mwkang opened a new pull request, #8798:
URL: https://github.com/apache/storm/pull/8798
## What this PR does
`DefaultScheduler.defaultSchedule` runs the idle-supervisor redistribute
pass twice per
scheduling round — once over the full topology set
(`DefaultScheduler.java:75`), then again
inside the per-leftover `EvenScheduler.scheduleTopologiesEvenly` call
(`EvenScheduler.java:336`).
So the per-topology `nimbus.even.rebalance.max.free.per.topology` cap is
applied twice, and an
under-assigned topology can relocate up to `2 × cap` workers in a single
round.
The fix adds a `redistributeOntoIdle` toggle to `scheduleTopologiesEvenly`:
- the public 2-arg entry point keeps redistribute on, so using
`EvenScheduler` directly is unchanged;
- `DefaultScheduler` runs the single full-set redistribute itself and
delegates per leftover
topology with the toggle off.
The cap now binds once per round, while the full-set round-robin fairness
that lets multiple
topologies share idle slots is preserved (simply dropping the full-set call
would break it).
Follow-up to #8778. The feature is off by default, so there is no behavior
change unless
`nimbus.even.rebalance.idle.supervisor.enabled=true`.
Closes #8797.
## How it was tested
Added `defaultSchedulerAppliesMaxFreeCapOncePerRound`: the DefaultScheduler
path with two idle
supervisors, `max.free.per.topology=1`, and an under-assigned topology — it
relocates two workers
before the fix and exactly one after. Existing
`TestEvenSchedulerIdleSupervisor` and the other
scheduler tests still pass.
--
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]
