hbase-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HBASE-1573) Holes in master state change; updated startcode and server go into .META. but catalog scanner just got old values
Date Wed, 29 Jul 2009 05:36:14 GMT

    [ https://issues.apache.org/jira/browse/HBASE-1573?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12736458#action_12736458
] 

stack commented on HBASE-1573:
------------------------------

In this case:

{code}
hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-07-21:2009-07-21 22:49:59,701 INFO
org.apache.hadoop.hbase.master.ProcessRegionOpen: updating row nyt,1xNceOaojLkfi-31ZTQSK-==,1248216593259
in region .META.,,1 with  with startcode 1248174024124 and server XX.XX.45.124:20020
hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-07-21:2009-07-21 22:50:00,625 DEBUG
org.apache.hadoop.hbase.master.BaseScanner: Current assignment of nyt,1xNceOaojLkfi-31ZTQSK-==,1248216593259
is not valid;  Server '' unknown.
{code}

... there is < 1 second between update and the scanner.

Looking at code, I see that region stays in state of transition until after its been updated
by the OpenRegionProcess update of .META. with new location.

if (info.isOffline() ||
        (serverName != null && this.master.regionManager.regionIsInTransition(info.getRegionNameAsString()))
||
          (serverName != null && this.master.serverManager.isDead(serverName))) {
        return;
      }

In 1518 I added the serverName to the 2nd line above making it so we fall into the reassignment
code if serverName is null (Its null on split and close_region from shell)

> Holes in master state change; updated startcode and server go into .META. but catalog
scanner just got old values
> -----------------------------------------------------------------------------------------------------------------
>
>                 Key: HBASE-1573
>                 URL: https://issues.apache.org/jira/browse/HBASE-1573
>             Project: Hadoop HBase
>          Issue Type: Bug
>            Reporter: stack
>            Priority: Blocker
>             Fix For: 0.20.0
>
>         Attachments: 1573-v2.patch, 1573-v3.patch, 1573.patch
>
>
> Here is example of a scan that takes a while because 6k regions acting on stale data
resulting in double assignment of region:
> {code}
> hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-06-22:2009-06-22 10:56:06,220
INFO org.apache.hadoop.hbase.master.ServerManager: Received MSG_REPORT_OPEN: enwikibase,Cwj1sehVeEbnrUDR_j0xok==,1245604540748:
safeMode=false from XX.XX.45.121:20020
> hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-06-22:2009-06-22 10:56:06,220
INFO org.apache.hadoop.hbase.master.RegionServerOperation: enwikibase,Cwj1sehVeEbnrUDR_j0xok==,1245604540748
open on XX.XX.45.121:20020
> hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-06-22:2009-06-22 10:56:06,220
INFO org.apache.hadoop.hbase.master.RegionServerOperation: updating row enwikibase,Cwj1sehVeEbnrUDR_j0xok==,1245604540748
in region .META.,,1 with  with startcode !?B and server XX.XX.45.121:20020
> hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-06-22:2009-06-22 10:56:06,397
DEBUG org.apache.hadoop.hbase.master.BaseScanner: Current assignment of enwikibase,Cwj0sehVeEbnrUDR_j0xok==,1245604540748
is not valid;  Server 'XX.XX.44.95:20020' unknown.
> hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-06-22:2009-06-22 10:56:06,582
INFO org.apache.hadoop.hbase.master.RegionManager: Assigning region enwikibase,Cwj1sehVeEbnrUDR_j0xok==,1245604540748
to XX.XX.45.97:20020
> hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-06-22:2009-06-22 10:56:09,587
INFO org.apache.hadoop.hbase.master.ServerManager: Received MSG_REPORT_PROCESS_OPEN: enwikibase,Cwj1sehVeEbnrUDR_j0xok==,1245604540748:
safeMode=false from XX.XX.45.97:20020
> hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-06-22:2009-06-22 10:56:12,614
INFO org.apache.hadoop.hbase.master.ServerManager: Received MSG_REPORT_OPEN: enwikibase,Cwj1sehVeEbnrUDR_j0xok==,1245604540748:
safeMode=false from XX.XX.45.97:20020
> hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-06-22:2009-06-22 10:56:13,549
INFO org.apache.hadoop.hbase.master.RegionServerOperation: enwikibase,Cwj1sehVeEbnrUDR_j0xok==,1245604540748
open on XX.XX.45.97:20020
> hbase-Powerset-master-aa0-000-8.u.powerset.com.log.2009-06-22:2009-06-22 10:56:13,549
INFO org.apache.hadoop.hbase.master.RegionServerOperation: updating row enwikibase,Cwj1sehVeEbnrUDR_j0xok==,1245604540748
in region .META.,,1 with  with startcode !?? and server XX.XX.45.97:20020
> {code}
> We've just updated the server info in the master because of the region open message but
the scan sees old info in the .META. table though .META. was just updated.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message