hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jean-Daniel Cryans (JIRA)" <j...@apache.org>
Subject [jira] Commented: (HBASE-2421) Put hangs for 10 retries on failed region servers
Date Fri, 07 May 2010 01:24:48 GMT

    [ https://issues.apache.org/jira/browse/HBASE-2421?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12864990#action_12864990
] 

Jean-Daniel Cryans commented on HBASE-2421:
-------------------------------------------

I'm looking at applying this to trunk and I see that this wasn't applied to branch:

{code}
--- src/java/org/apache/hadoop/hbase/LocalHBaseCluster.java
+++ src/java/org/apache/hadoop/hbase/LocalHBaseCluster.java
@@ -116,6 +116,8 @@ public class LocalHBaseCluster implements HConstants {
       try {
         server = regionServerClass.getConstructor(HBaseConfiguration.class).
           newInstance(conf);
+        // Servers should not shut down HDFS, since they share an instance.
+        server.shutdownHDFS.set(false);
{code}

Was it intentional?

> Put hangs for 10 retries on failed region servers
> -------------------------------------------------
>
>                 Key: HBASE-2421
>                 URL: https://issues.apache.org/jira/browse/HBASE-2421
>             Project: Hadoop HBase
>          Issue Type: Bug
>            Reporter: Jean-Daniel Cryans
>            Assignee: ryan rawson
>            Priority: Critical
>             Fix For: 0.21.0
>
>         Attachments: HBASE-2421-2.txt, hbase-2421.txt, HBASE-2421.txt
>
>
> Since MultiPut got in, instead of calling getRegionLocationForRowWithRetries we now call
getRegionServerWithRetries to send an array list of Puts. The problem is that if the region
server failed, we'll still retry the 10 times in a backoff fashion even tho we get connections
refused. This is also true for a single put since it's the same code path.
> Marking as critical since it almost disables our responsiveness to machine failures in
certain cases where we are already sending a batch of edits when the server fails. Assigning
to Ryan since he's been there recently.

-- 
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