0.19.3 and TRUNK roll the commit log every hour. Every 5 minutes might be a bit much though if no edits, we do not roll the log so maybe it'd be ok? You can't set it on a per-table basis since its a regionserver-level configuration. The configuration is hbase.regionserver.logroll.period. Its not in hbase-default.xml. I thought the configuration too exotic but that may have been a wrong call.
St.Ack On Sat, May 30, 2009 at 2:40 PM, Billy Pearson <[email protected]>wrote: > I had this happen to me I created a region then ran some import jobs all > the clients got stock on could not complete file for hlog and I had to kill > -9 them all and reload the data the table was missing when I did that > because the meta did not flush > could we not add a manual force flush to meta when we create or alter a > table so that data can be flushed to disk? or maybe we can setup where meta > flushes more option then the normal regions would like 5 mins or so to cover > splits etc. are covered. > I guess this would not be a problem if we have append in hadoop but would > be nice to have a good state on meta. > > Billy > > > "Yabo-Arber Xu" <[email protected]> wrote in message > news:[email protected]... > > Hi there, >> >> I had a single-node cluster up and running. And yesterday the node crashed >> for unknown reason, and when I restart it, everything appears to work >> except >> that all the tables are LOST( UI says that there is no user tables)!! I >> checked the log file and didn't find any clue; while I found the tables >> files are still there on HDFS. >> >> Anybody has any clue? >> >> It's quite urgent. Any help will be really appreciated. >> >> Best, >> Arber >> >> > >
