-
Notifications
You must be signed in to change notification settings - Fork 19
Bound job spools and surface disk logging failures #1600
Copy link
Copy link
Open
Labels
area:loggingThe debug record: call log, trace, journalThe debug record: call log, trace, journalarea:sessionThe engine — turns, tasks, the toolbelt, checkpointsThe engine — turns, tasks, the toolbelt, checkpointsbugSomething the code does that it should notSomething the code does that it should notsev:criticalData loss, money spent wrongly, a false done, or the merge queue blockedData loss, money spent wrongly, a false done, or the merge queue blocked
Milestone
Description
Activity
Metadata
Metadata
Assignees
Labels
area:loggingThe debug record: call log, trace, journalThe debug record: call log, trace, journalarea:sessionThe engine — turns, tasks, the toolbelt, checkpointsThe engine — turns, tasks, the toolbelt, checkpointsbugSomething the code does that it should notSomething the code does that it should notsev:criticalData loss, money spent wrongly, a false done, or the merge queue blockedData loss, money spent wrongly, a false done, or the merge queue blocked
Part of #1599. Scoped implementation task tracked in PlanDB as
t-tet4jq.Limit every job spool, keep a useful recent tail, and report dropped output or write failures honestly without breaking child pipe draining. Cover large writes, stdout/stderr concurrency, close, and failure injection with deterministic regressions. Update the manual and PERF.md in the same change.
Acceptance: deterministic regression through the relevant real job/search tool door; unaffected ordinary behavior; exact committed candidate passes affected-package acceptance on Spark and GitHub CI. Full suites and acceptance must never run on the laptop. Draft PR remains draft until production review and fixes are complete; then request @AbirAbbas review. Do not merge before that review.