Repository navigation
relay host connection: test both health-callback write failures; handle a rejecting close on the normal-disconnect path #791
Description
Follow-up from the #787 review (non-blocking, LOW).
What the fix does:
- Tests drive a failed host-record write from BOTH health callbacks (the initial health probe and the periodic one) and assert the diagnostic names the record path and reason and contains no record contents.
- The normal-disconnect path in the relay host connection handles a rejecting
close()the same way the reconnect path does since fix(cli): bound retained relay channels and report host record refresh failures #787: the channel is retained andstop()keeps the host record and reports the close failure, instead of the rejection being swallowed.
Each comes with a test (a mutation that reverts it turns its test red).
Metadata
Metadata
Assignees
Labels
No labels
Type
Projects
- Status
Activity
tps-flint commentedon Nov 6, 2026
Also from the #787 review (Sherlock, non-blocking): the connect-time host-record writes in
relay.ts(the unwrappedpersist()calls around the connect path, ~1279 and ~1283) sit outside the refresher, so a host lock held at connect throws into the loop's catch, tears the channel down and reconnects with no diagnostic. Either route them through the same first-failure diagnostic or say in the docs that the diagnostic covers only the heartbeat and health callbacks.