Fixed race condition on multi-topic consumer #11764
Merged
Conversation
This file contains 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
merlimat
added
type/bug
merlimat
requested review from
sijie,
rdhabalia,
jerrypeng,
hangc0276,
315157973 and
codelipenghui
315157973
approved these changes
Aug 25, 2021
There was a problem hiding this comment.
LGTM
pendingReceives also seems to have a race condition, two threads will operate pendingReceives :
receiveMessageFromConsumer is executed by client.getInternalExecutorService().execute
internalReceiveAsync is executed by the user's thread
It is better to put them all in pinnedExecutor, I am working on it. And it can solve the above scenario at the same time
I think the usage on
My concern there is that if you have a multi-topic consumer pulling from many topics, we shouldn't be bottlenecked by 1 single thread. |
hangc0276
pushed a commit
that referenced
this pull request
Aug 25, 2021
### Motivation Under certain conditions applications using the multi-topic consumers might get the consumption stalled: The conditions to reproduce the issue are: * Consumer is subscribed to multiple topics, but only 1 topic has traffic * Messages are published in batches (no repro if no batches) * Receiver queue size == 1 (or small, in order to exercise race condition) The problem is that there is race condition between 2 threads when we're deciding to put one of the individual consumers in "paused" state, when the shared queue is full. What happens is that, just after we checked the conditions and we decide to mark the consumer as paused, the application has emptied the shared queue completely. From that point on, there is no re-attempt to check whether we need to unblock that consumer. ### Modification Instead of introducing a sync block (contended by many consumers), we just double check the state of the shared queue after marking the consumer as "paused". If the other thread has emptied the queue in the meantime, we'll be guaranteed to unblock the consumer. (cherry picked from commit f1d66d1)
codelipenghui
pushed a commit
that referenced
this pull request
Dec 11, 2021
### Motivation Under certain conditions applications using the multi-topic consumers might get the consumption stalled: The conditions to reproduce the issue are: * Consumer is subscribed to multiple topics, but only 1 topic has traffic * Messages are published in batches (no repro if no batches) * Receiver queue size == 1 (or small, in order to exercise race condition) The problem is that there is race condition between 2 threads when we're deciding to put one of the individual consumers in "paused" state, when the shared queue is full. What happens is that, just after we checked the conditions and we decide to mark the consumer as paused, the application has emptied the shared queue completely. From that point on, there is no re-attempt to check whether we need to unblock that consumer. ### Modification Instead of introducing a sync block (contended by many consumers), we just double check the state of the shared queue after marking the consumer as "paused". If the other thread has emptied the queue in the meantime, we'll be guaranteed to unblock the consumer. (cherry picked from commit f1d66d1)
bharanic-dev
pushed a commit
to bharanic-dev/pulsar
that referenced
this pull request
Mar 18, 2022
### Motivation Under certain conditions applications using the multi-topic consumers might get the consumption stalled: The conditions to reproduce the issue are: * Consumer is subscribed to multiple topics, but only 1 topic has traffic * Messages are published in batches (no repro if no batches) * Receiver queue size == 1 (or small, in order to exercise race condition) The problem is that there is race condition between 2 threads when we're deciding to put one of the individual consumers in "paused" state, when the shared queue is full. What happens is that, just after we checked the conditions and we decide to mark the consumer as paused, the application has emptied the shared queue completely. From that point on, there is no re-attempt to check whether we need to unblock that consumer. ### Modification Instead of introducing a sync block (contended by many consumers), we just double check the state of the shared queue after marking the consumer as "paused". If the other thread has emptied the queue in the meantime, we'll be guaranteed to unblock the consumer.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Motivation
Under certain conditions applications using the multi-topic consumers might get the consumption stalled:
The conditions to reproduce the issue are:
The problem is that there is race condition between 2 threads when we're deciding to put one of the individual consumers in "paused" state, when the shared queue is full.
What happens is that, just after we checked the conditions and we decide to mark the consumer as paused, the application has emptied the shared queue completely. From that point on, there is no re-attempt to check whether we need to unblock that consumer.
Modification
Instead of introducing a sync block (contended by many consumers), we just double check the state of the shared queue after marking the consumer as "paused". If the other thread has emptied the queue in the meantime, we'll be guaranteed to unblock the consumer.