Repository navigation
Conversation
… origin outside the layer, no real clean in tests
The selection tests run from project directories under the system temp dir with --no-bind-cwd. When /tmp is a tmpfs, it is not part of the layer's overlay, so the project directory did not exist inside the layer and the layer failed to start with "failed to set CWD". Leave the cwd bind on, as it is for a user, and write the out-of-project file for the status check into the repository, which is always overlaid.
…mes in comments The internal rename pass turned the limits test's /sys/fs/cgroup/sandbox into /sys/fs/cgroup/layer, but the cgroup parent is devlayer, so the test never found the layer's cgroup. Also correct comments that named LAYER_IGNORED, LAYER_NO_DEFAULT_BINDS, .layer.toml, ~/.config/layer, /etc/layer.conf and /sys/fs/cgroup/layer.
The devlayer command waits on its child inside the layer's cgroup, so when the cgroup ran out of memory the kernel sometimes killed devlayer instead of (or as well as) tail, and the test got no exit status. tail now raises its own oom_score_adj to 1000 so it is always picked first.
… keep test layers up explicitly
… command is starting in them
… naming the layer's own PID 1
… run coverage on ubuntu-latest again
- Resolve the storage dir once; record the layer name's source per branch - Store a single LayerOrigin in Config; skip loading it for new layers, so a concurrent launch's origin isn't picked up - Only remove an unparseable pid file while holding the layer lock, so list can't remove one that is being written - Share pid/lock path helpers between Layer and util - Pass only PID 1's pid from the supervisor; drop redundant setresuid/gid - config: report bind_cwd's real source, drop the no_bind_cwd key - Check accept/reject patterns before choosing the layer - Parse true/false env vars in one loop; bad values error, empty ignored - Make LLVM_PROFILE_FILE absolute in the Makefile, so the supervisor's coverage data lands in coverage/profraw - Fix flaky ai_test_limits_applied; share pid file helpers in the fixture
After tail is OOM-killed, head can still be writing to the pipe before tail's memory is freed, which triggers a second OOM kill. With head at the default oom_score_adj, that second kill took the waiting devlayer command, so the test got no exit status (about 1 in 40 runs under load).
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.
No description provided.