Add a histogram to measure how long are we waiting to close...

Add a histogram to measure how long are we waiting to close
sockets with pending IO.

BUG=9258
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=14807

[wtc, 2009-04-28 22:15:50]

LGTM.  Thanks a lot for writing this CL.  I actually meant
to do this, but haven't got around to it.

Some comments below...

http://codereview.chromium.org/100129/diff/1/2
File net/base/tcp_client_socket_win.cc (right):

http://codereview.chromium.org/100129/diff/1/2#newcode160
Line 160: base::Time start = base::Time::Now();
Should we call base::TimeTicks::Now() instead?  We're
measuring time durations here, and TimeTicks is for that
purpose.

http://codereview.chromium.org/100129/diff/1/2#newcode168
Line 168: if (waiting_read_ || waiting_write_) {
More questions:

Should we create separate histograms for the waits on the
read and write events?  I think it's not necessary because
the log files from Sundar and Karen showed that they only
saw waiting_read_.

Should we put the whole thing, including the 'start'
variable, inside if (waiting_read_ || waiting_write_)?
This way we can avoid the cost of 'start' when no pending
IO needs to be canceled.

[rvargas (doing something else), 2009-04-28 22:34:08]

thanks.

http://codereview.chromium.org/100129/diff/1/2
File net/base/tcp_client_socket_win.cc (right):

http://codereview.chromium.org/100129/diff/1/2#newcode160
Line 160: base::Time start = base::Time::Now();
On 2009/04/28 22:15:50, wtc wrote:
> Should we call base::TimeTicks::Now() instead?  We're
> measuring time durations here, and TimeTicks is for that
> purpose.
sure.

http://codereview.chromium.org/100129/diff/1/2#newcode168
Line 168: if (waiting_read_ || waiting_write_) {
On 2009/04/28 22:15:50, wtc wrote:
> More questions:
> 
> Should we create separate histograms for the waits on the
> read and write events?  I think it's not necessary because
> the log files from Sundar and Karen showed that they only
> saw waiting_read_.

I don't think it makes a significant difference for us to have separate values
for reads and writes; just how long we block.

> 
> Should we put the whole thing, including the 'start'
> variable, inside if (waiting_read_ || waiting_write_)?
> This way we can avoid the cost of 'start' when no pending
> IO needs to be canceled.

done.

[wtc, 2009-04-28 22:38:52]

LGTM.  Thanks!