[
https://issues.apache.org/jira/browse/CASSANDRA-2034?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13080223#comment-13080223
]
Jonathan Ellis commented on CASSANDRA-2034:
-------------------------------------------
Remember our algorithm looks something like this:
- if we know a replica is down, hint it on the capped executor. if this is
full, throw timeout.
- perform writes to believed-to-be-healthy replicas
- return ok to client once CL is achieved
- wait for remaining replicas in bg and hint on uncapped executor if necessary
The last part is what Jake and I were talking about when I said "worst case is
already "you need to be able to buffer up to rpc_timeout's worth of writes in
memory." (Referring to how the write stage on a replica buffers up mutations.)
This only comes into play during the delay between a replica becoming
unreachable, and the coordinator detecting that. Otherwise it goes to the
first, capped-write stage. So scheduling hints after 1h or w/e doesn't really
matter.
> Make Read Repair unnecessary when Hinted Handoff is enabled
> -----------------------------------------------------------
>
> Key: CASSANDRA-2034
> URL: https://issues.apache.org/jira/browse/CASSANDRA-2034
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Jonathan Ellis
> Assignee: Patricio Echague
> Fix For: 1.0
>
> Attachments: 2034-formatting.txt, CASSANDRA-2034-trunk-v2.patch,
> CASSANDRA-2034-trunk-v3.patch, CASSANDRA-2034-trunk-v4.patch,
> CASSANDRA-2034-trunk-v5.patch, CASSANDRA-2034-trunk-v6.patch,
> CASSANDRA-2034-trunk-v7.patch, CASSANDRA-2034-trunk.patch
>
> Original Estimate: 8h
> Remaining Estimate: 8h
>
> Currently, HH is purely an optimization -- if a machine goes down, enabling
> HH means RR/AES will have less work to do, but you can't disable RR entirely
> in most situations since HH doesn't kick in until the FailureDetector does.
> Let's add a scheduled task to the mutate path, such that we return to the
> client normally after ConsistencyLevel is achieved, but after RpcTimeout we
> check the responseHandler write acks and write local hints for any missing
> targets.
> This would making disabling RR when HH is enabled a much more reasonable
> option, which has a huge impact on read throughput.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
