https://bz.apache.org/bugzilla/show_bug.cgi?id=69817

--- Comment #9 from Christopher Schultz <[email protected]> ---
(In reply to Mark Thomas from comment #7)
> I've only tested CLF so far but, for that case, switching to Instant
> improves performance (just for generating the access log entry) by about 3%.

:D

Does that include cached-formatted-timestamp up through the seconds value?

> I also think we have a bigger problem. We need to look at how we record
> request start time. Currently we subtract the duration from the current time
> to determine the start time but that doesn't account for the delay between
> when we calculate the request duration (in the CoyoteAdapter) and when we
> obtain the current time (in the AcessLogValve). Some crude testing suggests
> that offset is in the 2ms to 10ms range. There is no point providing
> microsecond precision for request start time if if is only accurate to
> within a few milliseconds.

:(

> I think we are going to need to record start time in the request as an
> Instant. I'll also measure what that does to performance as well.

+1

(In reply to Mark Thomas from comment #8)
> Good news. After porting to Instant all of the different access log
> timestamp formats are faster. Improvement ranges from ~0.9% to ~3.6% and
> that is with the change to Request to record the Instant at request start.

:D :D

> I am going to start merging the change but the plan is to do it in stages as
> much as I can.

I'm assuming you'll start by changing the way we record the start-of-request,
then commit small refactorings as you go? That had been my plan.

-- 
You are receiving this mail because:
You are the assignee for the bug.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to