Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

iOS: a canceled request releases the session lock while its runner command still runs on the device

Closed
#3,383 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

A pull request for this has already been merged.

  • #3385 by @thymikee — merged

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
22/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Stale
Tech stack
ios, typescript
Domain
backend, mobile

Research direction

Start in src/daemon/request-execution-locks.ts and packages/kernel/src/keyed-lock.ts to see when the session and device locks are released, then in packages/platform-apple/src/runner/runner-exchange.ts where an abandoned charge is marked. Done means a test in which a canceled runner command keeps the session lock until the runner journal reports it terminal, plus the live iOS press-then-relaunch check. Linked PR #3385 is already open against this, so check it before starting.

Written by the indexing model from the issue text.

Description

Problem

A per-call signal (#3178) cancels a request, but on iOS the daemon releases the session's lock before the device has finished the canceled work. The next request on the same session can then run while the aborted command is still executing on the device.

  • The per-session and per-device keyed locks are held by the handler's promise (request-execution-locks.ts, keyed-lock.ts).
  • On abort, the runner fetch rejects at once, the exchange marks the charge abandoned and rethrows (runner-exchange.ts), and the handler settles. The lock is released.
  • The runner never sees the disconnect. Its serial commandExecutionQueue runs the command to completion.
  • Only runner commands queue behind it. A next request whose device work skips the runner, such as a simctl launch from open --relaunch, simctl privacy, or a screenshot, can overtake the tap. So can the runner's inline status and uptime.

Android holds the lock until adb returns, so it is serialized already.

Consumer impact: tester-army/e2e cannot adopt signal. It keeps an inflight set of abandoned commands and waits for each response before the next attempt starts, so that a tap from a canceled step cannot land in the retry (packages/mobile/src/surface.ts). With signal, the client's promise rejects immediately, so that wait no longer proves the device finished.

Required behavior

  • A canceled request whose runner command was already sent keeps its session and device locks until the runner reports that command terminal (completed or failed in the command journal), or until the command's own deadline expires. The client still rejects at once with request_canceled.
  • A command that never reached the runner, or one the runner refused, releases at once, as it does today.
  • A drain that hits the deadline leaves the charge abandoned, as today, so the detach/handoff rules keep seeing it.

Done when

  • A test shows that a canceled runner command holds the session lock until the runner journal reports it terminal, and that a second request on the session starts only after that.
  • Live on an iOS simulator: abort a long press (or longpress --duration) through the client, then immediately run open --relaunch. The launch starts after the press finishes, as the request log shows.

Related: #3178 (per-call signal), #2965 (charge settlement by terminal evidence).

Dominant language
TypeScript
Stars
4.9k
Forks
328
Avg merge
12h 18m
Merged PRs (30d)
561

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from callstack/agent-device

All issues in callstack/agent-device

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.