Ted Dunning wrote:
That makes it wait, but I don't think it increases the urgency on the part
of the namenode.
As an interesting experiment, I had a cluster with lots of pending
replication to do that was happening slowly. Restarting the name node
caused the rate of replication to increase massively. The difference was
highly visible on the ganglia graph because the amount of I/O wait time on
the cluster increased to >15% from near zero.
I think we should file a jira on this. There is no reason for Namenode
replicate faster after restart than before, unless it is remembering
something that it should not. The issue Chris Kline reported earlier in
Jan was unresolved as well.. even after looking through multiple logs.
Ted, could you run 'dfsadmin -metasave' and attach relevant info to a
new jira? Any or all the logs help.
Raghu.
