We're using a number of Riak 1.4 nodes in a cluster. In front of it is our java-based "Storage-Backend" (our own version of Riak-CS, custom built to our requirements), which connects to localhost Riak. The Riak nodes are linked over IPSec-links. All of this is running on cheap commodity hardware, and thanks to the nature of Riak we can deal with node failures. We're learning to embrace the concept of enventual consistency and are using conflict resolvers to deal with siblings.
Recently we upgraded from 1.4.10 to 1.4.12, and we're having the impression that this version performs worse. The rate of sibling creation feels much higher (but I don't have solid data to back this up). We've also noticed a few occasions where our conflict resolver was unable to do its job. Manual investigation showed that an object had siblings with *identical vtags*, but only sometimes!! From memory, this is what happened: - Storage-Backend throws Unresolved Conflict Exception with object "abc123" of bucket "chunks" - Repeatedly issuing 'curl --head http://localhost:8087/buckets/chunks/keys/abc123' returns half the time HTTP 200 and half the time HTTP 300 - When getting the HTTP 300 response, the output of 'curl http://localhost:8087/buckets/chunks/keys/abc123' was something like this: Siblings: 4BaLvqdDSd8qYicDsn3DS4 4BaLvqdDSd8qYicDsn3DS4 The solution was to manually retrieve the object (specifying the vtag) and uploading it again (specifying the Vclock). My questions: - Is anyone else experiencing similar issues, maybe even with 1.4.12? - Is it possible (or even worthwhile) to downgrade to 1.4.10? Thanks a bunch! Cheers, Henning _______________________________________________ riak-users mailing list [email protected] http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
