Thats super useful! Thanks a lot Le ven. 24 juil. 2020 à 16:29, István Fajth <[email protected]> a écrit :
> Hi Michel, > > great to see your interest in Ozone and your testing results. > Please go through this blog post: > > https://blog.cloudera.com/multi-raft-boost-up-write-performance-for-apache-hadoop-ozone/ > > This will describe how write path is working in Ozone, and will give you > the tuning tips you need in order to change this behavior which is kind of > by design for such small clusters unfortunately. > > If you have further questions please feel free to ask anytime ;) > Pifta > > Michel Sumbul <[email protected]> ezt írta (időpont: 2020. júl. 24., > Pén 17:07): > > > Hi guys, > > > > I installed ozone 0.5 (with CDP but i don't think it changed anything) > and > > I enabled kerberos. The setup has 4 datanodes with a single rack. > > When I ingest data in a specific bucket using the HadoopFS with Nifi, I > can > > see that all the containers get always created on the 3 same datanodes. > > Is there any reason why the last datanode don't get any container? Is it > an > > expected behavior and how can I troubleshoot that? I didn't see anything > in > > the log that can give me a hint. > > > > I tried to stop one of the datanode having containers to see if the > > containers will get created on the datanode number4 but nothing > happened. I > > restarted the OM/SCM and the 3 datanodes still up and then it started to > > replicate the containers because there was a missing replica. > > > > After bringing back the datanode that I shutdown at the beginning, I > > restart the ingestion of data and again the new container gets created > only > > 3 specific datanode. > > > > If you have any tips or explanations that will be helpful. > > > > Thanks, > > Michel > > >
