[
https://issues.apache.org/jira/browse/HBASE-2361?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12848413#action_12848413
]
Jean-Daniel Cryans commented on HBASE-2361:
-------------------------------------------
bq. + Would it be worthwhile making the "optimization" that if no replication
config, do not create a scopes List? Or would this make things a little
involved?
Not sure how WALEdit could know that.
bq. REPLICATION_SCOPE_LOCAL should be in HLog rather than in HConstants? Is it
ever used outside of HLog?
Yes it is.
bq. For usual case, are we making two kvs arrays everytime we do a readFields?
Yeah had the same idea. And make those finals?
bq. + Is this right: + kvs = new ArrayList<KeyValue>(0);? Should be '1'?
Oversight, thx.
> WALEdit broke replication scope
> -------------------------------
>
> Key: HBASE-2361
> URL: https://issues.apache.org/jira/browse/HBASE-2361
> Project: Hadoop HBase
> Issue Type: Bug
> Reporter: Jean-Daniel Cryans
> Assignee: Jean-Daniel Cryans
> Fix For: 0.21.0
>
> Attachments: HBASE-2361.patch
>
>
> Before HBASE-2283, each KV had a HLogKey with a replication scope. Now a key
> applies to a list of KVs that spans multiple families so a single scope
> doesn't work anymore. Multiple possible solutions:
> - Each KV have their own scope. We already ruled that out in a previous jira
> since that means the scope would end up in the HFiles.
> - Store pairs of scope/KV in WALEdit instead of straight KVs.
> - Have 2 parallel lists in WALEdit, one for KVs and the other for scopes.
> - Subclass KV and add the scope there, those would be created when inserted
> in the WAL and would contain the KV stored in the HFiles.
> I'm sure there are other solutions, discuss.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
