mirror of
https://github.com/mattermost/mattermost.git
synced 2026-09-19 10:12:47 +08:00
9de0027868
* [MM-69982] Fix intermittent remote cluster ping failures from stale keep-alive reuse The remote cluster HTTP transport set IdleConnTimeout to 90s while pings fire every PingFreq (60s) and peers close idle keep-alive connections after their own IdleTimeout (default 60s). Because 90s > 60s, the pooled ping connection outlived the peer's reaping, so each 60s ping reused a connection the peer had already closed, racing the FIN and failing intermittently with EOF / connection reset. Set IdleConnTimeout to PingFreq/2 so the pool always discards the connection before the next ping fires, making the reuse race structurally impossible while keeping the invariant (IdleConnTimeout < PingFreq) explicit and tied to PingFreq. Co-authored-by: mattermost-code <matty-code@mattermost.com> * chore: retrigger Server CI after check-style Docker Hub flake Server CI check-style failed pulling buildenv (Docker Hub Client.Timeout / missing buildenv-image artifact); unrelated to IdleConnTimeout changes. Enterprise CI check-style already passed on the same commit. Co-authored-by: mattermost-code <matty-code@mattermost.com> --------- Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: mattermost-code <matty-code@mattermost.com>