-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/3980/#review5837
-----------------------------------------------------------


Thanks for the patch Brock. 

I am not in favor of an interface to be implemented by the exception to signal 
any specific cause. The reason being that it will be hard and error prone for 
us to bring this into compliance specially for components developed outside of 
Flume. Any explicit contract that cannot be enforced is likely to become a 
cause of headaches in the field. Hence, my suggestion is to do the following:

* Remove the FatalException interface
* Update the runners to catch(Exception) once they have dealt with all the 
known exception types, log it and do a BACKOFF wait simulation.

This will ensure that the runners do not bail out if an exception occurs and 
keep retrying. This may be futile in some cases but would make sense in other 
cases as once a process/thread exits there is no restoring it without more 
complex logic up top or via manual intervention.

- Arvind


On 2012-02-29 10:42:17, Brock Noland wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/3980/
> -----------------------------------------------------------
> 
> (Updated 2012-02-29 10:42:17)
> 
> 
> Review request for Flume.
> 
> 
> Summary
> -------
> 
> Catches, logs, and sleeps when a general exception occurs. This is similar to 
> PollingSourceRunner.
> 
> 
> This addresses bug FLUME-984.
>     https://issues.apache.org/jira/browse/FLUME-984
> 
> 
> Diffs
> -----
> 
>   flume-ng-core/src/main/java/org/apache/flume/FatalException.java 
> PRE-CREATION 
>   
> flume-ng-core/src/main/java/org/apache/flume/source/PollableSourceRunner.java 
> 264fce1 
>   flume-ng-core/src/main/java/org/apache/flume/SinkRunner.java 9e46d87 
> 
> Diff: https://reviews.apache.org/r/3980/diff
> 
> 
> Testing
> -------
> 
> All unit tests pass.
> 
> 
> Thanks,
> 
> Brock
> 
>

Reply via email to