siddhantsangwan opened a new pull request, #4643:
URL: https://github.com/apache/ozone/pull/4643

   ## What changes were proposed in this pull request?
   
   Problem: Container Balancer gets a reference to a Container's 
`ContainerInfo` from `ContainerManager` in several places throughout the 
balancer package. We need to investigate if there are possible race conditions 
with other parts of SCM changing the replicas of a container etc.
   
   In the investigation it was found that in most places, the `ContainerInfo` 
object's `usedBytes` field is being read through the getter 
`ContainerInfo#getUsedBytes`. `usedBytes` is written in 
`AbstractContainerReportHandler#updateContainerUsedAndKeys` when a container 
report is processed. Since `usedBytes` is a long primitive type and 
reads/writes to a long are not atomic, there's a race condition between the 
balancer thread reading this value and the report handling thread writing to 
it. This PR proposes making `usedBytes` volatile because reads/writes to a 
volatile long are atomic - 
https://docs.oracle.com/javase/specs/jls/se8/html/jls-17.html#jls-17.7. This 
should solve the race condition. 
   
   `AbstractFindTargetGreedy` passes a `ContainerInfo` to 
`PlacementPolicyValidateProxy#validateContainerPlacement`. This method reads 
the replication type and replication factor from `ContainerInfo`. These values 
should stay constant, so synchronization is not required here.
   
   I don't think locking on `ContainerInfo` is required in the balancer package 
except in `MoveManager` where it's already present.
   
   ## What is the link to the Apache JIRA
   
   https://issues.apache.org/jira/browse/HDDS-8347
   
   ## How was this patch tested?
   
   Ran existing tests.


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to