[mirror-admin] [Fwd: Do overlapping release dates matter for our mirrors?]

[J.H.]

This particular issue came up, rather early on, in the oss-infra mailing 
list (a, hopefully, neutral ground for core infrastructure people across 
distributions, projects, etc to meet) and one thing that came out of it was:

http://www.google.com/calendar/embed?src=pnjc2h5nmtpb02jqgbafgj51u8%40group.calendar.google.com&ctz=America/Los_Angeles

Which we are trying to keep reasonably up to date, but it would be nice 
if we can get other to help us maintain it.

> Here are the questions as it relates to the mirrors that carry Fedora 
> and other distros:
> 
> 1) What is the minimum distance in days that we have to put between our 
> release and other distros?

Preferably a full week from another release.  Why a week?  It gives time 
for both the traffic from the previous release to completely flush, and 
gives time for mirrors to sync up for the next one.  Traditionally 
Fedora has taken about a week from bit-flip to "normal" levels of 
traffic to return, this seems to have dropped in the last few releases 
down into the 4day range.  Giving mirrors an extra couple of days to 
make sure they are synced and ready for the next release would be 
immensely helpful

> 2) Is it the release date that matters most or the date the staging starts?

Both.  While the release date is an obvious flood of traffic things like 
staging matter too.  If your a tier1 mirror for a distro and you've got 
a major release, like Fedora, and someone stages their bits in the 
middle of your release you are now trying to service the hordes getting 
the currently released software, and trying to service the TierNs that 
are trying to sync from you.

To the disk usage this won't actually look a lot different than if you 
had both released simultaneously.  The only "upside" is that there are 
fewer TierNs than there are people who want the new distro and they have 
  bigger pipes, so they have a tendency to clear fairly quickly.

> 3) On a scale of 1 to 5 (1 = doesn't matter, 3 = moderate impact, 5 = 
> critical, we'd be out of our minds not to) how important is it that we 
> bend our schedule to not land as close as we might be potentially 
> landing with the dates below (assuming we do not slip)?

I'll give this a 3-5 because it's a little dependent.  If Fedora is 
going against the release of something like Debian or Archlinux the 
timings in that situation aren't *THAT* bad, mainly because Fedora is 
going to overshadow the other release so much.

However if your close to something like Ubuntu, OpenSuse or even CentOS 
this drifts quickly up into the 5 range and I would have to even say my 
above answer with respect to having a weeks gap starts cutting it really 
close.  Putting two huge distro releases back-to-back is iffy at best, 
for the simple fact that your likely to have one of them slip and the 
slipage (as happened originally in Fedora 12 here) moved directly into 
the release window of another big distro.  Now this ended up, 
ultimately, working out this time around but it gave me a nigh heart 
attack and I was convinced I would literally kill off one of my US boxes 
during the release because of the extra loads.

I hate to say it but assuming there won't be slippage to your release, 
or slippage to someone elses, particularly if your Fedora, Ubuntu, 
OpenSuse, etc is probably a bad assumption to make.  Plan for some extra 
room around, but I wouldn't go overboard.  If a release slips by a week 
that's fine, but if it suddenly slips by several weeks it might need to 
be re-checked if the new release window is "safe".  This can be helped 
dramatically by checking with releases around your date and seeing if 
they are going to slip as things get closer.

> 
> I'm really looking for a scale to get a sense of the severity and not 
> subjective responses like "it will be bad" or "people will hate us."  I 
> want to know "how much."
> 
> 4) Our current options as I understand them are:
>     a) Overlap closely (see below)
>     b) Add two weeks to Fedora 13
> 
>    Is there an "option C" that we could implement or go with so as not 
> to have to add two weeks to our scheduled GA date if the consensus score 
> from #3 is high?
> 
> ------------------------------------------------------
> 
> Debian                2010-March (Freeze, not an actual release date)
> 

Is the Fedora date listed here right? 
http://fedoraproject.org/wiki/Schedule shows the release not on 
2010-04-27 but on 2010-05-11

> Fedora 13             2010-04-27 (since Fedora 8 every release has been 
> late >= 2 weeks)
> 
> Ubuntu 10.4           2010-04-29 (I'm told they have never slipped)
> 
> OpenSUSE              2010-05-05 (no idea on their "on time arrival 
> history")


- John 'Warthog9' Hawley

--