[
https://issues.apache.org/jira/browse/HBASE-21863?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16765680#comment-16765680
]
stack commented on HBASE-21863:
-------------------------------
bq. ...but for assign/unassign we have to support the old protocol so this
haven't been done yet.
Are you talking pre-2.2 [~Apache9]?
Are you suggesting that on region server report, for each region, we'd add pid
that opened the region?
Thanks.
[~sershe]
bq. ....we will just, very infrequently, error out the call by mistake that
could have succeeded and retry it elsewhere.
Possibly double-assign?
Are you trying to put a procedure timeout inside the master rpc timeout
envelope?
IIRC, its still the case that open could edit hfiles and mark WALs. We should
fix this so no damage allowed until Master gives its assent.
On the patch, its getting nicer.
Should
public class RegionsStateInvalidException extends IOException {
extend YouAreDeadException if we are going to kill again?
This stuff right regards -1 and 0 in below?
71 // -1 means don't set the deadline
72 public static final String MASTER_RPC_DEADLINE_SAFETY_MS =
73 "hbase.master.rpc.deadline.safety.ms";
74 private static final int DEFAULT_MASTER_RPC_DEADLINE_SAFETY_MS = 0;
Have to head... will look at rest later.
> narrow down the double-assignment race window
> ---------------------------------------------
>
> Key: HBASE-21863
> URL: https://issues.apache.org/jira/browse/HBASE-21863
> Project: HBase
> Issue Type: Bug
> Reporter: Sergey Shelukhin
> Assignee: Sergey Shelukhin
> Priority: Major
> Attachments: HBASE-21863.01.patch, HBASE-21863.patch
>
>
> See HBASE-21862.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
