Skip to content

Orphan iOS devices stay on disk after a failed shutdown; cleanup doesn't see them #1801

Open

Description

@crockalet

What happened

On a 16 GB Mac under memory pressure, deleting a simulator at lease end runs simctl shutdown with a 30 s timeout. The daemon log shows it timing out, then the device is quarantined and the reclaim retries fail:

device.purge-failed  DriverCrashError: simctl shutdown <udid> timed out after 30000ms
device.quarantined   kind: shutdown-failed
quarantine reclaim retry failed (attempts 3)

Some of those simulators stay in ~/.simlock/devices/ios with no row in simlock list. I had three (1.2 to 1.5 GB each), left on 0.7.0. After upgrading to 0.8.0 and restarting the daemon they were still there, and only doctor saw them:

$ simlock --version
0.8.0
$ simlock list            # only the one live lease
$ simlock cleanup --dry-run
[]
$ simlock doctor
{"findings":[{"kind":"orphan-device", ... "platform":"ios"}, (x3) ...]}

A handful of these push the disk under the 10 GB pressure line. Two orphans in an earlier batch came from provisions whose requests were then rejected with unresolvable-spec, so a rejected request may also leave its new simulator behind.

This looks like the shutdown/delete side of #1546. #1563 bounded clone/create and deletes a late copy, but the shutdown at lease end still has the fixed 30 s limit, and a daemon restart does not reconcile the orphans.

What I expected

  • A quarantined device whose retries are used up gets a last forced simctl delete.
  • cleanup (or a periodic doctor reconcile) removes orphan-device findings.
  • A rejected request deletes the simulator it created.

How to reproduce

The shutdown timeout only happened under memory pressure (swap about 2.5 GB, pressure "warn"), so that part is intermittent. Once orphans exist, the list / cleanup / doctor check above shows them every time.

Environment

simlock 0.8.0 (orphans created on 0.7.0), macOS 27, iOS 27.2 runtime.

Workaround

simlock doctor --purge-orphans --yes removed all three on 0.8.0 without a daemon restart. A restart alone does not clear them.

Suggested fix

Give the shutdown at lease end the same kind of setting #1563 added for clone/create, and run the orphan reconcile from cleanup or on a timer.

Activity

  1. added
    bug:newReported. An agent may triage it (rule 8); the fast track may set it ready (rule 17).
    on Nov 22, 2026
  2. self-assigned this
    on Nov 22, 2026
  3. V3RON commented on Nov 22, 2026

    @V3RON
    Contributor

    Reproduction

    This is a decision, not a defect. ADR 0001 §6 rejects reaping orphans automatically. The three orphans match #1546 (a clone cut off at 30 s on 0.7.0), which #1563 fixed in 0.8.0 for new clones. The test's expectation is the reporter's proposal.

    src/core/doctor-unobservable-platform.test.ts, "a daemon start's reconcile destroys every orphan device in the owned root without --purge-orphans". Branch bug/1801-repro.

    pnpm exec vitest run --project unit src/core/doctor-unobservable-platform.test.ts
    AssertionError: expected [] to deeply equal [ …(2) ]   (no device.orphan-purged for orphan-1, orphan-2)
     ❯ src/core/doctor-unobservable-platform.test.ts:126:7
    

    Root cause

    src/core/doctor.ts:216-226 purges orphans only under --purge-orphans. cleanup rules (src/core/cleanup/rules.ts:5) read only registry records, so they never see a device with no record. Both follow ADR 0001 §6 and safety rule 1: an orphan has no record and no claim, so an automatic reaper could delete a device that is between driver.provision and registerDevice for a live request. Two tests pin this (doctor.test.ts:478, doctor-unobservable-platform.test.ts:65). A lease-end shutdown failure does not create orphans. The device stays registered and quarantined, and a retry deletes it (quarantine-coordinator.ts:306-327). A rejected unresolvable-spec is the second ProvisionFailedError (acquisition-workflow.ts:549-553): on 0.7.0 that was the timed-out clone.

    Simplest fix

    None inside the rules. Making reconcile or cleanup purge needs an ADR that supersedes ADR 0001 §6 and answers its mid-provision race, so it needs a spec session. If the decision stands, close this as covered by #1546/#1563, with doctor --purge-orphans --yes as the cleanup for 0.7.0 leftovers.

    Alternatives rejected

    Risk

    Fast track

    • Defect: no — a decision (ADR 0001 §6)
    • Fix tried: no — the change needs a superseding ADR
    • Nothing public added or broken: no — changes what doctor and cleanup destroy
    • No ADR or rule changed: no — supersedes ADR 0001 §6 and safety rule 1
    • One fix: no — reconcile, cleanup and a timer are three options

    Side findings

    Written by an agent.

  4. removed their assignment
    on Nov 22, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bug:newReported. An agent may triage it (rule 8); the fast track may set it ready (rule 17).

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions