[
https://issues.apache.org/jira/browse/MAILBOX-297?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Tellier Benoit closed MAILBOX-297.
----------------------------------
> Cassandra setFlags operation speed improvment for ranges
> --------------------------------------------------------
>
> Key: MAILBOX-297
> URL: https://issues.apache.org/jira/browse/MAILBOX-297
> Project: James Mailbox
> Issue Type: Bug
> Components: cassandra
> Affects Versions: master
> Reporter: Tellier Benoit
> Fix For: master
>
>
> We observed that flags updates are very slow on large ranges for Cassandra
> implementation.
> A quick audit shows that:
> - We read full messages, and not just specific metadata. Thus we endup
> performing a "join" on James side, which is costly and not needed
> - We generate a new modseq on every message, paying much synchronisation
> costs.
> - We process the given flags sequentially.
> This is the proposed algorithm changes:
> {code:java}
> 1. Generate MODSEQ
> 2. Read messages metadata to be updated
> 3. Update each message, with condition that MODSEQ should not change
> since read (Compulsory to not mess up conditional requests):
> - If OK, compute UpdatedFlags
> - If not OK, add UID to failed list
> 4. Restart from 1 and try to update the failed list. several time.
> Step 3 can be done in a parallel fashion.
> My answer continues to try to do the update with the failed list.
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
