J-D, here is the meta scan file - https://docs.google.com/document/d/1_g2Ce20H65rukrEe8i9UWaW9wskaV34_96ElpOQDsYE/edit?hl=en_US
I solved this problem "Chain of regions in table AkStats is broken; edges does not contain 277808094:1314921600:daily:Volume" using combination of check_meta and hbck -fix. But every now and then i hit this problem where java client hits WRE but hbase shell works. an example of the above is - get 'AkStats', '26696569976:1317081600:weekly:AudEng' I have a suspicion on why this could be happening. Please confirm if I should be concerned about the following - 1) 5-10% of my rows are really large. ~2.5MB+. Remaining are smaller - few KB. 2) Before I write the large rows, I delete the row if it exists and then write a new one. The reason I do it this way is because some of the columns in the existing row don't apply any more. So I delete the whole row and rewrite it. Ofcourse, this happens every few hours for 1-2K rows only based on my current load. Does hbase scale well for row deletes? thanks On Tue, Sep 27, 2011 at 1:23 PM, Jean-Daniel Cryans <[email protected]>wrote: > Yes, please send the info I asked. > > About the hbck errors you had, this is usually fixed with -fix: > > Region > AkStats,277808094:1314921600:daily:Volume,1317052667861.1ecc871503cd827934c3a9077b44e52b. > not deployed on any region server. > > This is "probably" a region that wasn't cleaned up so it's not really a > problem: > > Region > file:/media/ephemeral0/hbase-data/AkStats/3520c0cfbedcf212084379c0e41e7839 > on HDFS, but not listed in META or deployed on any region server. > > This is a real problem: > > Chain of regions in table AkStats is broken; edges does not contain > 277808094:1314921600:daily:Volume > > J-D > > On Tue, Sep 27, 2011 at 1:00 PM, Vinod Gupta Tankala > <[email protected]> wrote: > > J-D, > > I was getting these errors even after restarting the client. So it > probably > > is not straightforward. > > Also, I was able to run a combination of check_meta.rb and hbck with > their > > fix options and restore some of the inconsistencies. i still have 4 > > inconsistencies left (earlier it was 7) but check_meta thinks .META. is > > fine. After i did this, I am not getting java client errors any more. Not > > sure how to explain that. > > > > Do you still want me to send the information you requested? may be you > can > > help with remaining inconsistencies. > > > > thanks > > vinod > > > > On Tue, Sep 27, 2011 at 11:05 AM, Jean-Daniel Cryans < > [email protected]>wrote: > > > >> Hi vinod, > >> > >> Yeah WREs are never fun, hopefully we can help you fixing it. > >> > >> First, about the difference when querying from the shell and your java > >> client. > >> > >> - Is it a long lived client? Did you restart it since you got the WREs? > >> - If not, this could just be due to the fact that it has a cache of > >> regions that's different from what a new client would see. > >> - If you did restart it, then I would have to think a bit more about > >> it to find the difference. > >> > >> Either way, it'd be nice to see what you're doing. We need; > >> > >> - A full dump of your .META. (in the shell: scan '.META.'), please > >> put this on a web server or a pastebin > >> - The keys you are trying to reach > >> - The exceptions you are seeing that contain the row keys and regions > >> it's trying to reach > >> - Another thing that would be nice is to have the output of when you > >> are reaching that row key from the shell, but with the shell started > >> with the -d option (will show a lot more debug info). > >> - The master log of the day the exception started happening. > >> > >> J-D > >> > >> On Tue, Sep 27, 2011 at 7:57 AM, Vinod Gupta Tankala > >> <[email protected]> wrote: > >> > I find this hard to believe. For the same row key, my java client is > >> > throwing wrong region exception. But I can query the same using hbase > >> shell. > >> > Im on 0.90.2 version. > >> > > >> > Also note that I have inconsistencies in my regions that I am still > >> trying > >> > to figure out. But regardless, the inconsistencies should impact both > >> > methods of querying similarly. right? > >> > > >> > thanks > >> > vinod > >> > > >> > > >
