> On April 20, 2015, 11:18 p.m., Joel Koshy wrote: > > core/src/main/scala/kafka/server/OffsetManager.scala, line 124 > > <https://reviews.apache.org/r/32650/diff/1/?file=909897#file909897line124> > > > > I think there is an issue in relying on the metadata cache mainly due > > to start-up. E.g., when we start up the broker (and offset manager) the > > metadata cache will actually be empty so this would delete _all_ the > > offsets. Unfortunately even after start-up there is no _guarantee_ that you > > have the most current information in the cache (say, if the controller > > failed to send an UpdateMetadataRequest to the broker by the time the > > compactor task runs)
Actually - I think what you have is correct. The offset cache would be empty at start-up and would only be populated on becoming leader. However, we just need to make sure that we get the complete cluster topic metadata before the compactor thread runs. I'll take another look tomorrow. - Joel ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/32650/#review80857 ----------------------------------------------------------- On March 30, 2015, 9:47 p.m., Sriharsha Chintalapani wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/32650/ > ----------------------------------------------------------- > > (Updated March 30, 2015, 9:47 p.m.) > > > Review request for kafka. > > > Bugs: KAFKA-2000 > https://issues.apache.org/jira/browse/KAFKA-2000 > > > Repository: kafka > > > Description > ------- > > KAFKA-2000. Delete consumer offsets from kafka once the topic is deleted. > > > Diffs > ----- > > core/src/main/scala/kafka/server/OffsetManager.scala > 395b1dbe43a5db47151e72a1b588d72f03cef963 > > Diff: https://reviews.apache.org/r/32650/diff/ > > > Testing > ------- > > > Thanks, > > Sriharsha Chintalapani > >
