stacklok / toolhive / 23442649867

64%

Build Type

push

github

Committed by web-flow

Commit Message

Prevent permanent backend unhealthy marking after startup race (#4290)

fix(vmcp): prevent permanent backend unhealthy marking after startup race

Health checks could permanently mark backends as unhealthy due to two
issues: shared http.DefaultTransport kept stale keep-alive connections
to replaced K8s pods returning 4xx indefinitely, and mcp-go sentinel
errors (ErrUnauthorized, ErrLegacySSEServer) were not recognized by
the error classification chain, causing auth failures to surface as
generic backend unavailability.

- Clone http.DefaultTransport per client factory call to isolate
  connection pools and avoid stale connections after pod replacement
- Map transport.ErrUnauthorized to ErrAuthenticationFailed and
  transport.ErrLegacySSEServer to ErrBackendUnavailable in
  wrapBackendError before falling back to string-based detection
- Add "unauthorized (401)" pattern to IsAuthenticationError to match
  mcp-go's ErrUnauthorized string format
- Poll DynamicRegistry version every 2s and trigger an immediate
  status report when backends are added or removed, rather than
  waiting for the full 30s reporting interval

Closes: #4278

Co-authored-by: taskbot <taskbot@users.noreply.github.com>

Run Details

8 of 21 new or added lines in 1 file covered. (38.1%)

138 existing lines in 1 file now uncovered.

50591 of 78023 relevant lines covered (64.84%)

215.47 hits per line