launchpad-dev team mailing list archive

[Robert Collins <robert.collins@xxxxxxxxxxxxx>]

On Thu, Mar 3, 2011 at 10:12 PM, John Arbash Meinel
<john@xxxxxxxxxxxxxxxxx> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Seems like something you could push to failure, and then back off, if
> you really wanted to push it. 1s increments per day/week, then backoff
> when it starts increasing significantly, and then focus on performance
> of remaining timeouts. I just have a strong suspicion that 99.9% of your
> responses are way under the timeout, so you have a huge gap between
> where the threshold is, and where it has an actual effect. (The requests
> that fail today would be taking 20+s, the requests that succeed are down
> in the <10s range. Notching from 15 to 14 certainly doesn't seem like it
> would do much, which is certainly what you saw.)

https://devpad.canonical.com/~lpqateam/ppr/lpnet/latest-daily-timeout-candidates.html
(internal link - sorry) - is what I use to assess impact. This is all
pages over 10 seconds for their 99th percentile.

The drop of 14 to 13 *added* a bunch of timeouts, but we had *also
fixed a bunch*, so the absolute count stayed flat - which is good:
we're at a sweet spot: a higher timeout and we're being wasteful,
which leads to long lived locks, more contention etc. A lower timeout
and the impact is felt by more users.

I want us to ride this wave all the way down: not too hot, not to
cold, Just Right.

-Rob