fix: await owned daemon retirement on request timeout

[thymikee]

Summary

A local request timeout could remove daemon metadata before its daemon exited. Socket and HTTP timeouts now await the same ownership-checked retirement operation before reporting reset. Unconfirmed exit retains state and is included in the primary timeout error.

Removes the client’s raw metadata-removal and takeover-stop exports. Read-only command policies, remote behavior, Apple runner pattern bytes and request fallback rules stay intact. Health probes reserve time for a fallback and count requester loading and response completion within their budget.

Depends on #3130. Ref #3116. Eight files, 616 gross changed lines; production: 107 additions/143 deletions. The final enforcement commit acknowledges one implementation-only health-reader change, with no RPC payload or protocol-version change.

Validation

Tested 2afb5ec51ab198cc380f6db6a2c6d88d44965241:

• 61 focused tests, including actual socket/HTTP requests and owned children; zero skips.
• Removing the await fails both force-retirement controls. Removing fallback reservation or loader accounting each fails its control.
• Remaining retirement-owner tests catch premature unlink: 31 passes → 3 failures/28 passes. Obsolete wrapper tests and seams were removed together.
• 42 wire compatibility/closure/mutation controls pass.
• pnpm check:quick, Fallow, and pnpm check:affected --base fix/daemon-startup-contention --run pass; 925 related tests.
• Independent read-only review: no findings. CI-owned coverage, provider and device evidence remains pending.

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	4.96 MB	4.96 MB	+499 B
Package (unpacked)	4.95 MB	4.96 MB	+499 B
Package (download)	1.49 MB	1.49 MB	+111 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	27.1 ms	26.7 ms	-0.4 ms
CLI --help	83.4 ms	82.1 ms	-1.3 ms

[cubic-dev-ai]

All reported issues were addressed across 8 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[thymikee]

This PR is ready at 2afb5ec. The change now waits for the owned daemon to retire after a request timeout, and I found no problem in the code.

Not blocking, and you can take or leave these: (1) the local retirement branch in https://github.com/callstack/agent-device/blob/2afb5ec/src/daemon-client/daemon-client-transport.ts#L269 keys on the public daemon_transport_timeout reason, so only a timer this client armed should trigger it, for example through a module-private error class that requestTimeoutError creates and the sendRequest catch checks with instanceof; (2) the halved canConnect budget and the health-probe race in https://github.com/callstack/agent-device/blob/2afb5ec/src/daemon-client/daemon-client-transport.ts#L79 look like the startup-deadline work from #3130, so could that move there or be named in the summary; (3) the hint in https://github.com/callstack/agent-device/blob/2afb5ec/src/daemon-client/daemon-client-timeout.ts#L110 always says the daemon could not be safely retired, even when the daemon was killed and the result is lock-busy or registration-replaced, and it drops the Apple cleanup wording, so the wording could follow retirement.reason; (4) the timeout error now returns after the request budget plus up to 1s for exit after SIGKILL and up to 1s for the retirement lock, which is bounded but not documented or asserted.

CI is likely unrelated to this change. The Coverage failure is the slow-test gate on daemon-client-startup-race.test.ts:207 ("startup uses one deadline when the claim is held"), which came in with #3130 and is unchanged here. Its held-claim route never publishes daemon.json, so it never reaches canConnect or sendRequest. The Smoke failure is the iOS preflight daemon_startup_failed with hasInfo:false, so the daemon never published its registration within the 15s budget. The changed canConnect only runs after that info exists, so the changed probe was not reached. The same signature appears on #3129.

I did not run the mutations the PR describes (removing the await, the fallback reservation, the loader accounting). I judged the regression tests from a read of the pre-change code. I could not confirm that #3129 lacks this PR's canConnect change, so the Smoke attribution rests on the hasInfo:false evidence. I also did not confirm the Coverage failure on #3130's own CI run, so "inherited" rests on file and route provenance.

No conflicts. Before merge, the Coverage failure must be fixed in #3130 by injecting the startup clock or poll budget. Then Smoke should be rerun after #3130 lands.

[thymikee]

Consolidated into #3127 as part of reducing #3116 to seven PRs. The composition preserves the complete pre-consolidation source tree, including tests and later review corrections. This PR is superseded; its review discussion and native evidence remain available. Outstanding findings transfer to the owning keeper in the implementation record.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-10-03 21:16 UTC