[Act] Act services moving on a new server

[Sébastien Aperghis-Tramoni]

Salve J Nilsen wrote:

> Sébastien Aperghis-Tramoni said:
> >
> > Steps:
> > - shutdown all services on spectre
> > - dump all PostgreSQL databases
> > - rsync everything a last time from spectre to phantom
> > - reimport all PostgreSQL databases
> > - startup all services on phantom
> >
> > Executing these steps should roughly take an hour, given that
> > data have been rsync'd every 2 hours since weeks. But as said,
> > DNS propagation will make the disruption longer for people still
> > trying to reach spectre. This can be mitigated to some degree
> > by reducing the TTL of the records.
> >
> > Unless I oversaw something, migrating over the next days should
> > be an appropriate time.
> >
> > Opinions?
> 
> 0) Make a cronological task list, with ALL necessary steps written
> down. Even the minor ones!

Agreed.

> 1) Do the $NEW_TTL change right now, so people with $CURRENT_TTL can
> have time to update their DNS cache. Then wait with any migration
> until $CURRENT_TTL time has passed. :)

Yes, I didn't mention that in my previous mail because I already had
that in mind. The current TTL is currently 3h. I can reduce it to 1h.
Below usually is useless, because many ISP will cache the records
anyway.

> 2) Make a "this site is under maintainance" static webpage, that
> one can put up during the migration. Doesn't have to be more than
> a catch-all mod_rewrite rule pointing to a static webpage containing
> a few words of apology, a smiley and an ETA for when the sites 
> are expected to be up again.

Agreed.

> 3) IMHO, at least admins of currently active sites should get a
> separate notice mail telling about the impending doom, ETA,
> duration, etc. :)
> This is also so they can give notice to their users/staff as 
> they deem necessary.

Indeed, I didn't think of that.

> 4) Pick a downtime a few days *after* a conference, and as long
> as possible before the next conference, so signup disruption is
> minimized.

Yes. Given YAPC::NA is in roughly two months, I thought it was
far enough to avoid being an inconvenience. But given Todd's
answer, I guess it isn't. 


Todd E Rinaldo wrote:

> Yeah I'll be honest the timing on this really sucks for yapcna
> if there's really going to be an outage.
> 
> The other thing that you should have written down is what your
> back out plan is if something goes wrong

Another factor on which we don't have any grasp is when we will
loose the network on spectre. This is dependant on when the ISP
of Logilune decides to cut the cable.

So it is better to try and migrate softly while we still have
spectre running, instead of having to handle that as an emergency.
This makes for a kinda temporary rollback plan, but not a backup
plan per se.


-- 
Sébastien Aperghis-Tramoni

Close the world, txEn eht nepO.