jianghuazhu commented on PR #3806:
URL: https://github.com/apache/hadoop/pull/3806#issuecomment-1136659630

   Thanks @ZanderXu  for following.
   Here are some explanations:
   1. The main job of FsDatasetAsyncDiskService is to delete the replica files 
synchronously or asynchronously. The copy files to be deleted here are all 
files on the local DataNode, and the number is limited. Although the thread 
pool uses an unbounded queue, it will not be stored all the time, because it 
will always be consumed. And these copies have been loaded into memory when the 
DataNode is working, so the probability of OOM here is very low.
   2. If the copy is deleted asynchronously, the thread pool work will be 
started. Before this, each disk will correspond to a thread pool, and the 
thread pool will have at most 4 fixed threads to work, and this condition is 
fixed. In our cluster, DataNodes have different numbers of disks, 12 disks, 36 
disks, and 60 disks will exist. Take DataNode with 36 disks or 60 disks as an 
example, then during peak hours, DataNode needs to start a lot of thread work. 
Adjusting the number of threads flexibly will reduce the workload of the 
DataNode.


-- 
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