We just found the issue! It was an error in the Node library we've been working with/updating - seems to be formatting the binary packet wrong when encoding the offset. The dump segment tool was really useful to help confirm that the problem was external - thanks for the pointer to that and the responsiveness on the list and IRC.
I probably could have used dump segment several times in the past - it'd be great to have a script wrapped around it. On Thu, Nov 3, 2011 at 9:58 PM, Jun Rao <jun...@gmail.com> wrote: > If you are not using zk, are you using SimpleConsumer then? > > Jun > > On Thu, Nov 3, 2011 at 4:31 PM, Taylor Gautier <tgaut...@tagged.com> > wrote: > > > I'm working with Nalin on this…the offsets in the segment are really low. > > The client is submitting an offset that appears to be within the proper > > range and then when kafka gets it for some reason it gives an error with > > this huge number. We will double check that the offset provided is > exactly > > correct. > > > > Also there was not any zk involved, we've turned it off for the time > being. > > > > On Thu, Nov 3, 2011 at 4:24 PM, Jun Rao <jun...@gmail.com> wrote: > > > > > Nalin, > > > > > > It sounds like the consumer computed a wrong offset. At this moment, > you > > > will have to either manually change the offset in ZK to a valid one > (use > > > any offset returned in the DumpLogSegement tool), or use a new consumer > > > group (which will consume from either the head or the tail of the > queue). > > > > > > Jun > > > > > >
