Sourceware service migration; downtime friday/weekend 28/29/30 nov

[Mark Wielaard]

Hi Maciej,

On Thu, Nov 27, 2025 at 06:07:17AM +0000, Maciej W. Rozycki wrote:
> On Tue, 25 Nov 2025, Mark Wielaard wrote:
> 
> > But DNS propagation also takes time and might make the new server
> > inaccessible for some time after that.
> 
>  The usual approach is to reduce REFRESH in SOA and TTL of the individual 
> records to single minutes enough in advance in preparation and then bring 
> the old values afterwards.  With the current value of 3600 there's still 
> enough time left to arrange for that.

Thanks. We realized the same and that is why we lowered the TTL to
just one hour. We expect the final data sync between the datacenter to
take ~3 to ~4 hours. During which all services on either side will be
shutdown. So it seems having a refresh of 3600 seconds is low enough
to get everybody the new addresses. We would update DNS first around
16:00 UTC, make sure it propagates to all secondaries, shutdown the
services, do the final data sync and bring everything up on the other
side around 20:00 UTC.

But there are also some other network and domain changes we have to
coordinate (PTR records, spf updates, dns updates for cygwin.com,
elfutils.org, dwarfstd.org, gcc.gnu.org, valgrind.org) where we rely
on others to do timely updates. I think we prepared for everything,
but there can always be hickups.

BTW. https://forge.sourceware.org will not be affected by this, so
this might be an ideal time to play with it if you haven't yet (make
sure to register with your @sourceware.org account/email first though,
because email will not work during the downtime *).

Cheers,

Mark

*) See also https://gcc.gnu.org/wiki/ForgeExperiment (the wiki will be
 down though during the datacenter move ...)