25012343577
84%
master: 93%

Ran 27 Apr 2026 06:28PM UTC

Jobs 1

Files 23

Run time 1min

Badge

Embed ▾

Committed 27 Apr 2026 06:26PM UTC coverage: 83.886%. Remained the same

Build # 25012343577

Build Type

Pull #1224

github

Committed by

dunglas

Commit Message

fix: cap default topic selector cache to 100k entries

`TopicSelectorStore` keeps a `(topic_selector, topic) -> bool`
otter LRU. The previous default of 2_560_000 entries was sized for
very large bare-metal deployments. At ~100 B/entry (otter
overhead + the two strings) the cache can grow to roughly 256 MB
before evicting. On common K8s pod sizes (128-256 MiB) it pins
the heap to the cgroup ceiling and the runtime burns >90% of CPU
in `gcBgMarkWorker` trying to stay under GOMEMLIMIT (90% of
cgroup, set by Caddy's automemlimit). We hit this on a busy
production hub: 7 of 30 pods stuck at 600m CPU, no actual
applicative work, just GC trashing on the cache.

Drop the default to 100_000 (~10 MB max). Fits any reasonable
pod, still absorbs the typical hub working set, doesn't change
behavior for hubs that already configure
`topic_selector_cache <N>` explicitly.

Pull Request Pull Request #1224: fix: cap default topic selector cache to 100k entries

Coverage Stats

1744 of 2079 relevant lines covered (83.89%)

49.8 hits per line

Jobs

ID	Job ID	Ran	Files	Coverage
1	0 - 25012343577.1	27 Apr 2026 06:28PM UTC	23	83.89	GitHub Action Run

dunglas / mercure / 25012343577
84%
master: 93%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Jobs

Source Files on build 25012343577

dunglas / mercure / 25012343577 84% master: 93%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Jobs

Source Files on build 25012343577

dunglas / mercure / 25012343577
84%
master: 93%

README BADGES
x