Hi Brock, On Fri, Apr 13, 2012 at 12:12 PM, Brock Noland <[email protected]> wrote:
> On Fri, Apr 13, 2012 at 7:05 PM, Bruno Mahé <[email protected]> wrote: > > On 04/13/2012 11:00 AM, Brock Noland wrote: > >> > >> I have been thinking about this for a while. Let's say that we wanted > >> to write some integration tests, that is programmatically startup > >> flume node with a specific configuration. Given the current module > >> layout, where would we do this? It seems like the node module would be > >> the correct module but then it would have to have knowledge of all > >> other modules. > >> > > Could you define "integration"? > > Depending on your definition, Apache Bigtop may be more suitable for > > such integration tests. > > Start up a flume agent with a SequenceGeneratingSource, using > RecoverableMemoryChannel, and writing to HDFSSink using file:/// as > the FileSystem run it for a few seconds and shut it down. > In the past I have used the FIT framework successfully to do integration testing. It has a bit of a steep starting cost, but once the fixtures are developed and a few tests written, it becomes very easy to extend and maintain. http://fit.c2.com/ The other alternative is to use TestNG/JUnit categorized for integration test where you create the various processes in a controlled environment etc. This approach is a bit more tedious and harder to maintain, although easer to start with. Thanks, Arvind Prabhakar
