[ https://issues.apache.org/jira/browse/FLUME-238?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Ashish Paliwal resolved FLUME-238. ---------------------------------- Resolution: Won't Fix Fix Version/s: v0.9.5 Won't fix. 0.X branch not maintained anymore > Add more details in source/sink/decorator semantics > --------------------------------------------------- > > Key: FLUME-238 > URL: https://issues.apache.org/jira/browse/FLUME-238 > Project: Flume > Issue Type: Improvement > Affects Versions: v0.9.2 > Reporter: Jonathan Hsieh > Fix For: v0.9.5 > > > FLUME-155 is a good first cut but more improvements can be done. This jira > is the place holder for some of these issues. > Feedback from Vibhor on first cut. (FLUME-155) > """" > Most of it looks good and very interesting. I'll give the high level comments > here, gave minor comments on paper to Jon. > 1) Missing the main Threading detail, it should be explained that there is a > main driver thread, and different decorators can create their own threads too. > 2) In the section talking about the sinks with buffered memories it says > close() from a closing state "should blocks until resources are freed, it > should attempt to flush its buffers before returning. For example, so if a > network source has some buffered data, the network connection should be > closed to prevent new data from entering, and then the buferred data should > be flushed", > but at the same time it is said that the next() should keep on pulling data > from the buffer even in the closing state. As both these methods (close() and > next()) can be from different threads, it should be clearly mentioned how are > we handling the race conditions while operating on the same data. > 3) In both Source/Sink descriptions, it is said that if a timeout happens in > a closing state, then we move to a close state. But do we ensure that we have > given up all the resources before it? Else there will be an issue when > someone later will try to open this decorator. > 4) It should be clearly explained why and how we handle the > Thread-Interrupted-Exception when a append() on sink gets interrupted while > it is blocked. > """ -- This message was sent by Atlassian JIRA (v6.3.4#6332)