Announce larger Synapse caches and cache auto-tuning

CHANGELOG.md

@@ -1,3 +1,22 @@
+# 2024-02-14
+
+## Much larger Synapse caches and cache auto-tuning enabled by default
+
+Thanks to [FSG-Cat](https://github.com/FSG-Cat) (inspired by [Tom Foster](https://github.com/tcpipuk)'s [Synapse optimization guide](https://tcpipuk.github.io/synapse/deployment/synapse.html)), 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).
+
+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.
+
+
 # 2024-01-17
 
 ## Switching to Element's AGPLv3-licensed Synapse release