[
https://issues.apache.org/jira/browse/STORM-837?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14693521#comment-14693521
]
ASF GitHub Bot commented on STORM-837:
--------------------------------------
Github user revans2 commented on the pull request:
https://github.com/apache/storm/pull/644#issuecomment-130315702
For me internally the tests pass and everything builds. Although the rest
of storm seems a bit unstable with tests.
I think the changes look fine +1 for merging this in.
I don't think we have any split brain problems because of how we route
messages, so the begin command and commit messages should prevent this.
There are some things that I think we can do better, but are not wrong.
1) I am scared that the transaction file is going to put more of a load on
the Name Node than I like. I am mostly concerned that if we have thousands of
instances of hdfs state creating/deleting transaction files multiple times a
second the load is going to noticeable on a large cluster. But this is a
premature optimization so I would say we don't address it unless we see it as
an actual problem.
2) The RotationActions don't have a recovery interface, and things like the
MoveFile action could potentially fail after successfully rotating the file
which might leave the file in the wrong directory.
3) I don't really like the special case for the TimedRotationAction.start.
I would rather see start a part of the RotationAction interface, that can be
ignored by those that don't want/need it.
The last two could perhaps be a follow on JIRA. I am fine with the code as
is, but I would like another pair of eyes looking at this too before merging it
in.
> HdfsState ignores commits
> -------------------------
>
> Key: STORM-837
> URL: https://issues.apache.org/jira/browse/STORM-837
> Project: Apache Storm
> Issue Type: Bug
> Reporter: Robert Joseph Evans
> Assignee: Arun Mahadevan
> Priority: Critical
>
> HdfsState works with trident which is supposed to provide exactly once
> processing. It does this two ways, first by informing the state about
> commits so it can be sure the data is written out, and second by having a
> commit id, so that double commits can be handled.
> HdfsState ignores the beginCommit and commit calls, and with that ignores the
> ids. This means that if you use HdfsState and your worker crashes you may
> both lose data and get some data twice.
> At a minimum the flush and file rotation should be tied to the commit in some
> way. The commit ID should at a minimum be written out with the data so
> someone reading the data can have a hope of deduping it themselves.
> Also with the rotationActions it is possible for a file that was partially
> written is leaked, and never moved to the final location, because it is not
> rotated. I personally think the actions are too generic for this case and
> need to be deprecated.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
