CI flake: Smoketests (Windows)

[clockwork-labs-bot]

Detected by CI Gremlin after a failed GitHub Actions attempt succeeded on rerun.

Tests:

• Smoketests (Windows)

Error:
Could not fetch job logs: Command '['gh', 'api', '-X', 'GET', 'repos/clockworklabs/SpacetimeDB/actions/jobs/89957026823/logs']' returned non-zero exit status 1.

Failed job runs:

• https://github.com/clockworklabs/SpacetimeDB/actions/runs/30159025267/job/89957026823 (Smoketests (Windows), attempt 2 -> 5)

Timeout detection:

• none

Context:

• none

[clockwork-labs-bot]

Exact-attempt diagnosis for run 30159025267, job 89957026823 (Smoketests (Windows), attempt 2): the job failed before the Windows smoketests ran.

GitHub job metadata says:

runner: ghr-acb7d313
started: 2026-07-27T10:54:45Z
completed: 2026-07-27T11:04:45Z
conclusion: failure
last started step: Run dsherret/rust-toolchain-file@v1
later setup and Run smoketests (Windows): pending

The job-specific log endpoint is still unavailable and returns Azure BlobNotFound:

HTTP 404 The specified blob does not exist.
X-Ms-Error-Code: BlobNotFound

Root cause: CI/runner infrastructure failure during early Windows setup, before any smoketest assertion or test body ran. The job ended as failed while GitHub still reports the rust-toolchain action as in progress and all later setup/test steps as pending, and the log blob was never retained/uploaded. This is not diagnosable as a SpacetimeDB product test failure from the exact linked attempt.

Suggested title: CI infra flake: Windows Smoketests runner lost logs during rust-toolchain setup

Duplicate analysis: this matches the same Windows Smoketests missing-log / runner-loss family as #5346, which was closed with the note that it was probably caused by VM-killing cron, and #5545, a recent repeat whose exact job also never reached Run smoketests (Windows) and had a missing job log. Same-family older closed repeats include #5187, #5267, #5345, #5385, #5391, #5392, #5393, #5398, #5401, #5411, #5414, #5418, #5420, and #5423. Open #5260 is also Windows Smoketests setup-related, but it has a concrete dotnet workload update manifest-version failure, so I would not mark this as a duplicate of #5260.

Top duplicate candidate: #5346, if we are treating the Windows runner-lost-log class as one canonical issue. Otherwise, similar tickets exist but none contains recoverable logs for this exact failure, so this ticket should stay classified as runner/log infrastructure rather than a test flake.

[bfops]

run was canceled