[
https://issues.apache.org/jira/browse/CHUKWA-330?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12723672#action_12723672
]
Jerome Boulon commented on CHUKWA-330:
--------------------------------------
For my proposal:
1- if the client provide a name, then this name overwrite all the
AdaptorManager logic
2- If there's no name then the AdaptorManager rely on the underlying adaptor to
provide one.
3- >> uniqueness is a problem for whoever is issuing the "add" command
I still think that if the one that is doing the add is automatically able to
figure out a unique name in order to avoid duplicate in a distributed
environment without talking to anyone else, then the that logic can be pushed
down the adaptorManager, who is the only one to have the global vision.
4- >> isDuplicate() need to be called for each existing adaptor, every time we
create one
No. the AdaptorManager, receive a Add command then call a isDuplicate's method
to the factory,
Here the workfow:
4.1- At the factory level, if there's no adaptor already registered for this
AdaptorType then there's no duplicate
4.1.1- if it's not a duplicate The AdaptorManger call the factory to create the
adaptor
4.1.1.1- The factory create the new adaptor and register that instance as the
instance that will provide all answers for the isDuplicate call for that
specific AdaptorType (hash.put(AdaptorType, adaptor) )
4.1.2- if there's already one adaptor registered for the AdaptorType then the
factory return hash.get(AdaptorType).isDuplicate(...)
4.1.2.1- At the AdaptorManager level if it's a duplicate then it returns the
existing adaptor based on the name, otherwise call the factory
So there's no iteration over any adaptors, just a get from a hash based on
AdaptorType + 1 call to isDuplicate(..) and therefore there's no performance
issue
> controller should not create duplicates for adaptors with offsets
> -----------------------------------------------------------------
>
> Key: CHUKWA-330
> URL: https://issues.apache.org/jira/browse/CHUKWA-330
> Project: Hadoop Chukwa
> Issue Type: Bug
> Components: input tools
> Reporter: Ari Rabkin
> Assignee: Ari Rabkin
> Priority: Blocker
> Fix For: 0.2.0, 0.3.0
>
> Attachments: CHUKWA-330.patch
>
> Original Estimate: 1.5h
> Remaining Estimate: 1.5h
>
> The offset-of-first-byte param to FileTailingAdaptor is hashed into the
> default adaptor name. This changes the semantics of adaptor uniqueness, and
> broke metrics and the log4jappender.
> Proposed fix:
> Add a new "addByName()" method to ChukwaAgentController that lets callers
> specify names.
> Use this new method whenever finer control of adaptor uniqueness is needed.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
