HBase is telling you that writes to those datanodes are slow. Is it the same host names over and over? Probably they have high system load, a bad or dying disk, bad or dying network adapter, &c. Basically HBase is giving you a hint to go diagnose your cluster.
-n On Fri, Mar 13, 2015 at 2:44 AM, Serega Sheypak <serega.shey...@gmail.com> wrote: > Hi, finally I met HBase perfomance problems :( > I see this message: > org.apache.hadoop.hbase.regionserver.wal.FSHLog > > Slow sync cost: 345 ms, current pipeline: [] > > in log file > sometimes [] contains actual addresses of my datanodes. > What are the steps to undestand why HBase is so slow. > I have 7RS and 3K reads per second and 500 writes per second. Requests are > evenly distributed across cluster since I use hash for keys. >
