[ 
https://issues.apache.org/jira/browse/HBASE-22190?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16832097#comment-16832097
 ] 

Duo Zhang commented on HBASE-22190:
-----------------------------------

I have to merge it first to see if it works by checking the flaky dashboard. We 
are facing different problems but the solution may have conflicts, this is a 
common case, and this issue is a blocker one so I want to be hurry. And it 
seems that the flaky dashboard is good so feel free to revert the patch here, I 
will get a review and reapply it again.

On your question:

1. Since now we will always list the root directory get the FileStatus for all 
the sub directories, I do not think it has much difference to check the 
'modificationTime' first or on the fly, as you always need to go over the 
FileStatus for all the sub directories to get the newest modificationTime. I've 
added comments about this in the patch.

2. If we introduce a new method which calls refreshCache then we should add 
synchronized to the method. Now I need to add synchronized for RefreshCacheTask 
then all the places where we call refreshCache are already synchronized so I 
removed the synchronized on the private refreshCache method.

On the patch for HBASE-21070, another concern is that does S3 support atomic 
rename? If not maybe there are more works to do...

> SnapshotFileCache may fail to load the correct snapshot file list when there 
> is an on-going snapshot operation
> --------------------------------------------------------------------------------------------------------------
>
>                 Key: HBASE-22190
>                 URL: https://issues.apache.org/jira/browse/HBASE-22190
>             Project: HBase
>          Issue Type: Bug
>          Components: snapshots
>            Reporter: Duo Zhang
>            Priority: Blocker
>             Fix For: 3.0.0, 2.2.0, 2.3.0, 2.1.5
>
>
> And it seems that it is not only a test issue, we do delete the files under 
> the archive directory, which is incorrect.
> Need to find out why, this maybe a serious bug.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to