[
https://issues.apache.org/jira/browse/HBASE-1897?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12764132#action_12764132
]
Jean-Daniel Cryans commented on HBASE-1897:
-------------------------------------------
HBASE-1756 is about refactoring HLog and I have some stuff done already.
> WAL is spread all over code-base and presumption is its SequenceFile;
> encapsulate inside a wal subpackage
> ---------------------------------------------------------------------------------------------------------
>
> Key: HBASE-1897
> URL: https://issues.apache.org/jira/browse/HBASE-1897
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: stack
>
> Presumption is that WAL is a sequencefile. I just spent some time looking
> again at my old buddy SF and its kinda heavy-duty for our needs. Do we need
> the sync bytes it writes into the stream every time you call a sync? Maybe
> it'd help recovering logs of edits? We don't need the compression by record
> or block, metadata info, etc. Its also currently unsuited because its sync
> actually doesn't do a sync on the backing stream:
> {code}
> /** create a sync point */
> public void sync() throws IOException {
> if (sync != null && lastSyncPos != out.getPos()) {
> out.writeInt(SYNC_ESCAPE); // mark the start of the
> sync
> out.write(sync); // write sync
> lastSyncPos = out.getPos(); // update lastSyncPos
> }
> }
> {code}
> We should move all of our HLog stuff into a wal package and rename the
> classes as WAL, WALEdit, etc. Splitting code and replay code should
> reference a WAL.Reader rather than a SequenceFile.Reader, etc.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
