That was one of the examples I gave that doesn't cover all cases. We have admin tools that can move leader away selectively. In that case, the partitions would have different error codes.
Thanks, Neha On Wed, Oct 23, 2013 at 1:36 AM, xingcan <xingc...@gmail.com> wrote: > Neha, > > Thanks for your prompt reply. And I got another two questions. As I wrote > before, > my mechanism is to add all partitions belonging to the same leader broker > to > a single request. And then send these requests one by one for each broker. > Is this necessary? And if that, all topic and partitions in one > FetchResponse > should belong to the same broker. Will the broker's down lead to partial or > definitely total error for all topic and partitions? > > > > On Tue, Oct 22, 2013 at 10:12 PM, Neha Narkhede <neha.narkh...@gmail.com > >wrote: > > > We need to return an error code per partition since your fetch request > > could've succeeded for some but not all partitions. For example, if one > > broker fails, some partitions might temporarily return a > LeaderNotAvailable > > error code. So you have to go through the individual error codes to know > > which partitions you need to retry the operation for. > > > > Thanks, > > Neha > > On Oct 21, 2013 11:27 PM, "xingcan" <xingc...@gmail.com> wrote: > > > > > Hi, > > > > > > After migrating from 0.72 to 0.8, I still use SimpleConsumer to > construct > > > my own consumer. By FetchRequestBuilder, I add all partitions belonging > > to > > > the same broker to a single request and get a FetchResponse for all > these > > > partitions. However, I find the error code in FetchResponse is a > little > > > hard to retrieve for I must iterate all these partitions to check. Is > > there > > > any tips or suggestions to deal with this or maybe the API provided by > > > FetchResponse could be little changed? > > > > > > -- > > > *Xingcan* > > > > > > > > > -- > *Xingcan* >
