Yes this could be the problem as I testing one regions server vary hard to
find a breaking point with a datanode and 4 task trackers on the same
server.
The server is only a dual 2.4Ghz with 4Gb memory with a 147GB SCSI 10K rpm
drive
I got 16 threads on 4 server includeing the region server doing a bulk
insert on 1 table.
So far so good everything is holding up much better then it was a few months
ago.
I have to hand it to your guys its much more stable now.
I am getting a lot better memory useage and speed sense I switch from Rest
to thrift.
splitting, compactions, and flushes are running smoothly.
Also note this type of problem could cause the region server lose there
lease
I have had to update the hbase.master.lease.period to a much larger number
then default to keep the region servers from killing them self under heave
load like this
I maybe something to add to the FAQ.
Billy Pearson
|