cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ryan McGuire (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-6731) Requests unnecessarily redirected with CL=ONE
Date Wed, 19 Feb 2014 15:44:20 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-6731?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13905570#comment-13905570
] 

Ryan McGuire commented on CASSANDRA-6731:
-----------------------------------------

I've attached a script (ticket_6731.py) that runs a few tests:

* local_read_speed_test_without_shutdown
* local_read_speed_test_with_shutdown
* test_local_read_trace

The first two perform a small speed run comparing the case of one node vs three nodes. The
single node test was a few seconds faster (average 37.8s for 1 node vs 42.5s for 3 nodes.)

test_local_read_trace does things more methodically, and runs a trace on each query to determine
what percentage of queries actually contact nodes other than the coordinator. Of a sample
set of 1000 queries, 10.7% of them contacted other nodes. You can see those traces in 6731.traces.txt.

So, I do concur that a non-negligible amount of the queries do contact other nodes, but it
doesn't appear to have much of a performance impact. This test was done on a local ccm cluster,
so it may warrant testing on a physical cluster with different data loads.

[~arisp] can you post your schema and your approximate data size?

> Requests unnecessarily redirected with CL=ONE
> ---------------------------------------------
>
>                 Key: CASSANDRA-6731
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-6731
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>         Environment: Linux
>            Reporter: Aris Prassinos
>            Assignee: Ryan McGuire
>         Attachments: 6731.traces.txt, ticket_6731.py
>
>
> Three-node cluster with RF=3. All data currently in sync.
> Network topology strategy. Each node defined to be on a different rack.
> endpoint_snitch: PropertyFileSnitch
> dynamic_snitch_update_interval_in_ms: 100
> dynamic_snitch_reset_interval_in_ms: 600000
> dynamic_snitch_badness_threshold: 0
> All tables defined with read_repair_chance=0
> From cqlsh when querying with Consistency=ONE tracing shows that requests get frequently
redirected to other nodes even though the local node has the data.
> There is no other activity on the cluster besides my test queries.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Mime
View raw message