On Sun, Jan 2, 2011 at 6:04 PM, Wayne <[email protected]> wrote: > The LZO did not seem to work with the 1GB region size. It was causing > several minute pauses followed by 5 seconds of requests being processed and > then again 30+ second pauses (is this GC, compaction, or splits??...all > regions go to 0 requests but the Meta region which has a few hundred > requests).
Whats it say in the regionserver at this time? There are a few reasons for our blocking incoming writes; if we've not been compacting fast enough or if memory is full. You can mess with configs. to take on more before the barriers go up (IIRC, reading your configs., you'd upped the memstore size to match the bigger 1GB regions? If logs say blocking because too many storefiles, you could up the upper count...hbase.hstore.blockingStoreFiles). Once going back to the default region size the pauses seemed to > go away. I still see all nodes going to zero requests except the Meta table > region, but it only lasts for a few seconds at the most. We also had the max > open files problem (which has now been fixed), so don't know if that could > have caused that. > Max open files? Ulimit? St.Ack
