Billy Pearson wrote:
could it be from the global memcache limit
I set my hbase.hregion.memcache.flush.size =
hbase.regionserver.globalMemcacheLimit
So that memcache flushes are only as needed.
That would probably explain it. The global memcache limit will likely
be reached before a flush has happened. Under load, memcache will
usually exceed hbase.hregion.memcache.flush.size before the flush
actually kicks in. Looks like in your case the global memcache limit
and handler ran before the flush had completed.
Make hbase.regionserver.globalMemcacheLimit at least
hbase.hregion.memcache.flush.size * hbase.hregion.memcache.block.multiplier.
St.Ack
