[
https://issues.apache.org/jira/browse/TEZ-1076?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14057237#comment-14057237
]
Siddharth Seth commented on TEZ-1076:
-------------------------------------
bq. There was some discussion on having versioning inside TezUserPayload and
making it public?
This has nothing to do with payload versioning. Instead, is representative of
the attemptId (speculation for example) - similar to DataMovementEvents. We
should likely rename the field there as well.
bq. Please follow the 80 char line as much as possible. Some of the javadocs
and method signatures are quite long.
Following the 100 char line convention, which I believe is what Tez is
following - even though we don't have a HowToContribute page. Should probably
be closer to 120 characters per line.
bq. Renaming the member var to wrapper or initializerWrapper will reduce
confusion
Will do
bq. getEntityName() can just be getName(). The object on which its being called
is known. So the method does not have to be explicit about it.
WIll look into this when I change the patch.
bq. Can the vertex name be invalid? Existing code may be suffering from this
issue too
The vertex name can be invalid, the input name can be invalid. It's also
possible that the user code which handles the actual event throws an exception.
We have no error checking for this - and such errors end up causing the AM to
die, and recovery to kick in. I'd opened a jira a couple of days ago for this.
Not addressing as part of this patch. Will look into the error handling jira
soon.
bq. Is this check relevant for the new event?
It could be used for the initial routing. Don't see it as required though,
since the source-meta should be generated by Tez itself. (An assert maybe)
bq. Log is probably wrong
Will fix.
bq. The check is odd. We should know whether we will stabilize in RUNNING or
INITED state and check for that state only, right?
We move the INITED, and then generate a START event right ?, which is being
processed separately. Since that isn't intercepted - the vertex could end up in
either state, but should at least move out of INITIALIZING.
> Allow events to be sent to InputInitializers
> --------------------------------------------
>
> Key: TEZ-1076
> URL: https://issues.apache.org/jira/browse/TEZ-1076
> Project: Apache Tez
> Issue Type: Improvement
> Reporter: Siddharth Seth
> Assignee: Siddharth Seth
> Attachments: TEZ-1076.1.txt
>
>
> Events from tasks / vertices to InputInitializers.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
