[ 
https://issues.apache.org/jira/browse/CASSANDRA-8752?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14310479#comment-14310479
 ] 

Kevin Ye edited comment on CASSANDRA-8752 at 2/7/15 2:36 AM:
-------------------------------------------------------------

Hi Aleksey. 

Switching to another version is not an easy decision for my project. So we hope 
to fix the issuse in our 2.1.2 code. 
It will be very helpful if you can tell the issuse number you mentioned.

At first we need to reproduce this problem. One possible cause(not sure) may 
like this:
1. +100 to the counter, counter become (1,+100)
2. Shutdown and reboot the node, which will read the data file asynchronously.
3. Before data file is load, rpc service is started and -33 to the counter 
comes.
4. Since the data file is not loaded, reading of old counter(1,+100) returns 
nothing. Then an (1, -33) is written.
5. (1,+100) and (1, -33) conflict occurs when compact happens.

What's your opinion about it?


was (Author: simsky):
Hi Aleksey. 

Switching to other version is not an easy decision for my project. So we hope 
to fix the issuse in our 2.1.2 code. 
It will be very helpful if you can tell the issuse number you mentioned.

At first we need to reproduce this problem. One possible cause(not sure) may 
like this:
1. +100 to the counter, counter become (1,+100)
2. Shutdown and reboot the node, which will read the data file asynchronously.
3. Before data file is load, rpc service is started and -33 to the counter 
comes.
4. Since the data file is not loaded, reading of old counter(1,+100) returns 
nothing. Then an (1, -33) is written.
5. (1,+100) and (1, -33) conflict occurs when compact happens.

What's your opinion about it?

> invalid counter shard detected in Version 2.1.2
> -----------------------------------------------
>
>                 Key: CASSANDRA-8752
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-8752
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>         Environment: SUSE11 SP1, Cassandra 2.1.2, java version "1.7.0_55".
> 4 node cluster, vnode = 1, replication = 2
>            Reporter: Kevin Ye
>            Assignee: Aleksey Yeschenko
>
> I was doing counter test (first +100 several times, then -33) on a 4 nodes 
> cluster while below log appear at 2 nodes.There is no concurrent access to 
> same counter.
> WARN  [CompactionExecutor:757] 2015-02-02 13:02:33,375 
> CounterContext.java:431 - invalid global counter shard detected; 
> (9cca9262-934a-4275-963b-66802471b0c2, 1, -33) and 
> (9cca9262-934a-4275-963b-66802471b0c2, 1, 100) differ only in count; will 
> pick highest to self-heal on compaction
> Anyone has encounter this problem? I thought Cassandra 2.1.2 had solved this 
> counter problem, but it appeared.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to