On Fri, Mar 4, 2011 at 8:51 PM, Otis Gospodnetic <[email protected]> wrote: > I'm wondering if writes are not being spread well, and if this server is just > being hotspotted. What is the best way to check this? >
Simplest way is watching the master UI. Refresh a bunch. Look at the counts for requests. Is this server carrying .META. If so, it'll have extra read load. You have ganglia or opentsdb setup? Thats good for ye ol' cluster view. Could be something wrong w/ the server or something else running on that box? > Also, the RS log on that machine is full of messages like these: > > 011-03-05 04:50:01,608 DEBUG org.apache.hadoop.hbase.io.hfile.LruBlockCache: > Block cache LRU eviction started; Attempting to free 20.02 MB of total=169.78 > MB > 2011-03-05 04:50:01,610 DEBUG org.apache.hadoop.hbase.io.hfile.LruBlockCache: > Block cache LRU eviction completed; freed=20.03 MB, total=149.75 MB, > single=93.37 MB, multi=75.47 MB, memory=0 KB > You doing lots of reads off this server? St.Ack
