Hello Stan

 Which version of Cassandra are you using ? There are some known issues of
streaming failure that prevent a node from finishing joining

 Regards

On Wed, Nov 19, 2014 at 3:57 PM, Stan Lemon <sle...@salesforce.com> wrote:

> Hello,
> I'm working on a two data center cluster with 12 nodes in each data
> center. I recently wanted to add a thirteenth node to one of the data
> centers to try and validate some load improvements to our hardware
> configuration. I added the node following DataStax directions (
> http://www.datastax.com/documentation/cassandra/2.0/cassandra/operations/ops_add_node_to_cluster_t.html)
> and the node appeared to bootstrap correctly and start joining.
>
> I monitored the load and watched it increase, periodically checking iotop
> to make sure there was still a pulse. Eventually the load topped out at
> roughly 85% of the average of the other nodes, iotop showed lots of
> activity.  After a few hours iotop stopped showing activity and the node's
> load had gone down a small amount, ~50-100mb.  Average load on the other
> nodes is about ~550gb
>
> The first time I tried this I let the process run through the weekend,
> periodically checking on it.  Something happened Monday morning which
> caused Cassandra to die, so I restarted the process. The load immediately
> began growing, eventually doubling that 85% marker and settling in around
> ~935gb, way more than any other node. When it reached this point it did the
> same thing though, basically stalled out.
>
> The whole time nodetool status just showed "UJ".
>
> Finally I aborted and cleared the node's data directory and started over,
> but again experienced the same stall out at the 85% mark. The node tool no
> time at all to get to that point, it was only a few hours. It's not been
> sitting at 85% for roughly 20 hours and iotop shows no activity.
>
> I am wondering a few things...
> 1. What's going on?
> 2. How do I get more information about what is happening with the join
> process?
> 3. Has anyone seen this before?
>
> Thanks for your help,
> Stan
>
>

Reply via email to