whole](https://github.com/matrix-org/synapse/issues/3939).
+
+The playbook now uses **a 20x larger cache factor** (currently `10`), adjusts a few other cache-related variables, and **enables cache auto-tuning** via the following variables:
+
+- `matrix_synapse_cache_autotuning_max_cache_memory_usage` - defaults to 1/8 of total RAM with a cap of 2GB; values are specified in bytes
+- `matrix_synapse_cache_autotuning_target_cache_memory_usage` - defaults to 1/16 of total RAM with a cap of 1GB; values are specified in bytes
+- `matrix_synapse_cache_autotuning_min_cache_ttl` - defaults to `30s`
+
+These values should be good defaults for most servers, but may change over time as we experiment further.
+
+Refer to our new [Tuning caches and cache autotuning](docs/maintenance-synapse.md#tuning-caches-and-cache-autotuning) documentation section for more details.
+
+
 # 2024-01-31
 
 ## (Backward-compatibility break) Minor changes necessary for some people serving a static website at the base domain

docs/maintenance-synapse.md

@@ -74,8 +74,32 @@ Synapse's presence feature which tracks which users are online and which are off
 
 If you have enough compute resources (CPU & RAM), you can make Synapse better use of them by [enabling load-balancing with workers](configuring-playbook-synapse.md#load-balancing-with-workers).
 
-Tuning Synapse's cache factor can help reduce RAM usage. [See the upstream documentation](https://github.com/element-hq/synapse#help-synapse-is-slow-and-eats-all-my-ram-cpu) for more information on what value to set the cache factor to. Use the variable `matrix_synapse_caches_global_factor` to set the cache factor.
-
 [Tuning your PostgreSQL database](maintenance-postgres.md#tuning-postgresql) could also improve Synapse performance. The playbook tunes the integrated Postgres database automatically, but based on your needs you may wish to adjust tuning variables manually. If you're using an [external Postgres database](configuring-playbook-external-postgres.md), you will aslo need to tune Postgres manually.
 
+### Tuning caches and cache autotuning
+
+Tuning Synapse's cache factor is useful for performance increases but also as part of controlling Synapse's memory use. Use the variable `matrix_synapse_caches_global_factor` to set the cache factor as part of this process.
+
+**The playbook defaults the global cache factor to a large value** (e.g. `10`). A smaller value (e.g. `0.5`) will decrease the amount used for caches, but will [not necessarily decrease RAM usage as a whole](https://github.com/matrix-org/synapse/issues/3939).
+
+Tuning the cache factor is useful only to a limited degree (as its crude to do in isolation) and therefore users who are tuning their cache factor should likely look into tuning autotune variables as well (see below).
+
+Cache autotuning is **enabled by default** and controlled via the following variables:
+
+- `matrix_synapse_cache_autotuning_max_cache_memory_usage` - defaults to 1/8 of total RAM with a cap of 2GB; values are specified in bytes
+- `matrix_synapse_cache_autotuning_target_cache_memory_usage` - defaults to 1/16 of total RAM with a cap of 1GB; values are specified in bytes
+- `matrix_synapse_cache_autotuning_min_cache_ttl` - defaults to `30s`
+
+You can **learn more about cache-autotuning and the global cache factor settings** in the [Synapse's documentation on caches and associated values](https://matrix-org.github.io/synapse/latest/usage/configuration/config_documentation.html#caches-and-associated-values).
+
+To **disable cache auto-tuning**, unset all values:
+
+```yml
+matrix_synapse_cache_autotuning_max_cache_memory_usage: ''
+matrix_synapse_cache_autotuning_target_cache_memory_usage: ''
+matrix_synapse_cache_autotuning_min_cache_ttl: ''
+```
+
+Users who wish to lower Synapse's RAM footprint should look into lowering the global cache factor and tweaking the autotune variables (or disabling auto-tuning). If your cache factor is too low for a given auto tune setting your caches will not reach autotune thresholds and autotune won't be able to do its job. Therefore, when auto-tuning is enabled (which it is by default), it's recommended to have your cache factor be large.
+
 See also [How do I optimize this setup for a low-power server?](faq.md#how-do-i-optimize-this-setup-for-a-low-power-server).

roles/custom/matrix-synapse/defaults/main.yml

@@ -547,8 +547,23 @@ matrix_synapse_event_cache_size: "100K"
 
