liubingxing opened a new pull request #4033:
URL: https://github.com/apache/hadoop/pull/4033


   When SPSPathIdProcessor thread call getNextSPSPath(), it get the pathId from 
namenode and namenode will also remove this pathId from pathsToBeTraveresed 
queue.
   ```java
   public Long getNextPathId() {
     synchronized (pathsToBeTraveresed) {
       return pathsToBeTraveresed.poll();
     }
   } 
   ```
   If SPS process restart, this path will not continue the move operation until 
namenode restart.
   
   So we want to provide a way for the SPS to continue performing the move 
operation after SPS restart.
   
   First solution: 
   
   1) When SPSPathIdProcessor thread call getNextSPSPath(), namenode return 
pathId and then move this pathId to a pathsBeingTraveresed queue;
   
   2) After SPS finish a path movement operation, it call a rpc to namenode to 
remove this pathId from pathsBeingTraveresed queue;
   
   3) If SPS restart, SPSPathIdProcessor thread should call a rpc to namenode 
to get all pathId from pathsBeingTraveresed queue;
   
   Second solution:
   
   We added timeout detection in the application layer, if a path does not 
complete the movement within the specified time, we can re-satisfy this path 
even though it has "hdfs.sps" xattr already.
   
   We choose the second solution because the first solution will add more rpc 
operation and may affect namenode performance.


-- 
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