[
https://issues.apache.org/jira/browse/CASSANDRA-20732?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17986423#comment-17986423
]
Caleb Rackliffe edited comment on CASSANDRA-20732 at 6/26/25 8:33 PM:
----------------------------------------------------------------------
|[5.0|https://github.com/apache/cassandra/pull/4213]|CI TODO|
|[trunk|https://github.com/apache/cassandra/pull/4215]|CI TODO|
was (Author: maedhroz):
|[5.0|https://github.com/apache/cassandra/pull/4213]|CI TODO|
|trunk|CI TODO|
> Avoid availability gap between UP and queryability marking for already built
> SAI indexes on bounce
> --------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-20732
> URL: https://issues.apache.org/jira/browse/CASSANDRA-20732
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Feature/SAI
> Reporter: Caleb Rackliffe
> Assignee: Caleb Rackliffe
> Priority: Normal
> Fix For: 5.0.x, 6.x
>
>
> On startup, already built SAI indexes are made queryable in two distinct
> ways. The first is asynchronous, as the secondary index initialization task
> is run. (SAI automatically repairs local column indexes that are incomplete
> and marks them queryable when this finishes.) The second is synchronous, in
> what are called "pre-join tasks". These run before a node is marked NORMAL
> after bootstrap. The problem with this design is that, on a node that is
> already NORMAL/bootstrapped, we don't actually attempt to make already built
> indexes queryable before the node starts accepting requests. The problem is
> transient, but it should also be easy to address, at least for SAI, by
> synchronously validating the index when the initialization task is requested
> and marking it queryable, if possible, before CFS initialization finishes.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]