cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Yuki Morishita (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (CASSANDRA-6024) sstableloader uploads full range sstable into some single range node(s) only
Date Mon, 23 Sep 2013 16:14:03 GMT

     [ https://issues.apache.org/jira/browse/CASSANDRA-6024?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Yuki Morishita updated CASSANDRA-6024:
--------------------------------------

    Attachment: 6042-test-result.txt

I tested with 1.2.9 and the latest 1.2 branch but I cannot reproduce the issue(logs from 1.2.9
attached).

* set up 3 nodes, RandomPartitioner, RF=2 ring using ccm
* generate SSTable containing 15 rows covering 5 rows for each node.
* load SSTable using 1.2.9 sstableloader to the cluster
* verified each nodes contains 10 rows (5 for the node itself, and 5 for replica).

there is only one change in sstableloader between 1.2.6 and 1.2.9, and it's CASSANDRA-5555.

                
> sstableloader uploads full range sstable into some single range node(s) only
> ----------------------------------------------------------------------------
>
>                 Key: CASSANDRA-6024
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-6024
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core, Tools
>         Environment: CentOS x64, Oracle JDK 7u17 x64, Cassandra 1.2.9
>            Reporter: Viktor Jevdokimov
>            Assignee: Yuki Morishita
>         Attachments: 6042-test-result.txt
>
>
> sstableloader from 1.2.9 uploads full-size sstables (not divided by token ranges) generated
with SSTableSimpleUnsortedWriter (no compression options) into Cassandra 1.2.9 nodes of some
single token range (on different clusters may be different range, but the same on the same
cluster). 
> sstableloader from Cassandra 1.2.6 uploads to Cassandra 1.2.9 nodes as expected, sstables
sent divided by token ranges.
> VNodes are not used, partitioner is RandomPartitioner.
> In both cases with 1.2.6 and 1.2.9 all sstables files generated the same, no binary difference.
> All target nodes returns correct token ranges.
> example 1:
> 3 nodes, 1 DC, RF=2, 2 out of 3 nodes receives 100% of sstable data instead of 66.6%,
rest node receives nothing.
> example 2:
> 32 nodes, 4 DCs, 8 nodes per DC, each DC RF=1 (total RF=4), same range node in each DC
receives 100% of sstable instead of 1/8 (cleanup will remove 7/8 of received data), other
nodes receives nothing.
> excerpts from sstableloader log when node does not receive a data:
>  INFO [main] 2013-09-13 01:03:41,336 Stream context metadata [], 2 sstables.  
>  INFO [main] 2013-09-13 01:03:41,338 Streaming to /1.2.3.4
> DEBUG [main] 2013-09-13 01:03:41,349 Files are   
> DEBUG [Streaming to /1.2.3.4:1] 2013-09-13 01:03:41,448 Received StreamReply StreamReply(sessionId=90e67680-1bff-11e3-b6f0-7df633a2f7a1,
file='', action=SESSION_FINISHED)  
> DEBUG [Streaming to /1.2.3.4:1] 2013-09-13 01:03:41,448 closing with status true  
>  INFO [Streaming to /1.2.3.4:1] 2013-09-13 01:03:41,449 Finished streaming session to
/1.2.3.4  
> DEBUG [Streaming to /1.2.3.4:1] 2013-09-13 01:03:41,450 Done streaming null  

--
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