Timeouts in Riak Search

Ryan Zezeski rzezeski at basho.com
Mon Nov 14 09:19:32 EST 2011

Hi Spike,

I think the reason the query is timing out is because of some other failure
that is occurring.  In your error gist there is a supervisor report @
19:16:07 about a riak_pipe_vnode_worker dying.  There isn't much in the way
of explanation, just "fitting_died."  Unfortunately, it looks like pipe
just swallows the reason the fitting died [1].  The next supervisor report
(for riak_core_vode_sup) actually occurred _before_ the previous one
according to the timestamp (@ 19:14:27).  It looks as if the ETS table
riak_core_node_watcher doesn't exist which would be a sign that it's owning
process crashed at some point.  The question is what caused the
riak_core_vnode_watcher to crash?  Do you have any other errors in your
crash/error.log that might help shed some light on this?

In summary, I believe the search timeout is a symptom of something else.



On Wed, Nov 9, 2011 at 2:46 PM, Spike Gronim <spike at wavii.com> wrote:

> Hello,
> My Riak search cluster is timing out very often. I am indexing text
> content extracted from web pages containing news articles. My articles
> range in size from a few KB to tens of KB.  I have put about 4.4 million
> articles into Riak for an average article size of 15 KB. The keys are MD5
> ASCII hex hashes and the values are JSON. When I set this system up I
> loaded it with 1GB or so of data and played with the search system.
> Everything was kosher, it responded quickly and the search relevance was
> fine. Now that I've imported 100x as much data I am getting timeouts. For
> example the query "steve jobs died" times out. When I put in extremely
> specific conjunctive queries like "+steve +jobs +died +cupertino +apple" I
> get no results but it runs quickly. While the system is running a query
> that will time out I see the coordinator Riak node consuming between one
> and two cores worth of CPU.
> How can I configure Riak to stop timing out searches? I am open to
> changing my schema and query pattern if that's what I need to do.
> app.config - https://gist.github.com/1352608
> schema - https://gist.github.com/1352616
> selected errors - https://gist.github.com/1c0976ced0f05ef0d5d6
> Nodes in the cluster: 4
> Hardware: EC2 m1.large with two disks in a RAID-0 on /mnt
> Operating system: Linux ip-XXXX 2.6.38-11-virtual #50-Ubuntu SMP Mon Sep
> 12 21:51:23 UTC 2011 x86_64 x86_64 x86_64 GNU/Linux
> Disk space consumed:
> 66G  /mnt/riak/leveldb
> 36G  /mnt/riak/merge_index
> Disk space available: 800G
> Spike Gronim
> spike at wavii.com
> _______________________________________________
> riak-users mailing list
> riak-users at lists.basho.com
> http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.basho.com/pipermail/riak-users_lists.basho.com/attachments/20111114/abb52b25/attachment.html>

More information about the riak-users mailing list