fix(memory): cap SQLite pools and shrink the event cache periodically #159

Merged

jlxq0 merged 3 commits from fix/memory-growth into main 2026-11-09 02:46:27 +00:00

Conversation 0 Commits 3 Files changed 3 +190 -6

jlxq0 commented 2026-11-09 02:15:35 +00:00

Owner

• Copy link
• Copy Markdown

v0.10.17 already frees replaced clients (#158), but the pod (limit 1Gi) still climbed from about 10Mi to 1Gi in steps within 1–2 days, and was OOMKilled 5 times in 6 days. One Matrix client is active, syncing about 500–900 responses an hour.

Two remaining contributors, both in how matrix-sdk 0.19 is configured:

1. SQLite pools. SqliteStoreConfig::new gives each store 4 x physical CPUs connections with a 2 MB page cache, and a client has four stores. Connections keep their cache once opened, which fits the step-shaped growth. This switches to the SDK's with_low_memory_config preset (one connection per physical CPU, 500 KB cache), with the same passphrase. A test proves the store still refuses a wrong passphrase.
2. Event cache. Each room's in-memory linked chunk is only unloaded when the room's last subscriber drops (Subscriber::drop → auto_shrink_if_no_subscribers), and nothing here subscribes to rooms (#158 notes the same). A 15-minute task per client now takes and drops a subscriber for each joined room. It holds only a Weak<Client> and is aborted in CachedClient::drop next to the sync task, so #158's freed-client test still holds.

Plus fix(deps): vodozemac 0.11.2 and ctutils 0.4.8 for RUSTSEC-2026-0445/0451/0385, which were published after v0.10.17 and fail cargo audit on main.

On toolchain 1.98.0 (CI's): fmt, clippy -D warnings and cargo test --all-features all pass (389 + 6 tests), and cargo audit is clean apart from the 2 allowed warnings.

Note: v0.10.18 was tagged by mistake on eccda5d, the same commit as v0.10.17. This change ships as v0.10.19.

🤖 Generated with Claude Code

https://claude.ai/code/session_01E2xQRZnoBbY9SEC5nF6J4U

`v0.10.17` already frees replaced clients (#158), but the pod (limit 1Gi) still climbed from about 10Mi to 1Gi in steps within 1–2 days, and was OOMKilled 5 times in 6 days. One Matrix client is active, syncing about 500–900 responses an hour. Two remaining contributors, both in how matrix-sdk 0.19 is configured: 1. **SQLite pools.** `SqliteStoreConfig::new` gives each store `4 x physical CPUs` connections with a 2 MB page cache, and a client has four stores. Connections keep their cache once opened, which fits the step-shaped growth. This switches to the SDK's `with_low_memory_config` preset (one connection per physical CPU, 500 KB cache), with the same passphrase. A test proves the store still refuses a wrong passphrase. 2. **Event cache.** Each room's in-memory linked chunk is only unloaded when the room's last subscriber drops (`Subscriber::drop` → `auto_shrink_if_no_subscribers`), and nothing here subscribes to rooms (#158 notes the same). A 15-minute task per client now takes and drops a subscriber for each joined room. It holds only a `Weak<Client>` and is aborted in `CachedClient::drop` next to the sync task, so #158's freed-client test still holds. Plus `fix(deps)`: vodozemac 0.11.2 and ctutils 0.4.8 for RUSTSEC-2026-0445/0451/0385, which were published after v0.10.17 and fail `cargo audit` on main. On toolchain 1.98.0 (CI's): fmt, clippy `-D warnings` and `cargo test --all-features` all pass (389 + 6 tests), and `cargo audit` is clean apart from the 2 allowed warnings. Note: `v0.10.18` was tagged by mistake on `eccda5d`, the same commit as `v0.10.17`. This change ships as `v0.10.19`. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01E2xQRZnoBbY9SEC5nF6J4U

jlxq0 added 1 commit 2026-11-09 02:15:35 +00:00

fix(memory): cap SQLite pools and shrink the event cache periodically ... 2819ca4dc5

The pod climbed from about 10 MiB to its 1 GiB limit in steps and was
OOM-killed five times in six days. Two contributors, both matrix-sdk
configuration:

- SqliteStoreConfig defaults to 4 x physical CPUs connections per store
  with a 2 MB page cache each, across four stores, each connection keeping
  its cache once opened. Use the SDK's low-memory preset (one connection
  per physical CPU, 500 KB cache), same passphrase.
- The event cache keeps every synced event of a room in memory until its
  last subscriber drops, and nothing here subscribes to rooms. Every 15
  minutes take and drop a subscriber per joined room, the SDK's own
  trigger to unload all but the last chunk. Events stay in SQLite.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01E2xQRZnoBbY9SEC5nF6J4U

jlxq0 added 1 commit 2026-11-09 02:25:42 +00:00

fix(deps): clear 17 RUSTSEC advisories with lockfile updates ... 4156f4ea44

rustls 0.23.46, rustls-webpki 0.103.16, vodozemac 0.11.2, quinn-proto
0.11.19, webpki-roots and webpki-root-certs 1.0.11, pulldown-cmark 0.13.6,
ctutils 0.4.8. cargo audit now reports only the three allowed warnings.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01E2xQRZnoBbY9SEC5nF6J4U

jlxq0 force-pushed fix/memory-growth from 4156f4ea44 to 0c4fc04f95 2026-11-09 02:38:27 +00:00 Compare

jlxq0 added 1 commit 2026-11-09 02:38:50 +00:00

release: v0.10.19 ... cdc1a07b11

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01E2xQRZnoBbY9SEC5nF6J4U

jlxq0 merged commit 8d38e9b4ca into main 2026-11-09 02:46:27 +00:00

jlxq0 deleted branch fix/memory-growth 2026-11-09 02:46:28 +00:00

jlxq0 referenced this pull request from a commit 2026-11-09 02:46:29 +00:00

Merge pull request 'fix(memory): cap SQLite pools and shrink the event cache periodically' (#159) from fix/memory-growth into main

jlxq0 referenced this pull request from a commit 2026-11-09 03:35:59 +00:00

fix(matrix-mcp): v0.10.20, caps SQLite pools and shrinks the event cache

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

blocked

Held on a named, checkable blocker: an issue number, a credential or a measurement. Not "held".

  

bug

Something isn't working

  

documentation

Improvements or additions to documentation

  

duplicate

This issue or pull request already exists

  

enhancement

New feature or request

  

good first issue

Good for newcomers

  

help wanted

Extra attention is needed

  

invalid

This doesn't seem right

  

question

Further information is requested

  

waiting-on-julian

Blocked on Julian. The ask is in the issue.

  

wontfix

This will not be worked on

No labels

blocked

bug

documentation

duplicate

enhancement

good first issue

help wanted

invalid

question

waiting-on-julian

wontfix

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

jlxq0/matrix-mcp!159

No description provided.