Neilk1021 opened a new pull request, #7391:
URL: https://github.com/apache/texera/pull/7391
### What changes were proposed in this PR?
Closes #7031 by keeping the coordinator from sending `queryStatistics` to
workers of a region whose teardown it has already started.
- `RegionExecutionManager` sets an `endWorkerSentRef` flag as the first
statement of `terminateWorkers` and exposes it as `isTerminating`, plus a
`containsPhysicalOp` membership test. The flag is deliberately *not* derived
from the existing `terminationFutureRef`. That reference is compared and
swapped only after `terminateWorkersWithRetry` has been constructed, and
constructing it already runs `terminateWorkers`.
- `WorkflowExecutionManager.isRegionTerminating(opId)` resolves an
operator to a terminating region. Documented as must-be-called at the point of
use rather than cached: a caller spanning several coordinator rounds would
otherwise act on an answer from before a region started tearing down, which is
exactly the window being
closed.
- `QueryWorkerStatisticsHandler` consults it alongside the existing
filters and skips those operators.
### Any related issues, documentation, discussions?
Closes #7031.
Part of #6891. This is the request-side half. #6960 fixed the reply-side
half; the two are independent and this PR does not change anything #6960
introduced.
### How was this PR tested?
Three unit tests added to the existing harnesses, no new harness:
- `WorkflowExecutionManagerSpec`: drives a real two-region schedule to the
state where region 1's `EndWorker` is on the wire and unanswered, then asserts
`isRegionTerminating` is true for that region's operator, false for the
not-yet-started region, and still true after termination completes. It also
asserts the operator aggregates as `RUNNING` at that moment, which is what
pins the point of the fix: the window is real, and the pre-existing `COMPLETED`
skip could not have covered it.
- `RegionExecutionManagerSpec`: `isTerminating` is false while the region
is merely running and observably true by the time `EndWorker` is emitted
(checked from inside the RPC probe's callback, not merely afterwards); and a
region recognises only its own operators.
```
sbt "WorkflowExecutionService/testOnly *RegionExecutionManagerSpec
*WorkflowExecutionManagerSpec *PortCompletedHandlerSpec *EndHandlerSpec"
```
### Was this PR authored or co-authored using generative AI tooling?
Generated-by: Claude Code (Opus 5)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
