You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
fix: drop wall-clock total from MCP relay stream timeouts (#2034)
* fix: drop wall-clock total from MCP relay stream timeouts
An MCP response stream is long-lived by design (the upcoming spec's
subscriptions/listen holds one open indefinitely); the relay's
total=300 wall-clock bound cut healthy streams every 5 minutes and
forced clients to re-subscribe. sock_read idle detection still bounds
a dead stream.
Applied to the ha_mcp_tools component webhook relay and the webhook
proxy dev flavor (2.0.5.dev2); the stable proxy picks it up on the
next promote.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017iTWsCSMPqSHm7Po8yGzfk
* fix: address Codex relay-timeout findings
Add a finite connect bound (30 s) to both relay ClientTimeouts: it
covers connection-POOL acquisition, so a pool occupied by long-lived
streams fails a new request instead of hanging it forever now that
total is gone. Replace the hard-coded stable relay_timeout_total test
expectation with a source-derived predicate that flips automatically
when the promote workflow copies the dev tree — a hard-coded 300 would
have failed every generated promotion PR.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017iTWsCSMPqSHm7Po8yGzfk
* chore: re-bump webhook-proxy-dev to 2.0.5.dev3 after #2033 merged dev2
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017iTWsCSMPqSHm7Po8yGzfk
---------
Co-authored-by: kingpanther13 <kingpanther13@users.noreply.github.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
0 commit comments