cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sylvain Lebresne (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-7925) TimeUUID LSB should be unique per process, not just per machine
Date Fri, 22 May 2015 14:28:18 GMT


Sylvain Lebresne commented on CASSANDRA-7925:

bq. It's currently used for paxos.

Right, and that's not a problem for Paxos.

bq. Also, I imagine we will still allow the clients to specify timestamps (esp for thrift)

We certainly will want to preserve backward compatibility (both for thrift and CQL), but doing
so mean that we will need to guarantee to 2 updates with the same (user provided) timestamp
actually *do* conflict, and this no matter what node the update hits. So in fact, we'll probably
have to hardcode a LSB to use for all update with user provided timestamp. In any case, I
think anticipating problems for CASSANDRA-7919 is a bit premature. It's not like we can't
change this if we really need to later.

In general, I'd prefer keeping it to a fixed LSB for a given process if possible: it's a tad
simpler, better for compression and a bit closer to the timeuuid RFC imo. And as of now, I
think that's good enough.

> TimeUUID LSB should be unique per process, not just per machine
> ---------------------------------------------------------------
>                 Key: CASSANDRA-7925
>                 URL:
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Peter M├Ądel
>            Assignee: T Jake Luciani
>             Fix For: 2.2.x
>         Attachments: cassandra-uuidgen.patch
> as pointed out in [CASSANDRA-7919|]
lsb collisions are also possible serverside.
> a sufficient solution would be to include references to pid and classloader within lsb.

This message was sent by Atlassian JIRA

View raw message