Igor Sirotin bad12e44b6

fix(filter): correct inverted idle-subscription cleanup check ...

The periodic cleanup in SubscribersMap.cleanUp deleted subscribers whose
elapsed time since last activity was *less* than the idle timeout, and kept
those that had gone idle. This is backwards: active subscribers (that keep
their subscription alive via pings, which call Refresh) were being dropped
within a cleanup cycle, while genuinely idle subscribers were never reaped
and leaked forever.

Flip the comparison so a subscriber is removed only once it has been idle
for longer than the timeout.

Also rewrite TestCleanup (which previously encoded the buggy behavior) and
add TestCleanupKeepsActiveSubscribers, which fails on the old inverted check.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

2026-07-06 09:40:12 +01:00

..

cliutils

chore: remove bridging topics feature (#949)

2023-12-09 13:59:35 -04:00

metrics

fix: return error in relay publish if exceeding max-msg-size (#939)

2023-12-01 06:53:28 +05:30

persistence

feat: log error and stacktrace when panic in goroutine (#1225)

2024-09-25 17:15:20 +08:00

tools

chore: switch to Google's Protobuf library

2023-02-16 11:37:59 -04:00

try

feat: Implement logic for publish from node

2021-12-07 14:32:02 +01:00

v2

fix(filter): correct inverted idle-subscription cleanup check

2026-07-06 09:40:12 +01:00