launchpad-dev team mailing list archive

[Robert Collins <robert.collins@xxxxxxxxxxxxx>]

On Tue, Jul 19, 2011 at 4:43 AM, Jeroen Vermeulen <jtv@xxxxxxxxxxxxx> wrote:
>> I agree that it makes it more conservative; how does that lead to more
>> downtime?
>
> Through more downtime deployments.  My understanding was that we will still
> have downtime when we can't deploy a patch within the time budget.

If the patch is slower than 15 seconds, it requires sign off from me /
francis. This is an escape valve for things that are too hard to make
fast - but we don't expect to use it.

> If that is the case, when we over-estimate the deployment time of a patch
> such that the estimate is over budget but actual deployment time is under
> budget, we will plan downtime that isn't actually needed.

This is true; however Francis and I can dig deeper (e.g. by asking you
and stub :P) about the likely prod performance of a db patch thats
slow.

15 seconds is room - with cold cache - to do quite a lot; its about
2ms to bring back a row cold, and with a write-back cache we shouldn't
need to block on writes, so - 2ms * 2 nodes -> about 3500 rows can be
updated.

-Rob