From beb7a951f4b5dffda74428d039be42fe3e37334b Mon Sep 17 00:00:00 2001 From: Henrique Date: Mon, 7 Oct 2024 05:37:39 -0300 Subject: [PATCH] docs: add note about PYTHONMALLOC for accurate jemalloc memory tracking (#17709) Added a note in the documentation suggesting that users may set `PYTHONMALLOC=malloc` when using `jemalloc`. This allows jemalloc to track memory usage more accurately by bypassing Python's internal small-object allocator (`pymalloc`), helping to ensure that `cache_autotuning` functions as expected. This doc change aims to provide more clarity for users configuring jemalloc with Synapse. Based on: https://github.com/element-hq/synapse/blob/4ac783549c5bac7a490a715d359f330bb0b1a161/synapse/metrics/jemalloc.py#L198-L201 --- changelog.d/17709.doc | 1 + docs/usage/administration/admin_faq.md | 2 ++ 2 files changed, 3 insertions(+) create mode 100644 changelog.d/17709.doc diff --git a/changelog.d/17709.doc b/changelog.d/17709.doc new file mode 100644 index 0000000000..8fffc1ca0c --- /dev/null +++ b/changelog.d/17709.doc @@ -0,0 +1 @@ +Add documentation note about PYTHONMALLOC for accurate jemalloc memory tracking. Contributed by @hensg. diff --git a/docs/usage/administration/admin_faq.md b/docs/usage/administration/admin_faq.md index a1184d0375..0dce3d3e37 100644 --- a/docs/usage/administration/admin_faq.md +++ b/docs/usage/administration/admin_faq.md @@ -255,6 +255,8 @@ line to `/etc/default/matrix-synapse`: LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so.2 +*Note*: You may need to set `PYTHONMALLOC=malloc` to ensure that `jemalloc` can accurately calculate memory usage. By default, Python uses its internal small-object allocator, which may interfere with jemalloc's ability to track memory consumption correctly. This could prevent the [cache_autotuning](../configuration/config_documentation.md#caches-and-associated-values) feature from functioning as expected, as the Python allocator may not reach the memory threshold set by `max_cache_memory_usage`, thus not triggering the cache eviction process. + This made a significant difference on Python 2.7 - it's unclear how much of an improvement it provides on Python 3.x.