hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Robert Gonzalez <g...@maxpointinteractive.com>
Subject HBase is not ready for Primetime
Date Tue, 12 Apr 2011 16:49:54 GMT
We've been using HBase for about a year, consistenly running into
problems where we lost data.  After reading forums and some back and
forth with other Hbase users, we changed our data methodology to save
less data per row.  This last time, we upgraded to 0.90 at the
recommendation of the hbase community, cleared off all our data, and
started over.  Seemed to be running ok for a couple of months, until
this morning.  One of the regionservers stopped responding to data
requests and we tried to restart it to no avail.  Then we shutdown our
processes so that nothing was using HBase and we shut down HBase and
brought it back up.  We waited a little bit, until hbase status
indicated that all the servers were back up.  We turned on our
processes and lo and behold, HBase is broken, getting
org.apache.hadoop.hbase.NotServingRegionException: Region is not
online: -ROOT-,,0
    at org.apache.hadoop.hbase.regionserver.HRegionServer.getRegion(HRegionServer.java:2319)
    at org.apache.hadoop.hbase.regionserver.HRegionServer.getClosestRowBefore(HRegionServer.java:1607)
    at sun.reflect.GeneratedMethodAccessor2.invoke(Unknown Source)
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:25)
    at java.lang.reflect.Method.invoke(Method.java:597)
    at org.apache.hadoop.hbase.ipc.HBaseRPC$Server.call(HBaseRPC.java:570)
    at org.apache.hadoop.hbase.ipc.HBaseServer$Handler.run(HBaseServer.java:1036)

And now we can't even shut it down.

Seems that Hbase is just too flaky to depend on for a serious system,
we've not had this type of problem to this degree with conventional DB
systems. Now that we are not saving that much data (we are using large
hdfs files for that) in Hbase, we are probably going to move back to a
conventional SQL system for our control data.  We just can't afford to
be constantly fighting problems like this.


Robert Gonzalez

Maxpoint Interactive

View raw message