You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@zookeeper.apache.org by Vishal Kher <vi...@gmail.com> on 2011/04/01 23:46:32 UTC
Recording time taken for disk IO
Hi,
I was wondering if there is a way to measure the disk IO latency at ZK
server for reads/writes performed to the transaction log and snapshots. In
case ZK
is experiencing higher transaction latencies it would be useful to
verify whether it is due to disk or network.
We need to add this for debugging our application since we do not have
ZK logs on dedicated disks. If the disk IO is not currently monitored, I
was thinking of extending the streams used in FiLeSnap.java and
FileTxnLog.java, and measure the time taken to do the IO.
Does this sound like a good place to measure the latency?
Will it useful to have this in the tree?
Thanks,
-Vishal
Re: Recording time taken for disk IO
Posted by Benjamin Reed <br...@apache.org>.
measuring these latencies and recording them in the stats structure
would be great! i don't think we need to measure the time to take
snapshots though since they happen outside of the critical path and in
the background.
ben
On Fri, Apr 1, 2011 at 2:46 PM, Vishal Kher <vi...@gmail.com> wrote:
> Hi,
>
> I was wondering if there is a way to measure the disk IO latency at ZK
> server for reads/writes performed to the transaction log and snapshots. In
> case ZK
> is experiencing higher transaction latencies it would be useful to
> verify whether it is due to disk or network.
>
> We need to add this for debugging our application since we do not have
> ZK logs on dedicated disks. If the disk IO is not currently monitored, I
> was thinking of extending the streams used in FiLeSnap.java and
> FileTxnLog.java, and measure the time taken to do the IO.
> Does this sound like a good place to measure the latency?
> Will it useful to have this in the tree?
>
> Thanks,
> -Vishal
>