cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Pavel Yaskevich (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-6373) describe_ring hangs with hsha thrift server
Date Thu, 19 Dec 2013 22:19:09 GMT


Pavel Yaskevich commented on CASSANDRA-6373:

Is the same happening with vnodes turned off? I just don't get why if you send any other command
it works fine but if you send describe_ring it fails, it shouldn't be any different from the
server perspective... Can you also try increasing min number of rpc threads from 2 to 4, wonder
if it would make any difference.

> describe_ring hangs with hsha thrift server
> -------------------------------------------
>                 Key: CASSANDRA-6373
>                 URL:
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Nick Bailey
>            Assignee: Pavel Yaskevich
>             Fix For: 2.0.4
>         Attachments: describe_ring_failure.patch, jstack.txt, jstack2.txt
> There is a strange bug with the thrift hsha server in 2.0 (we switched to lmax disruptor
> The bug is that the first call to describe_ring from one connection will hang indefinitely
when the client is not connecting from localhost (or it at least looks like the client is
not on the same host). Additionally the cluster must be using vnodes. When connecting from
localhost the first call will work as expected. And in either case subsequent calls from the
same connection will work as expected. According to git bisect the bad commit is the switch
to the lmax disruptor server:
> I've attached the patch I used to reproduce the error in the unit tests. The command
to reproduce is: 
> {noformat}
> PYTHONPATH=test nosetests --tests=system.test_thrift_server:TestMutations.test_describe_ring
> {noformat}
> I reproduced on ec2 and a single machine by having the server bind to the private ip
on ec2 and the client connect to the public ip (so it appears as if the client is non local).
I've also reproduced with two different vms though.

This message was sent by Atlassian JIRA

View raw message