cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Carl Yeksigian (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-10405) MV updates should optionally wait for acknowledgement from view replicas
Date Fri, 02 Oct 2015 16:13:26 GMT


Carl Yeksigian commented on CASSANDRA-10405:

It was one of the requests/concerns that came up after the talk at summit. Async writes to
the MV break the ability to write and read at quorum (or even all) and get consistent data

This would work OK if there were just a handful of views that you were waiting on ACKs from;
waiting for a large number of views would kill your availability. That's why this should be
a per-view option instead of per-node or per-base table.

> MV updates should optionally wait for acknowledgement from view replicas
> ------------------------------------------------------------------------
>                 Key: CASSANDRA-10405
>                 URL:
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Carl Yeksigian
>              Labels: materializedviews
>             Fix For: 3.x
> MV updates are currently completely asynchronous in order to provide parallelism of updates
trying to acquire the partition lock. For some use cases, leaving the MV updates asynchronous
is exactly what's needed.
> However, there are some use cases where knowing that the update has either succeeded
or failed on the view is necessary, especially when trying to allow read-your-write behavior.
In those cases, we would follow the same code path as asynchronous writes, but at the end
wait on the acknowledgements from the view replicas before acknowledging our write. This option
should be for each MV separately, since MVs which need the synchronous properties might be
mixed with MV which do not need this.

This message was sent by Atlassian JIRA

View raw message