[
https://issues.apache.org/jira/browse/HDFS-1073?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12924103#action_12924103
]
Robert Chansler commented on HDFS-1073:
---------------------------------------
Is the file 100-100 _forbidden_? What if the service is stopped when the most
recent file has zero records? (I'd always write a "I'm quitting" record,
otherwise you can never know if you have lost the last edits.) And what if
there are files 100-200 and 100-300? Rather than different special cases, why
not make the general case just work? Roll means roll regardless, and starting
up finds the latest image and _any_ consistent sequence of edits that -starts
with- includes the very next transaction, reporting whether the last available
edit record is "I'm quitting!".
And catching up with Sanjay's comment about tx ids in every record, it would
seem that the principal benefits are really obtained only if the tx id is
assigned to requests as they are _received in sequence_. Just doing
{{log.write(id++)}} doesn't offer much real protection.
If there is a tx id per record, would it make sense for the actual bits be the
record check sum+id? Years ago we discussed having record check sums, but it
never became a priority.
(In file N-M, I might have expected that the first record, if any, has tx id N,
not N+1.)
> Simpler model for Namenode's fs Image and edit Logs
> ----------------------------------------------------
>
> Key: HDFS-1073
> URL: https://issues.apache.org/jira/browse/HDFS-1073
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: Sanjay Radia
> Assignee: Todd Lipcon
> Attachments: hdfs-1073.txt, hdfs1073.pdf
>
>
> The naming and handling of NN's fsImage and edit logs can be significantly
> improved resulting simpler and more robust code.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
