LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1700275531
I submit the pr many times and [hadoop-yetus] always report the same file
hadoop-hdfs-project/hadoop-hdfs/hs_err_pid52701.log. Does the
hs_err_pid52701.log not delete after new submit？
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1683622535
> If true, is it enough to make IBR and FBR asynchronously together and use
only one thread? Thanks.
I tried and I've given up. The reason are as follows:
IBR and FBR are not on the
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1663510246
Revert the previous commit because of problems. Can we review the ibr and
fbr in different thread first？
--
This is an automated message from the Apache Git Service.
To respond to the
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1657869992
Apache Yetus(jenkins) error: mvninstallCould not transfer artifact
org.codehaus.mojo:extra-enforcer-rules:pom:1.5.1 from/to central
(https://repo.maven.apache.org/maven2): Transfer
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1655422522
> Suggest to submit at one PR (rather than different PR) , thus we could
focus and discuss at the same place.
OK , done. Thanks
--
This is an automated message from the
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1655248903
> If true, is it enough to make IBR and FBR asynchronously together and use
only one thread? Thanks.
Please review https://github.com/apache/hadoop/pull/5898 , make IBR and FBR
into
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1652883289
> If true, is it enough to make IBR and FBR asynchronously together and use
only one thread?
Variable fullBlockReportLeaseId updates from heartbeat and used by fbr and
then update to
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1652806518
@Hexiaoqiao ， well, I user original sendIBRLock to make right order for ibr
and fbr. The FBR will execute ibr first, so the sendIBRLock will be
successful for it. Please review for it,
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1651805316
Yes，I know it. It is a problem with mis-order.
a. heartbeat got a delete cmd on replication named r
b. generate fbr
c. datanode delete repliction r and send ibr
d. send fbr
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1651437873
And maybe it not a problem. The FBR will only process the reported block
list，the replication r will not discard. FBR
finally execute BlockManager.processReportedBlock method ,
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1651382745
@Hexiaoqiao ，if the order of ibr and fbr is must be guaranteed at datanode
side， [HDFS-16016](https://issues.apache.org/jira/browse/HDFS-16016) also can
not guarantee it.
Thus if it
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1651228974
> Good improvement. Not think carefully but my first feeling, IBR - FBR
mis-order could trigger some issues, such as miss some block report? One case,
a. generate full block report, b.
12 matches
Mail list logo
