My opinion is that HADOOP-5744 does nor change the major semantics that were laid out in HADOOP-1700... e.g single appender, new reads see flushed data, etc.etc If you can post your design on hdfs byzantine fault tolerant and then describe the assumptions of hdfs-appends that it relies on, that will be great.
thanks, dhruba On Mon, May 4, 2009 at 10:29 AM, Sangmin Lee <[email protected]> wrote: > Dhruba, > > I'm working on making hdfs byzantine fault tolerant and was wondering if I > have to account for current implementation of append. > If all append semantic and design will be changed (as HADOOP-5744 > indicated), > I might have to wait for it to come out. > If the change won't be significant, I will probably continue with current > implementation. > > Do you have any comment on this? > Thanks, > > Sangmin > > On Mon, May 4, 2009 at 3:20 AM, Dhruba Borthakur <[email protected]> wrote: > > > The code for "append" is not supported in stable releases. The .doc from > > HADOOP-1700 explains the semantics of "append". There are quite a few > open > > issues that need to be addressed before "append" is production ready for > > general applications, i.e. HADOOP-4663, HADOOP-4379, HADOOP-5027 among > > others. > > > > what type of application do you have that wants to use "append"? > > > > thanks > > dhruba > > > > > > On Mon, May 4, 2009 at 12:55 AM, Sangmin Lee <[email protected]> > > wrote: > > > > > Hi all, > > > > > > I am wondering if append for hdfs is supported in stable releases. > > > If not, what is the issue? ( I see code for append is already there in > > > repository ) > > > Also where can I find the documentation of current semantic of append? > > > (I see some .doc files from HADOOP-1700 but not sure if they are > > out-dated) > > > Can anyone shed some light on this? > > > > > > Thanks, > > > Sangmin > > > > > >
