[FLINK-9427] Fix registration and request slot race condition in TaskExecutor #6067
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…Executor This commit fixes a race condition between the TaskExecutor and the ResourceManager. Before, it could happen that the ResourceManager sends requestSlots message before the TaskExecutor registration was completed. Due to this, the TaskExecutor did not have all information it needed to accept task submissions. The problem was that the TaskExecutor sent the SlotReport at registration time. Due to this, t he SlotManager could already assign these slots to pending slot requests. With this commit, the registration protocol changes such that the TaskExecutor first registers at the ResourceManager and only after completing this step, it will announce the available slots to the SlotManager.
tillrohrmann
added a commit
to tillrohrmann/flink
that referenced
this pull request
May 24, 2018
…Executor This commit fixes a race condition between the TaskExecutor and the ResourceManager. Before, it could happen that the ResourceManager sends requestSlots message before the TaskExecutor registration was completed. Due to this, the TaskExecutor did not have all information it needed to accept task submissions. The problem was that the TaskExecutor sent the SlotReport at registration time. Due to this, t he SlotManager could already assign these slots to pending slot requests. With this commit, the registration protocol changes such that the TaskExecutor first registers at the ResourceManager and only after completing this step, it will announce the available slots to the SlotManager. This closes apache#6067.
tillrohrmann
added a commit
to tillrohrmann/flink
that referenced
this pull request
May 24, 2018
…Executor This commit fixes a race condition between the TaskExecutor and the ResourceManager. Before, it could happen that the ResourceManager sends requestSlots message before the TaskExecutor registration was completed. Due to this, the TaskExecutor did not have all information it needed to accept task submissions. The problem was that the TaskExecutor sent the SlotReport at registration time. Due to this, t he SlotManager could already assign these slots to pending slot requests. With this commit, the registration protocol changes such that the TaskExecutor first registers at the ResourceManager and only after completing this step, it will announce the available slots to the SlotManager. This closes apache#6067.
GJL
approved these changes
May 24, 2018
|
|
||
| log.debug(message); | ||
| throw new SlotAllocationException(message); | ||
| throw new TaskManagerException(message); |
Member
There was a problem hiding this comment.
Why not return an exceptional future here?
Member
There was a problem hiding this comment.
We already talked about it offline.
| ResourceManagerGateway resourceManagerGateway = resourceManagerConnection.getTargetGateway(); | ||
| resourceManagerGateway.disconnectTaskManager(getResourceID(), cause); | ||
| if (log.isDebugEnabled()) { | ||
| log.debug("Close ResourceManager connection {}.", |
Member
There was a problem hiding this comment.
I was wondering whether cause can get logged twice:
log.debug("Close ResourceManager connection {}.",
resourceManagerResourceId, cause);
log.debug("Terminating registration attempts towards ResourceManager {}.",
resourceManagerConnection.getTargetAddress(), cause);
Member
There was a problem hiding this comment.
We already talked about it offline.
|
|
||
| final TaskExecutorGateway taskExecutorGateway = taskExecutor.getSelfGateway(TaskExecutorGateway.class); | ||
|
|
||
| assertThat(registrationQueue.isEmpty(), is(true)); |
Member
There was a problem hiding this comment.
nit: assertThat(registrationQueue, is(empty())); gives better failure messages. Now it's just a fancy way of writing assertTrue().
Contributor
Author
There was a problem hiding this comment.
good point. Will change it.
|
|
||
| final TestingTaskExecutorGateway testingTaskExecutorGateway = new TestingTaskExecutorGatewayBuilder() | ||
| .setRequestSlotFunction(slotIDJobIDAllocationIDStringResourceManagerIdTuple5 -> { | ||
| requestSlotQueue.offer(slotIDJobIDAllocationIDStringResourceManagerIdTuple5); |
Member
There was a problem hiding this comment.
I think you are using spaces for indentation here.
Contributor
Author
There was a problem hiding this comment.
Arrg, spaces.... How did they make it into this PR? Will get rid of them.
| final BlockingQueue<ResourceID> registrationQueue = new ArrayBlockingQueue<>(1); | ||
|
|
||
| testingResourceManagerGateway.setRegisterTaskExecutorFunction(stringResourceIDSlotReportIntegerHardwareDescriptionTuple5 -> { | ||
| registrationQueue.offer(stringResourceIDSlotReportIntegerHardwareDescriptionTuple5.f1); |
Member
There was a problem hiding this comment.
Same here: spaces are used for indentation.
Contributor
Author
There was a problem hiding this comment.
Arrg, the same.
tillrohrmann
added a commit
to tillrohrmann/flink
that referenced
this pull request
May 24, 2018
…Executor This commit fixes a race condition between the TaskExecutor and the ResourceManager. Before, it could happen that the ResourceManager sends requestSlots message before the TaskExecutor registration was completed. Due to this, the TaskExecutor did not have all information it needed to accept task submissions. The problem was that the TaskExecutor sent the SlotReport at registration time. Due to this, t he SlotManager could already assign these slots to pending slot requests. With this commit, the registration protocol changes such that the TaskExecutor first registers at the ResourceManager and only after completing this step, it will announce the available slots to the SlotManager. This closes apache#6067.
asfgit
pushed a commit
that referenced
this pull request
May 24, 2018
…Executor This commit fixes a race condition between the TaskExecutor and the ResourceManager. Before, it could happen that the ResourceManager sends requestSlots message before the TaskExecutor registration was completed. Due to this, the TaskExecutor did not have all information it needed to accept task submissions. The problem was that the TaskExecutor sent the SlotReport at registration time. Due to this, t he SlotManager could already assign these slots to pending slot requests. With this commit, the registration protocol changes such that the TaskExecutor first registers at the ResourceManager and only after completing this step, it will announce the available slots to the SlotManager. This closes #6067.
sampathBhat
pushed a commit
to sampathBhat/flink
that referenced
this pull request
Jul 26, 2018
…Executor This commit fixes a race condition between the TaskExecutor and the ResourceManager. Before, it could happen that the ResourceManager sends requestSlots message before the TaskExecutor registration was completed. Due to this, the TaskExecutor did not have all information it needed to accept task submissions. The problem was that the TaskExecutor sent the SlotReport at registration time. Due to this, t he SlotManager could already assign these slots to pending slot requests. With this commit, the registration protocol changes such that the TaskExecutor first registers at the ResourceManager and only after completing this step, it will announce the available slots to the SlotManager. This closes apache#6067.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What is the purpose of the change
This commit fixes a race condition between the TaskExecutor and the ResourceManager. Before,
it could happen that the ResourceManager sends requestSlots message before the TaskExecutor
registration was completed. Due to this, the TaskExecutor did not have all information it needed
to accept task submissions.
The problem was that the TaskExecutor sent the SlotReport at registration time. Due to this, t
he SlotManager could already assign these slots to pending slot requests. With this commit, the
registration protocol changes such that the TaskExecutor first registers at the ResourceManager
and only after completing this step, it will announce the available slots to the SlotManager.
cc @GJL
Brief change log
TaskExecutorResourceManagerregistration protocol to announce the available slots after the completion of the registrationTaskExecutor#requestSlotto only accept the call if there is an established connection to aResourceManagerVerifying this change
SlotManagerTest#testSlotRequestFailureTaskExecutorTest#testIgnoringSlotRequestsIfNotRegistered,testReconnectionAttemptIfExplicitlyDisconnected,testInitialSlotReportandtestInitialSlotReportFailureDoes this pull request potentially affect one of the following parts:
@Public(Evolving): (no)Documentation