restart the app: 'reconnect with pre-shared token: agent not reachable within 20s' in about 1 of 25 restarts, on fixture builds by both Flutter 3.44.8 and 3.50.0

Open
#823 1 comment 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
3/5
Estimated time
1-2 days
Newbie friendliness
68/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
android, dart, flutter, go
Domain
mobile, testing

Research direction

Start with the reconnect loop in internal/runner/device_context.go, then trace how agent.token_read_timeout controls its deadline. First measure time from app relaunch to the first successful dial on idle and loaded machines, as the issue suggests. Done means identifying why reconnects exceed the 20-second timeout and adding a focused regression test for the failure.

Written by the indexing model from the issue text.

Description

bug

What happens

In a suite that has before each test / restart the app, the restart sometimes fails before the test starts:

before each: line N: restart the app: restart the app: reconnect with pre-shared token: agent not reachable within 20s

The test never runs; the other tests of the same run pass. It is a failed reconnect after the app was force-stopped and relaunched, not an assertion. It showed up while running the new widget-family suites on one Android emulator, on builds of the fixture by both Flutter 3.44.8 and 3.50.0.

Measurements (everything I kept; not a controlled experiment)

Setup: the Flutter fixture app of this repository (native-test-apps/flutter, --dart-define=PROBE_AGENT=true --dart-define=PROBE_FIXTURE_WIDGETS=<family>), one emulator (Android 15, API 35, emulator-5600), the CLI and agent built from main of the day (the agent is a path dependency). Suites of 10 to 12 tests, each test starting with restart the app and a wait until #<id> appears.

probe.yaml of the runs (the same in all of them): project.app: com.alphawavesystems.probe_flutter_fixture, agent.port: 48690, agent.device_port: 48686, agent.dial_timeout: 30s, agent.token_read_timeout: 20s. No defaults.retry_failed_tests, no --retry: nothing re-ran a test. probe test <file> --device <serial> -y.

Flutter build runs kept runs with at least one restart failure tests run restart failures
3.44.8 (suites) 15 9 178 9
3.44.8 (negative files and small diagnostic files) 15 5 124 5
3.50.0 (suites and negative files), first 9 runs 9 0 101 0
3.50.0, 4 more runs made later (2 suites, 2 negative files) 4 3 42 4

Added after the first version of this issue: the 4 later 3.44.8 runs (2 suites, 2 negative files, 42 tests) had 2 more failures, so on 3.44.8 16 of about 344 restarts failed (about 5%). On 3.50.0, 4 of about 143 failed (about 3%): the first 9 runs had none, the 4 later ones 4 (one run had 2 failures). So this is not specific to one Flutter version; my first reading (3.44.8 only) came from the 3.50.0 runs happening to be the quiet ones. Apart from that run, each run had at most one failure; the failing test was a different one each time (first test, middle test, last test), and it did not follow a test that left a dialog or a snack bar open (it also happened after plain tests).

What I do not know

  • Machine load was not measured. Other sessions were building and testing on the same Mac, and the 3.44.8 runs were mostly made later in the day than the 3.50.0 ones, so the Flutter version and the load are confounded. Another developer measured the same failure on main before my runs: 3 of 10 runs with a heavy test run in parallel and 0 of 10 when the machine was idle. I did not repeat that.
  • Whether the app comes up late (a slow cold start after am force-stop), or the agent port is bound late, or the dial loop gives up too early: the error only says that the agent was not reachable within token_read_timeout (20 s).
  • Whether it happens on iOS.

Where to look

internal/runner/device_context.go (the reconnect after a restart: a plain dial loop that fails when the app does not come up within agent.token_read_timeout). A first measurement would be to log the time from the relaunch to the first successful dial in a run that fails and in a run that does not, on an idle machine and on a loaded one, for both Flutter builds.

Not this

  • #793 (steps right after a restart, the screen still being built) is a different problem: here the reconnect itself fails.
  • The 0x0 / first-frame problem fixed in #803 is not this either, according to the developer who measured it (a failed WebSocket dial).
Dominant language
Go
Stars
6
Forks
3
Avg merge
6h 28m
Merged PRs (30d)
362

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 AlphaWaveSystems/flutter-probe

All issues in AlphaWaveSystems/flutter-probe

Similar issues

More Go issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.