hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Lars Hofhansl (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (HBASE-7545) [replication] Break out TestReplication into manageable classes
Date Wed, 30 Jan 2013 18:03:13 GMT

    [ https://issues.apache.org/jira/browse/HBASE-7545?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13566688#comment-13566688
] 

Lars Hofhansl edited comment on HBASE-7545 at 1/30/13 6:02 PM:
---------------------------------------------------------------

Increased NB_RETRIES to 15 for now (TIMEOUT was already 1000 in 0.94).
Looking at the log for TestReplicationSmallTests.loadTesting() I see that edits are mostly
replicated in batches of only 3 (presumably due to the really tiny log file size set for the
test). In that run a roundtrip for the these 3 edits took 100-200ms. So for 1000 edits that
can take a while (1000/3*100ms=33s).

The jenkins machines seem to be super slow and that also seems to be a recent phenomenon.
I see this in other tests too, where ZK changes take seconds to make it to the RSs or the
master via the asynchronous trackers when that should take only a few milliseconds.

                
      was (Author: lhofhansl):
    Increased NB_RETRIES to 15 for now (TIMEOUT was already 1000 in 0.94).
Looking at the log for TestReplicationSmallTests.loadTesting() I see that edits are mostly
replicated in batches of only 3 (presumably due to the really tiny log file size set for the
test). In that run a roundtrip for the these 3 edits took 100-200ms. So for 1000 edits that
can take a while (1000/3*100ms=33s).

The jenkins machine seem to be super slow and also seems to be a recent phenomenon.
I see this in other tests too, where ZK changes take seconds to make it to the RSs or the
master via the asynchronous trackers when should take only a few milliseconds.

                  
> [replication] Break out TestReplication into manageable classes
> ---------------------------------------------------------------
>
>                 Key: HBASE-7545
>                 URL: https://issues.apache.org/jira/browse/HBASE-7545
>             Project: HBase
>          Issue Type: Bug
>            Reporter: Jean-Daniel Cryans
>            Assignee: Jean-Daniel Cryans
>             Fix For: 0.96.0, 0.94.5
>
>         Attachments: 7545-0.94.txt, 7545-0.94-v1.txt, HBASE-7545.patch
>
>
> This has been discussed before but after trying to debug the last failure on Jenkins
where I saw the time go back and forth (if you don't care about your own sanity do checkout
https://builds.apache.org/job/HBase-TRUNK/3726/testReport/junit/org.apache.hadoop.hbase.replication/TestReplicationWithCompression/testDeleteTypes/)
I think it is time to break out TestReplication.
> The difficulty is that the setup for the 2 clusters is a lot of code I don't want to
duplicate. I'm thinking that we can keep {{setUpBeforeClass}} there and have the other classes
extend TestReplication (which should also change name). I'm thinking of the following new
classes:
>  - TestReplicationSmallTests, contains the easy methods that don't mess around too much.
>  - TestReplicationQueueFailover, contains one test of the same name
>  - TestReplicationDisableInactivePeer, contains one test of the same name
>  - Rename TestReplicationWithCompression TestReplicationQueueFailoverWithCompression
and make it extends TestReplicationQueueFailover.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message