Summary
The :darwin: 26 aarch64 - test-bun Buildkite jobs are failing on most PR builds. The runner never starts a single test; it fails downloading the build artifact:
Error: buildkite-agent artifact download timed out after 120s for step 'darwin-aarch64-build-bun'. Refusing to continue with a partial download (would silently fall back to the wrong binary).
at getExecPathFromBuildKite (file:///Users/administrator/Library/Services/buildkite-agent/builds/darwin-aarch64-26-5-1-1/bun/bun/scripts/runner.node.mjs:2182:13)
at async runTests (.../scripts/runner.node.mjs:481:16)
Scope
Of the 30 most recent failed/failing builds on the bun pipeline, 22 have a failed :darwin: 26 aarch64 - test-bun job, across unrelated branches. All of them hit the same artifact download timeout:
| build |
branch |
| 67012 |
farm/fc45f163/react-compiler-hoisting-overreach |
| 67008 |
farm/833a1f2f/win-appcontainer-paths |
| 67007 |
claude/json-structural-index |
| 67004 |
claude/node-http-http2-compat |
| 66987 |
claude/navigator-clipboard-readtext-writetext |
| 66982 |
farm/dc5e99a7/serve-route-target-normalization |
| 66979 |
farm/9a83c90d/fix-symlink-fd-cache |
| 66976 |
farm/e3947f67/bump-ci-service-containers |
| 66971 |
farm/ed7cafd7/node-sqlite |
| 66968 |
claude/port-node-inspector-tests |
| ... |
(12 more) |
I saw it reproduce on two consecutive builds of the same PR (66995 and 67012), both on the agent darwin-aarch64-26-5-1-1, about 35 minutes apart. So it is not a one-off blip.
Likely causes
- The
darwin-aarch64-26 agent fleet (or just darwin-aarch64-26-5-1-1) has slow or broken connectivity to the Buildkite artifact store, so the 120s download budget in scripts/runner.node.mjs:getExecPathFromBuildKite is not enough.
- Or the
darwin-aarch64-build-bun artifacts are larger or slower to serve than they used to be.
Impact
Every PR that lands on an affected agent goes red on a lane that ran zero tests, which both hides real failures and blocks merges. The job does not get an automatic Buildkite retry (exit status 1 from the runner), so each occurrence needs a manual "Retry job" click.
Summary
The
:darwin: 26 aarch64 - test-bunBuildkite jobs are failing on most PR builds. The runner never starts a single test; it fails downloading the build artifact:Scope
Of the 30 most recent
failed/failingbuilds on thebunpipeline, 22 have a failed:darwin: 26 aarch64 - test-bunjob, across unrelated branches. All of them hit the same artifact download timeout:I saw it reproduce on two consecutive builds of the same PR (66995 and 67012), both on the agent
darwin-aarch64-26-5-1-1, about 35 minutes apart. So it is not a one-off blip.Likely causes
darwin-aarch64-26agent fleet (or justdarwin-aarch64-26-5-1-1) has slow or broken connectivity to the Buildkite artifact store, so the 120s download budget inscripts/runner.node.mjs:getExecPathFromBuildKiteis not enough.darwin-aarch64-build-bunartifacts are larger or slower to serve than they used to be.Impact
Every PR that lands on an affected agent goes red on a lane that ran zero tests, which both hides real failures and blocks merges. The job does not get an automatic Buildkite retry (exit status 1 from the runner), so each occurrence needs a manual "Retry job" click.