cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sergio Bossa (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-7870) Cannot execute logged batch when only the coordinator node is alive
Date Wed, 03 Sep 2014 18:40:51 GMT


Sergio Bossa commented on CASSANDRA-7870:

bq. batchlog is concerned with permadeath and not just temporary outages

Right, but the "single node alive" case is still a valid degenerate case. 
If we allow logged batches in such case, we have the following failure scenarios:
1) Permanent death of the coordinator node mid-batch: no one else has seen the partial batch
application, so any node coming back alive after it will not suffer any inconsistencies.
2) Temporary death with restart: the node will replay the batch (with outdated mutations being

There may be implementation details I'm missing or oversimplifying (and I'd argue replaying
on restart at a later unknown point in time sounds a bit dangerous so I wouldn't actually
do it and make this degenerate case fall in the same category as "I want ANY consistency"),
but in terms of "contract", I don't see how it is violated at all.

> Cannot execute logged batch when only the coordinator node is alive
> -------------------------------------------------------------------
>                 Key: CASSANDRA-7870
>                 URL:
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>            Reporter: Sergio Bossa
>            Priority: Critical
> As per issue summary.
> This is probably a bug, rather than a consequence of needing to replicate the batchlog,
as if only the coordinator is alive the batch cannot be partially executed on other nodes
(as there are no other nodes).

This message was sent by Atlassian JIRA

View raw message