Yes. You should analyze HBase and hadoop logs See Stack's comments in HBASE-4058
On Fri, Jul 8, 2011 at 8:40 PM, Xu-Feng Mao <[email protected]> wrote: > Thanks Ted, I thought the patch to trunk is OK to 0.90 branch. > So we have to wait for the patch. > > Another question, our observation is much like HBASE-3872, we lost the file > on hdfs, and have a > hole in the meta. But any other way to confirm this problem? For example, > can we find some clues > in the logs? > > > On Sat, Jul 9, 2011 at 10:54 AM, Ted Yu <[email protected]> wrote: > >> Please direct question related to cdh to cdh-dev >> >> Patch of HBASE-3872 for 0.90 branch isn't posted yet. Do you have a patch >> already ? >> >> On Fri, Jul 8, 2011 at 7:43 PM, Xu-Feng Mao <[email protected]> wrote: >> >> > Hi, >> > >> > Since we've run into HBASE-3872 issue, we're considering upgrade a >> > production system from cdh3u0 to 0.90.3+HBASE-3872 patched. >> > Is it safe to just replace the hbase directory, and restart all the >> > regionservers and master? We have no chance to stop the whole cluster >> > together, can we restarted the regionservers partly, ie., can cdh3u0 and >> > 0.90.3 cooperate for a while? >> > >> > Thanks and regards >> > >> > Mao Xu-Feng >> > >> > >
