[jira] [Commented] (CASSANDRA-21767) Accord: first read of untouched key times out

["Michael Semb Wever (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-21767?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18128182#comment-18128182 ] 

Michael Semb Wever commented on CASSANDRA-21767:
------------------------------------------------

[~aleksey], [~benedict], 
I took a stab at this too, again you obviously cannot include me as a committer here.
The jvm-dtest in CASSANDRA-21765's patch demonstrates the use-case where this was first encountered, while the unit test in this patch is the minimal reproduction.

https://github.com/apache/cassandra-accord/pull/292

> Accord: first read of untouched key times out
> ---------------------------------------------
>
>                 Key: CASSANDRA-21767
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-21767
>             Project: Apache Cassandra
>          Issue Type: Bug
>          Components: Accord
>            Reporter: Michael Semb Wever
>            Priority: Normal
>
> On a new cluster I have seen null dereferences of the value returned by  {{SafeCommandStore.get(RoutingKey)}}, at two call sites:
> * {{SafeCommandStore.updateManagedCommandsForKey:521}}, invoking {{SafeCommandsForKey.callback}}.
> * {{Commands.setDurable:1384}}, invoking {{SafeCommandsForKey.setDurable}} through {{InformDurable}}.
> {{AbstractSafeCommandStore.canExecute()}} used a non-null reference as proof that a key was ready. Incremental Cassandra tasks retain references for keys outside their current initialised batch. Those references pass the presence check but {{SafeCommandStore.get(key)}} returns {{null}}.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cassandra.apache.org
For additional commands, e-mail: commits-help@cassandra.apache.org