 # Controls cache sizes for Synapse.
 # Raise this to increase cache sizes or lower it to potentially lower memory use.
-# To learn more, see: https://github.com/matrix-org/synapse/issues/3939
-matrix_synapse_caches_global_factor: 0.5
+# To learn more, see:
+# - https://matrix-org.github.io/synapse/latest/usage/configuration/config_documentation.html#caching
+# - https://github.com/matrix-org/synapse/issues/3939
+# Defaults for timings of caches is from https://tcpipuk.github.io/synapse/deployment/synapse.html
+# The idea with the timings used is that you get to evict soon but also you keep stuff around for a long time when its not forced out.
+# Long cache lifetimes together with the low minimum TTL allows autotune to be the primary eviction method assuming size of cache is hit before we hit other caps.
+matrix_synapse_caches_global_factor: 10
+matrix_synapse_caches_expire_caches: true
+matrix_synapse_caches_cache_entry_ttl: "1080m"
+matrix_synapse_caches_sync_response_cache_duration: "2m"
+matrix_synapse_cache_autotuning_min_cache_ttl: "30s"
+# The Cache tune math used here is a derivative of the same math used to autotune sizes for postgres.
+# The memtotal variable can in theory be overiden to make Synapse think it has less ram to work with.
+# But if your at the point of considering that just override the math or put static values in.
+matrix_synapse_memtotal_kb: "{{ ansible_memtotal_mb*1024|int }}"
+matrix_synapse_cache_autotuning_max_cache_memory_usage: "{{ 2097152 if (matrix_synapse_memtotal_kb|int/8)/1024 >= 2048 else matrix_synapse_memtotal_kb|int/8 }}"
+matrix_synapse_cache_autotuning_target_cache_memory_usage: "{{ 1048576 if (matrix_synapse_memtotal_kb|int/16)/1024 >= 1024 else matrix_synapse_memtotal_kb|int/16 }}"
 
 # Controls whether Synapse will federate at all.
 # Disable this to completely isolate your server from the rest of the Matrix network.

roles/custom/matrix-synapse/tasks/validate_config.yml

@@ -89,6 +89,9 @@
     - {'old': 'matrix_synapse_send_federation', 'new': '<unnecessary - Synapse relies on federation_sender_instances now>'}
     - {'old': 'matrix_synapse_start_pushers', 'new': '<unnecessary - Synapse relies on pusher_instances now>'}
     - {'old': 'matrix_synapse_spam_checker', 'new': '<superseded by matrix_synapse_modules>'}
+    - {'old': 'matrix_synapse_caches_autotuning_max_cache_memory_usage', 'new': 'matrix_synapse_cache_autotuning_max_cache_memory_usage'}
+    - {'old': 'matrix_synapse_caches_autotuning_target_cache_memory_usage', 'new': 'matrix_synapse_cache_autotuning_target_cache_memory_usage'}
+    - {'old': 'matrix_synapse_caches_autotuning_min_cache_ttl', 'new': 'matrix_synapse_cache_autotuning_min_cache_ttl'}
 
 - name: (Deprecation) Catch and report renamed settings in matrix_synapse_configuration_extension_yaml
   ansible.builtin.fail:

roles/custom/matrix-synapse/templates/synapse/homeserver.yaml.j2

@@ -760,7 +760,7 @@ federation_domain_whitelist: {{ matrix_synapse_federation_domain_whitelist|to_js
 # The number of events to cache in memory. Not affected by
 # caches.global_factor.
 #
-event_cache_size: "{{ matrix_synapse_event_cache_size }}"
+event_cache_size: {{ matrix_synapse_event_cache_size | to_json }}
 
 caches:
   # Controls the global cache factor, which is the default cache factor
@@ -773,8 +773,7 @@ caches:
   #
   # Defaults to 0.5, which will half the size of all caches.
   #
-   global_factor: {{ matrix_synapse_caches_global_factor }}
-
+  global_factor: {{ matrix_synapse_caches_global_factor | to_json }}
   # A dictionary of cache name to cache factor for that individual
   # cache. Overrides the global cache factor for a given cache.
   #
@@ -796,13 +795,13 @@ caches:
   # Controls whether cache entries are evicted after a specified time
   # period. Defaults to true. Uncomment to disable this feature.
   #
-  #expire_caches: false
+  expire_caches: {{ matrix_synapse_caches_expire_caches | to_json }}
 
   # If expire_caches is enabled, this flag controls how long an entry can
   # be in a cache without having been accessed before being evicted.
   # Defaults to 30m. Uncomment to set a different time to live for cache entries.
   #
-  #cache_entry_ttl: 30m
+  cache_entry_ttl: {{ matrix_synapse_caches_cache_entry_ttl | to_json }}
 
   # Controls how long the results of a /sync request are cached for after
   # a successful response is returned. A higher duration can help clients with
@@ -811,7 +810,12 @@ caches:
   # By default, this is zero, which means that sync responses are not cached
   # at all.
   #
-  #sync_response_cache_duration: 2m
+  sync_response_cache_duration: {{ matrix_synapse_caches_sync_response_cache_duration | to_json }}
+
+  cache_autotuning:
+    max_cache_memory_usage: {{ matrix_synapse_cache_autotuning_max_cache_memory_usage | to_json }}
+    target_cache_memory_usage: {{ matrix_synapse_cache_autotuning_target_cache_memory_usage | to_json }}
+    min_cache_ttl: {{ matrix_synapse_cache_autotuning_min_cache_ttl | to_json }}
 
 
 ## Database ##