lsm / HyperNeo / 32188213040

80%

push

github

test(daemon): condition-based kickoff/probe waits in task-agent-lifecycle (#2573)

* test(daemon): condition-based kickoff/probe waits in task-agent-lifecycle (#998)

The kickoff and probe tests waited for agent 'idle' right after the
spawn poll, then gave SDK messages a fixed 5s to appear. Both signals
are wrong-side: the execution's agentSessionId is persisted before the
kickoff injection runs, and a fresh session's processing state IS
'idle' until the queue picks the kickoff up — so the idle wait returns
vacuously and the 5s message budget alone races kickoff injection +
SDK cold start. Under CI scheduling jitter (PR #2549, run 31980100328)
it starved and the adjacent message-asserting tests failed as a pair
(space-a flaked twice on different adjacent pairs: kickoff+spawn-probe,
then spawn-probe+check-status-probe — same shared wait structure).

Wait on the asserted conditions instead: kickoff context = first SDK
message persisted; probe processed = an assistant message beyond the
pre-send baseline. Budgets are cold-start scaled (40s mock, covering
the 30s startup gate), the spawn budget doubles to 30s mock (observed
starving 15s locally; the runtime also retries spawn delivery once on
'delivery not consumed'), and TEST_TIMEOUT rises to cover the summed
worst-case waits. Verified locally: two consecutive 5/5 dev-proxy runs
with targeted probe mocks still matching.

* test(daemon): tolerate transient fetch failures in lifecycle wait loops (#998 r1)

message.sdkMessages hard-throws 'Session not found' when the RPC-side
session load has not completed (SessionCache.getAsync funnels transient
load errors into null), so one transient failure inside the new 40-90s
poll windows aborted the test with an error misread as a spawn bug.
Both loops now fetch through a tolerant wrapper that treats a throw as
"condition not yet met" and keeps polling to the deadline — the same
transient-error tolerance waitForProcessingState already applies.

43723 of 59339 branches covered (73.68%)

Branch coverage included in aggregate %.

59583 of 71514 relevant lines covered (83.32%)

3794.21 hits per line