launchpad-dev team mailing list archive

[Robert Collins <robertc@xxxxxxxxxxxxxxxxx>]

On Mon, Apr 4, 2011 at 5:54 PM, John Arbash Meinel
<john@xxxxxxxxxxxxxxxxx> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 4/4/2011 6:49 AM, Robert Collins wrote:
>> We've merged rev 10381 (the highest qa'd revision) of db-stable to
>> devel and its wending its way through buildbot now.
>>
>> There were two revisions - a publisher change and comment hiding for
>> questions - that were in db-stable but not qa'd. They have *not* been
>> included.
>>
>> However, as neither contain DB changes, they can both be landed again
>> on devel once we qa past the db-stable merge; and then qa'd etc and
>> deployed subsequent to the complete downtime deploy.
>>
>> As part of making the downtime more reliable we're going to make the
>> db downtime be precisely that: we'll deploy the db changing revision
>> *only* during the downtime window, and do a regular nodowntime deploy
>> subsequent to that. If you are changing a service which isn't part of
>> the nodowntime set, it either needs to go only through db-devel, or
>> organise a *partial downtime* window for it via mrevell. My strong
>> preference is *partial downtime* because the DB downtime deploys are
>> the single most risky event we have: making that go really well is
>> crucial for decreasing the duration of complete downtime.
>>
>> Cheers,
>> Rob
>>
>> https://dev.launchpad.net/Downtime for the definitions of complete and
>> partial downtime.
>
> Should it still say that we only do scheduled complete downtime once a
> month?  I know you scheduled downtime just a bit ago, but I don't know
> whether that was partial or complete.

We have this monthly scheduled complete downtime coming up on
Wednesday; we had partial downtime today to replace the old zope FTP
service for uploads of source packages with the shiny new twisted
service.

[yes, once a month complete scheduled downtime is correct for now]

-Rob