Martin,
The error you are seeing is caused by the death of some of map phase worker
processes. It's hard to tell exactly why they would be dying from the limited
output, but I suspect since you mention that it happens only under heavy load
that it's hitting the map reduce timeout or perhaps (but more unlikely) even
exceeding the erlang process limit. Could you check for any other error output
in the riak log directory that might be relevant? You could also just try
increasing the timeout used in your mapreduce queries and see if that
alleviates the problem.
FWIW, I would expect you to have a better time with Riak 1.0 using the
riak_pipe mapreduce. It was designed to be more robust for long running jobs
and in the case when the Riak cluster is heavily loaded. Cheers.
Kelly
On Sep 15, 2011, at 9:49 AM, Martin Woods wrote:
> Hi all
>
> When running Riak recently on a development platform (Ubuntu 9.10 Server
> 64-bit with Riak 0.14.2) the Riak error logs reported the following
> termination:
>
>
> ** Reason for termination =
> ** {error,
> {phase_error,
> {error,
> {no_candidate_nodes,exhausted_preflist,
> [{riak_kv_mapred_planner,claim_keys,3},
> {riak_kv_map_phase,handle_info,2},
> {luke_phase,handle_info,3},
> {gen_fsm,handle_msg,7},
> {proc_lib,init_p_do_apply,3}],
>
>
> At the time we were heavily loading Riak using the riakc client with multiple
> Erlang mapreduce requests, each involving key filters. Could that load have
> caused this error? We can see in the code where this is raised but don't know
> enough about Riak under the hood to work out what the underlying cause is.
>
>
>
> Any help would be much appreciated!
>
>
>
> Kind regards,
>
> Martin.
>
>
>
> _______________________________________________
> riak-users mailing list
> [email protected]
> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
_______________________________________________
riak-users mailing list
[email protected]
http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
