sumitagrawl commented on code in PR #4988:
URL: https://github.com/apache/ozone/pull/4988#discussion_r1409485378
##########
hadoop-hdds/server-scm/src/main/java/org/apache/hadoop/hdds/scm/block/DeletedBlockLogImpl.java:
##########
@@ -437,15 +350,26 @@ public DatanodeDeletedBlockTransactions getTransactions(
throws IOException {
lock.lock();
try {
+ // Here we can clean up the Datanode timeout command that no longer
+ // reports heartbeats
+ getSCMDeletedBlockTransactionStatusManager().cleanAllTimeoutSCMCommand(
Review Comment:
@xichen01
"SCMBlockDeletingService timeout": this is period after which it will create
block set for further blocks for deletion. So this is running every 5 min.
scmCommandTimeoutMs: this is waiting time after command is sent in HB for
retry.
So normally DN should execute delete blocks within "5" minute.
With both having "5 min", and scmCommandTimeoutMs being initialized after
some time, it will avoid sending same command for one more retry, and
effectively get 10 min for blocks deletion.
This is more useful for aggressive deletion, keeping timeout to "1 min" for
"SCMBlockDeletingService timeout", so that new blocks are prepared quickly even
at background, old blocks deletion in progress at DN. And also resolving
duplicate blocks to that extent.
IMO, scmCommandTimeoutMs by default can be set to "20 min" as default value
in corresponding to "SCMBlockDeletingService timeout" which is default '5 min",
i.e. it will get 4-5 cycle waiting time for sending same blocks again.
OR can have relative configuration as scmCommandTimeoutMs = 4 *
"SCMBlockDeletingService timeout"
--
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]
