Repository navigation
harness-opencode/test: the crashed drive's stand-in host stops sending at the crash (fixes CI on main) - #15
Merged
Merged
Conversation
…g at the crash CI on main failed every run since 89c0e14 on this test: after the `restart` event, the resumed reply came interleaved with deltas of the reply the crash cut off. The chaos crash is in-process, so the first drive's stand-in OpenCode host kept streaming that reply (12 characters every 150 ms) after the crash, and it raced the next drive's recovered reply. A real crash takes the host with it: the stand-in now stops sending at the crash, and the test checks the resumed text exactly. Locally the old test failed 1 run in 6; the new one passed 15 of 15. docs/followups.md: live writes aren't fenced, so a driver still running after a takeover could do the same. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
CI on
mainhas failed every run since 89c0e14 (the push and three daily runs), always on one test: harness-opencode's "a crash while the model streams, after a never tool…" (added in #4). It also failed on one of #10's runs.Why: the test's crash is in-process (the chaos step). The first drive is killed, but its stand-in OpenCode host keeps streaming the reply the crash cut off:
unsafe_send_email returned: email sent to a@b.c (message id …), 12 characters every 150 ms. That's about half a second of deltas after the crash. When the next drive's recovered reply starts streaming within that window, the two interleave after therestartevent. Examples from the CI logs:"nt to a@b.c Recovered from the log: the email we(message id nt out.msg_qnjjsm74""Recovered from the log: the email went to a@b.c nt out."The test's own comment said a real crash writes nothing more; the stand-in didn't model that. So it was a race. It passed when #4 merged and has lost every run since.
Fix: the stand-in host stops sending at the crash, as a real crash takes the host with it. With that, the test checks the resumed text exactly. The text that streamed before the crash is still checked. Locally, the old test failed 1 run in 6; the new one passed 15 of 15. The whole harness-opencode suite passes.
Also in
docs/followups.md: live events aren't fenced. What this stand-in did (a drive's host still streaming after the next drive took over) could happen with a driver still running after a takeover, such as a stalled process.Test and docs only. harness-opencode isn't published, so there's no changeset.
🤖 Generated with Claude Code