Offsets always begin at 0 for each partition and increase sequentially from there. Offsets aren't unique within a topic. As old data is discarded the first retained offset will not remain 0. The behavior of what is retained is controlled by your retention settings.
In trunk there is a feature that allows removing messages from the log ( https://cwiki.apache.org/confluence/display/KAFKA/Log+Compaction). This would leave holes between offsets for topics where you enabled compaction. -Jay On Mon, Oct 7, 2013 at 12:45 PM, Francis Dallaire < francis.dalla...@ubisoft.com> wrote: > Hi, > > I have question regarding the offset in kafka (0.8). I've gone through the > documentation and did some tests, but I want to make sure I'm on the right > track. > > > * Are the offsets guaranteed to be sequential in a partition ? > > o Can it contain holes ? > > * How offsets are distributed amongst partitions ? > > o Unique inside a partition, not inside a topic ? > > o What determines where it starts for each partition ? > > If you have a link to some documentation, that could be useful. > > Thanks, > > Francis >
