rkhachatryan commented on code in PR #21981:
URL: https://github.com/apache/flink/pull/21981#discussion_r1112741455
##########
flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/adaptive/allocator/SlotSharingSlotAllocator.java:
##########
@@ -121,16 +133,13 @@ public Optional<VertexParallelismWithSlotSharing>
determineParallelism(
slotSharingGroupParallelism.get(
slotSharingGroup.getSlotSharingGroupId()));
- final Iterable<ExecutionSlotSharingGroup>
sharedSlotToVertexAssignment =
+ final List<ExecutionSlotSharingGroup> sharedSlotToVertexAssignment
=
createExecutionSlotSharingGroups(vertexParallelism);
- for (ExecutionSlotSharingGroup executionSlotSharingGroup :
- sharedSlotToVertexAssignment) {
- final SlotInfo slotInfo = slotIterator.next();
-
- assignments.add(
- new
ExecutionSlotSharingGroupAndSlot(executionSlotSharingGroup, slotInfo));
- }
+ SlotAssigner.AssignmentResult result =
+ slotAssigner.assignSlots(freeSlots,
sharedSlotToVertexAssignment);
+ assignments.addAll(result.assignments);
+ freeSlots = result.remainingSlots;
Review Comment:
Because `assignSlots` is called in a loop, and internally iterates over the
`freeSlots`, the complexity here is suboptimal. Lower-priority slots are added
over and over into the queue inside `assignSlots`, resulting in quadratic
complexity.
I was thinking about making free slots part of `SlotAssigner` state (passed
into constructor).
WDYT @dmvk ?
--
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]
