azagrebin commented on a change in pull request #12278:
URL: https://github.com/apache/flink/pull/12278#discussion_r441343582
##########
File path:
flink-runtime/src/main/java/org/apache/flink/runtime/jobmaster/slotpool/SlotPoolImpl.java
##########
@@ -592,6 +589,31 @@ private PendingRequest findMatchingPendingRequest(final
AllocatedSlot slot) {
return null;
}
+ private void maybeRemapOrphanedAllocation(
+ final AllocationID allocationIdOfRequest,
+ final AllocationID allocationIdOfSlot) {
+
+ final AllocationID orphanedAllocationId =
allocationIdOfRequest.equals(allocationIdOfSlot)
+ ? null : allocationIdOfRequest;
+
+ // if the request that initiated the allocation is still
pending, it should take over the orphaned allocation
+ // of the fulfilled request so that it can fail fast if the
remapped allocation fails
+ if (orphanedAllocationId != null) {
+ final SlotRequestId requestIdOfAllocatedSlot =
pendingRequests.getKeyA(allocationIdOfSlot);
+ if (requestIdOfAllocatedSlot != null) {
+ final PendingRequest requestOfAllocatedSlot =
pendingRequests.getByKeyA(requestIdOfAllocatedSlot);
+
requestOfAllocatedSlot.setAllocationId(orphanedAllocationId);
+
+ // this re-insertion of initiatedRequestId will
not affect its original insertion order
+ pendingRequests.put(requestIdOfAllocatedSlot,
orphanedAllocationId, requestOfAllocatedSlot);
+ } else {
+ // cancel the slot request if the orphaned
allocation is not remapped to a pending request.
+ // the request id can be null if the slot is
returned by scheduler
+
resourceManagerGateway.cancelSlotRequest(orphanedAllocationId);
Review comment:
Alright, so this is something like release the slot at the RM at the
same time.
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
