You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@pulsar.apache.org by GitBox <gi...@apache.org> on 2021/05/25 10:11:41 UTC
[GitHub] [pulsar] massakam opened a new pull request #10696: [broker] Fix issue where StackOverflowError occurs when trying to redeliver a large number of already acked messages
massakam opened a new pull request #10696:
URL: https://github.com/apache/pulsar/pull/10696
### Motivation
The other day, some of our broker servers got the following StackOverflowError:
```
13:44:17.410 [pulsar-io-21-6] WARN o.a.pulsar.broker.service.ServerCnx - [/xxx.xxx.xxx.xxx:58438] Got exception StackOverflowError : null
java.lang.StackOverflowError: null
at java.util.stream.ReferencePipeline$2$1.accept(ReferencePipeline.java:174)
at java.util.TreeMap$KeySpliterator.forEachRemaining(TreeMap.java:2746)
at java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:482)
at java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:472)
at java.util.stream.ForEachOps$ForEachOp.evaluateSequential(ForEachOps.java:150)
at java.util.stream.ForEachOps$ForEachOp$OfRef.evaluateSequential(ForEachOps.java:173)
at java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:234)
at java.util.stream.ReferencePipeline.forEach(ReferencePipeline.java:485)
at org.apache.bookkeeper.mledger.impl.ManagedCursorImpl.asyncReplayEntries(ManagedCursorImpl.java:1086)
at org.apache.bookkeeper.mledger.impl.ManagedCursorImpl.asyncReplayEntries(ManagedCursorImpl.java:1066)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.asyncReplayEntries(PersistentDispatcherMultipleConsumers.java:341)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:309)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
at org.apache.pulsar.broker.service.persistent.PersistentDispatcherMultipleConsumers.readMoreEntries(PersistentDispatcherMultipleConsumers.java:318)
```
This phenomenon can be reproduced by the following procedure:
1. Store a large number of messages in the backlog of a topic
2. Connect some Shared consumers to the topic. These consumers receive messages but do not acknowledge at all
3. Run skip-all to remove all messages from the backlog
4. Add another consumer whose receiver queue size is small
5. Close all the consumers added in step 2
6. StackOverflowError occurs on the broker
If broker receives a large number of redelivery requests for messages that have already been deleted, `PersistentDispatcherMultipleConsumers#readMoreEntries()` is called recursively many times. As a result, we get a StackOverflowError.
https://github.com/apache/pulsar/blob/v2.4.2/pulsar-broker/src/main/java/org/apache/pulsar/broker/service/persistent/PersistentDispatcherMultipleConsumers.java#L310-L329
### Modifications
- Avoid recursive calls of `readMoreEntries()` on the same thread
- If the dispatcher receives redelivery requests for messages whose positions are earlier than the mark delete position, it should not add them to `messagesToRedeliver`
--
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:
users@infra.apache.org
[GitHub] [pulsar] eolivelli commented on a change in pull request #10696: [broker] Fix issue where StackOverflowError occurs when trying to redeliver a large number of already acked messages
Posted by GitBox <gi...@apache.org>.
eolivelli commented on a change in pull request #10696:
URL: https://github.com/apache/pulsar/pull/10696#discussion_r638723521
##########
File path: pulsar-broker/src/main/java/org/apache/pulsar/broker/service/persistent/PersistentDispatcherMultipleConsumers.java
##########
@@ -248,7 +249,9 @@ public synchronized void readMoreEntries() {
// next entries as readCompletedEntries-callback was never called
if ((messagesToReplayNow.size() - deletedMessages.size()) == 0) {
havePendingReplayRead = false;
- readMoreEntries();
+ // We should not call readMoreEntries() recursively in the same thread
+ // as there is a risk of StackOverflowError
+ topic.getBrokerService().executor().execute(() -> readMoreEntries());
Review comment:
I have two high level questions about this change:
1) is it possible/does it make sense to not schedule the execution of `readMoreEntries` if there is already another pending request of executing `readModeEntries` ?
2) is this changing the semantics of how this method works ?
--
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:
users@infra.apache.org
[GitHub] [pulsar] massakam commented on a change in pull request #10696: [broker] Fix issue where StackOverflowError occurs when trying to redeliver a large number of already acked messages
Posted by GitBox <gi...@apache.org>.
massakam commented on a change in pull request #10696:
URL: https://github.com/apache/pulsar/pull/10696#discussion_r638971507
##########
File path: pulsar-broker/src/main/java/org/apache/pulsar/broker/service/persistent/PersistentDispatcherMultipleConsumers.java
##########
@@ -248,7 +249,9 @@ public synchronized void readMoreEntries() {
// next entries as readCompletedEntries-callback was never called
if ((messagesToReplayNow.size() - deletedMessages.size()) == 0) {
havePendingReplayRead = false;
- readMoreEntries();
+ // We should not call readMoreEntries() recursively in the same thread
+ // as there is a risk of StackOverflowError
+ topic.getBrokerService().executor().execute(() -> readMoreEntries());
Review comment:
@eolivelli Those are difficult questions...
Even now, if the message dispatch rate reaches the threshold, another thread can execute `readMoreEntries()` recursively after a certain period of time, so it seems to me that this change doesn't break the behavior of the dispatcher. However, there is no clear evidence for it.
If anyone has any knowledge, I would like to hear opinions.
https://github.com/apache/pulsar/blob/a6aed551026825ef2de6b1ac5916d17daf1af5c3/pulsar-broker/src/main/java/org/apache/pulsar/broker/service/persistent/PersistentDispatcherMultipleConsumers.java#L225
https://github.com/apache/pulsar/blob/a6aed551026825ef2de6b1ac5916d17daf1af5c3/pulsar-broker/src/main/java/org/apache/pulsar/broker/service/persistent/PersistentDispatcherMultipleConsumers.java#L275-L308
--
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:
users@infra.apache.org
[GitHub] [pulsar] sijie merged pull request #10696: [broker] Fix issue where StackOverflowError occurs when trying to redeliver a large number of already acked messages
Posted by GitBox <gi...@apache.org>.
sijie merged pull request #10696:
URL: https://github.com/apache/pulsar/pull/10696
--
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:
users@infra.apache.org