I think something changed on GitHub runners (maybe the runner image?) around June 11th that is causing runners launching the emulator to hang.
I have a workflow job that runs tests that using the emulator. The job is defined in https://github.com/pachli/pachli-android/blob/main/.github/workflows/ci.yml#L168.
The last successful run of this job on any PR was https://github.com/pachli/pachli-android/actions/runs/27281635488
- At: 2026-06-11 10:29 GMT+2
- Emulator test logs:
On 2026-06-11 03:47 GMT+2 (i.e,. before the last successful run) this job started failing on PR pachli/pachli-android#2328. This PR doesn't change code, it changes Android string resources, so there's nothing obvious in there that would cause the runner to hang.
The first failing run is https://github.com/pachli/pachli-android/actions/runs/27309207203. The job uses a matrix to launch three sub-jobs, of those three, one passed, one failed, and one was cancelled because one failed.
Logs for the passing job: https://github.com/pachli/pachli-android/actions/runs/27318340445/job/80703732950
Logs for the failed job: https://github.com/pachli/pachli-android/actions/runs/27318340445/job/80703732925
The failed job doesn't show an overt failure message, it just appears as though the runner is hanging. Not the emulator process, the runner as a whole. The logs just stop, and importantly, they stop before logging any tests starting on the emulator. At this point the emulator is running, but Gradle is still building the APK that will be tested.
This behaviour -- the failures initially being intermittent, before occurring 100% of the time -- is consistent with a gradual rollout of a new runner image with changes that break the runner.
Since then, any attempt to run this job hangs, with similar symptoms.
Thinking this might be a GitHub resource issue on the default runners I also tried this runners provided by https://www.blacksmith.sh/, including their blacksmith-2vcpu-ubuntu-2404 runner (roughly the same as a GitHub runner) and blacksmith-4vcpu-ubuntu-2404 runners (roughly 2x the resources of a GitHub runner), but the tests hang there too. I believe they use the same runner image as GitHub, which would explain the identical behaviour.
I think something changed on GitHub runners (maybe the runner image?) around June 11th that is causing runners launching the emulator to hang.
I have a workflow job that runs tests that using the emulator. The job is defined in https://github.com/pachli/pachli-android/blob/main/.github/workflows/ci.yml#L168.
The last successful run of this job on any PR was https://github.com/pachli/pachli-android/actions/runs/27281635488
On 2026-06-11 03:47 GMT+2 (i.e,. before the last successful run) this job started failing on PR pachli/pachli-android#2328. This PR doesn't change code, it changes Android string resources, so there's nothing obvious in there that would cause the runner to hang.
The first failing run is https://github.com/pachli/pachli-android/actions/runs/27309207203. The job uses a matrix to launch three sub-jobs, of those three, one passed, one failed, and one was cancelled because one failed.
Logs for the passing job: https://github.com/pachli/pachli-android/actions/runs/27318340445/job/80703732950
Logs for the failed job: https://github.com/pachli/pachli-android/actions/runs/27318340445/job/80703732925
The failed job doesn't show an overt failure message, it just appears as though the runner is hanging. Not the emulator process, the runner as a whole. The logs just stop, and importantly, they stop before logging any tests starting on the emulator. At this point the emulator is running, but Gradle is still building the APK that will be tested.
This behaviour -- the failures initially being intermittent, before occurring 100% of the time -- is consistent with a gradual rollout of a new runner image with changes that break the runner.
Since then, any attempt to run this job hangs, with similar symptoms.
Thinking this might be a GitHub resource issue on the default runners I also tried this runners provided by https://www.blacksmith.sh/, including their
blacksmith-2vcpu-ubuntu-2404runner (roughly the same as a GitHub runner) andblacksmith-4vcpu-ubuntu-2404runners (roughly 2x the resources of a GitHub runner), but the tests hang there too. I believe they use the same runner image as GitHub, which would explain the identical behaviour.