Repository navigation
Conversation
This comment was marked as off-topic.
This comment was marked as off-topic.
This comment was marked as off-topic.
This comment was marked as off-topic.
This comment was marked as off-topic.
This comment was marked as off-topic.
| ) -> color_eyre::Result<Option<Vec<Build>>> { | ||
| let jobs = jobs | ||
| .into_iter() | ||
| .filter(|job| !self.name_to_build.contains_key(&job.name)) |
There was a problem hiding this comment.
Restart now preserves running and succeeded builds. The retry reads current jobset inputs and writes their newly computed hash here, but this filter skips incoming jobs solely because their names already exist.
The consequence of this is that evaluation can complete using the old build while being cached under the new inputs hash, i.e., metric ton of cache misses. Previously restart deleted all builds before re-eval.
| circus_notification::dispatch_build_created( | ||
| pool, | ||
| build, | ||
| project, | ||
| &eval.commit_hash, | ||
| notifications_config, | ||
| notification_secret_key, | ||
| ) |
There was a problem hiding this comment.
An early batch can synthesize dependency build B, which finishes before a later batch discovers B's top-level job name. Its original completion notification is suppressed because dependency-prefixed jobs are internal. The callback above then invokes only the created notification path, which excludes Slack, email, and generic-webhook channels, those channels never receive B's named-job completion
|
CI is failing by the way. |
Agents sat idle for the whole evaluation on large jobsets. Each batch commits with its dependency builds and edges, so list_pending never sees a build before what it waits on.
b867a32 to
79d4410
Compare
4 participants
No description provided.