On Mon, Mar 21, 2011 at 7:19 PM, Adam Phelps <[email protected]> wrote: > It looks like we've come up against a problem that looks identical to the > one you described. How did you go about manually inserting the two child > regions? >
You know the daughter regions because they should be listed when you look at the parent in .META. It should have info:splitA and info:splitB columns with the daughters listed. Take the encoded name of the daughters. Look in hdfs. Are the regions there? If so, insert regions named the same as those in info:splitA and info:splitB. Take the parent region for the template making the HRegionInfo. Poke around in bin/*rb scripts to see examples of reading HRegionInfo, amending it, and insert into .META. > Our current thought on fixing it is to use the hbase shell to remove the > entries for the child regions and rewrite the region's entry such that > OFFLINE => false and SPLIT => false (ie both currently true) but we're not > sure if thats a good solution. > You could change the split flag to true and then try onlining parent again (try calling assign). That might get it back up. Before doing this though, you should remove daughters from hdfs if they are present (see above for how to figure daughter regions -- or go to the regionserver that was hosting parent and find the split message, it'll list the daughters). St.Ack
