cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jonathan Ellis (Commented) (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-3958) Remove random HH delay
Date Mon, 27 Feb 2012 17:50:49 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-3958?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13217329#comment-13217329
] 

Jonathan Ellis commented on CASSANDRA-3958:
-------------------------------------------

Not really.  I'm thinking in terms of the common case (I have a few hints to deliver from
a GC pause or network hiccup) and worst case (you went down hard so I have {{max_hint_window_in_ms}}
worth of writes).

In the common case you're not going to overload the recipient with or without the delay. 
And in the worst case 60s is negligible compared to your hint volume.

And in both cases the extra delay is meaningless with {{hinted_handoff_throttle_delay_in_ms}}
around to do a better job of throttling.

Basically the 60s delay was a bad hack that never really worked and post-{{hinted_handoff_throttle_delay_in_ms}}
only confuses people who actually read the logs.
                
> Remove random HH delay
> ----------------------
>
>                 Key: CASSANDRA-3958
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-3958
>             Project: Cassandra
>          Issue Type: Improvement
>          Components: Core
>            Reporter: Jonathan Ellis
>            Assignee: Jonathan Ellis
>            Priority: Trivial
>              Labels: hintedhandoff
>             Fix For: 1.1.0
>
>         Attachments: 3958.txt
>
>
> {code}
> .       // sleep a random amount to stagger handoff delivery from different replicas.
>         // (if we had to wait, then gossiper randomness took care of that for us already.)
>         if (waited == 0)
>         {
>             // use a 'rounded' sleep interval because of a strange bug with windows:
CASSANDRA-3375
>             int sleep = FBUtilities.threadLocalRandom().nextInt(2000) * 30;
>             logger_.debug("Sleeping {}ms to stagger hint delivery", sleep);
>             Thread.sleep(sleep);
>         }
> {code}
> This is obsolete now that we have the per-hint configurable delay.  And large hint loads
(which are the ones that matter most) are going to overlap anyway even with the maximum 60s
difference.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message