fix(client): acknowledge signals after process exit

This commit is contained in:
cabbage committed 2026-10-27 16:18:32 +00:00
1 parent 789342ab4f
commit 30e7bb8545
7 files changed
+465 -8

No files matched your search

+18
View File
@@ -188,6 +188,24 @@ loss marker plus terminal result, restart the same client spool, discard the
server-confirmed terminal row, and complete a fresh command. It also requires
prompt server-session cleanup rather than waiting for a heartbeat tick.
`BH-RUNTIME-09..11` and `HP-RUNTIME-05` exercise signal requests after process
exit, accepted work with no process, terminal state before active-map removal,
revision mismatches, unsupported Windows signals, actual TERM/KILL outcomes,
and supervisor failure. Expected command-level rejection must persist a typed
`SignalResult` and notify the sender without returning a session-fatal error.
Do not claim graceful/forced delivery when no process signal was attempted.
Malformed requests, unknown command ownership, and store failures still fail
explicitly; the fix is not a blanket suppression of callback errors.
`REC-RUNTIME-05` uses actual WebSockets and durable stores. Persist an active
command, disconnect, queue a signal while offline, reopen the same client spool,
and finish the process locally just as the pending signal arrives. Require the
actual executor's negative result to acknowledge the server intent after the
terminal event, then complete fresh work on that same session. Reopen/reconnect
again and require more fresh work with no replay of the acknowledged signal or
original execution. This reproduces the prior endless `signal command ...:
command is not running` reconnect loop without needing a Windows OS fixture.
```sh
docker compose -f deploy/compose.yaml run --rm toolchain \
go test -race ./internal/client/agent ./test/integration/clientagent