Hi Brock, The current implementation is half-baked when it comes to error handling during initialization. The main issue stopping this implementation is that we are yet to migrate fully to Guava Service implementation for lifecycle management of the various components. Once we do migrate, we will have to implement initialization time error handling so that the components can be disabled gracefully while others continue to work.
Thanks for bringing this up. I have filed the following issue to track this work: https://issues.apache.org/jira/browse/FLUME-966 Thanks, Arvind On Sun, Feb 12, 2012 at 7:04 PM, Brock Noland <[email protected]> wrote: > Hi, > > I am looking at Flume NG source and sink implementations. I have > noticed a common pattern, one of a few specific examples is: > > https://reviews.apache.org/r/3741 > > In the start method, if we get an IOException in the start method, we > log the error. However, when the process method is called a NPE will > be throw because Htable is still null. Is there a reason we don't wrap > the error in a runtime exception and throw it up? Another example is: > > https://reviews.apache.org/r/3831/ > > I am just curious if there is a reason we do log the error as opposed > to throwing it as a RuntimeException. > > Cheers > Brock > > -- > Apache MRUnit - Unit testing MapReduce - > http://incubator.apache.org/mrunit/ >
