hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-2195) Support cyclic replication
Date Thu, 08 Sep 2011 04:51:09 GMT

    [ https://issues.apache.org/jira/browse/HBASE-2195?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13099893#comment-13099893

stack commented on HBASE-2195:

@Ted I see what you are saying.  But if we add in VersionedWritable, he'll not be able to
do his trick where he can deserialize this version of HLogKey and the previous one (since
it will read a byte off the stream and treat it as the version regardless).  As I see it,
we can:

+ Leave patch as is.  Then later, if we need to change HLogKey, we'll up the version to '0'
('0' is an unlikely length on an vint for encoded name).  So we've squeezed an extra version
in.  Thereafter we'd have to move to use versions 1, 2, 3, etc.  If we ever encountered a
WAL that had HLogKeys written before this patch, we'd not be able to deserialize them since
we'd confuse them with these newer 1, 2, 3.. versions.  I think that should be fine since
a good bit of time will have passed between the two version and deserializing old stuff will
be less likely.
+ We move to VersionedWritable now but that'll break our being able to read WALs written before
this patch went in.

Like I say, I think the second option above tolerable given as we're going to a major version
and we can state we can't read WALs across versions but it seems like leaving the patch as
is, option 1, gives us most flexibility (we can read old WALs).

What you think Ted?

> Support cyclic replication
> --------------------------
>                 Key: HBASE-2195
>                 URL: https://issues.apache.org/jira/browse/HBASE-2195
>             Project: HBase
>          Issue Type: Sub-task
>          Components: replication
>            Reporter: Jean-Daniel Cryans
>            Assignee: Lars Hofhansl
>         Attachments: 2195-v10.txt, 2195-v12.txt, 2195-v5.txt, 2195-v6.txt, 2195.txt
> We need to support cyclic replication by using the cluster id of each HlogKey and stop
replicating when it goes back to the original cluster.

This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira


View raw message