AFAICT, the implementation in hbase-asyncfs is too specific. If we want to push it back to Hadoop/HDFS, we should implement a more generic version.
We have tried this before but with no luck, you can see HDFS-9924, and also HDFS-13572, where we even posted a design doc. And finally we just stopped at implementing async IPC... For me, I think hadoop will likely to have less active branches so it should be easier for us to maintain hbase-asyncfs by ourselves. Of course if we can finally push this back to Hadoop/HDFS, it is also very good. Thanks. Istvan Toth <st...@apache.org> 于2024年11月11日周一 14:28写道： > > Hi! > > We have a lot of issues with keeping hbase-asyncfs working with changes in > different Hadoop versions, and have to resort to more and more hacks to > keep it working. > > AFAIU, hbase-asyncfs itself is mature, and most, if not all recent changes > have been keeping up with the Hadoop changes, and fixing issues already > fixed in Hadoop. > > AFAIU Hadoop is open to adding the required functionality. > > This is quite a bit of work now, but in the long run it would make our > lives much easier, and would remove the main source of incompatibilities > with Hadoop patch releases. > > Should we push the relevant code to Hadoop ? > Are there any cons that I'm not aware of ? > > Istvan
