MMirelli commented on issue #3751:
URL: https://github.com/apache/bookkeeper/issues/3751#issuecomment-1423882617

   Unfortunately, I cannot share full memory dumps / logs generated by our 
internal testing. 
   
   Looking at the locking section of the broker JFR file I generated, the main 
bottleneck seems to be in the `BookKeeperClientScheduler-OrderedScheduler` 
thread having the maximum block time of 209microseconds. That is executing 
`BookieClientImpl.monitorPendingOperations()`. 
   
   The method which is called more often is instead `.__lock_text_start(k)` (53 
calls in 1 minute), which is likely to be the way a process acquires locks in 
Linux. That is called by 
`PulsarRegistrationClient.updatedBookies(Notification)` in a `pulsar-io` thread.
   
   The bookie which, from the broker's logs appears to have problems, is 
instead pretty much idling. The highest blocking time other than the one to 
create Object is 36 microseconds and it's caused by 
`org.apache.zookeeper.ClientCnxnSocketNIO`. Also there there are few calls of 
`.__lock_text_start(k)`, but only 15 calls in 1 minute. These are all related 
to communication with ZK (8 using directly 
`org.apache.zookeeper.ClientCnxnSocketNIO`).
   


-- 
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]

Reply via email to