cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Ariel Weisberg (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-7392) Abort in-progress queries that time out
Date Mon, 27 Jul 2015 17:26:07 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-7392?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14643077#comment-14643077
] 

Ariel Weisberg commented on CASSANDRA-7392:
-------------------------------------------

Not done reviewing. Just wanted to drop a thought on design/performance. Right now it schedules
a monitoring task for every operation. This has a cost linear to the number of operations
and adds a contended queue with a not great queue implementation (priority queue). My wild
guess is that there may be a workload we care about that will highlight this. If we do our
jobs it should bottleneck eventually as it is shared state accessed for every single operation.

What I would suggest is inverting the process. Have a thread wake up periodically and walk
the list of in progress operations looking for ones that need to be timed out. You can store
the in progress operation in a thread local. When the thread local is created it inserts into
a COW array list that the monitoring thread can iterate. That way there is no mutation of
shared state and it only enters shared state periodically when the the monitoring thread walks
the list.

This alternative approach has a cost linear to the number of threads dispatching tasks and
almost no coherence cost.

> Abort in-progress queries that time out
> ---------------------------------------
>
>                 Key: CASSANDRA-7392
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-7392
>             Project: Cassandra
>          Issue Type: New Feature
>          Components: Core
>            Reporter: Jonathan Ellis
>            Assignee: Stefania
>             Fix For: 3.x
>
>
> Currently we drop queries that time out before we get to them (because node is overloaded)
but not queries that time out while being processed.  (Particularly common for index queries
on data that shouldn't be indexed.)  Adding the latter and logging when we have to interrupt
one gets us a poor man's "slow query log" for free.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message