Meeting Minutes - Office Hours for CTI - 2026-08-28

[Zack Weinberg]

On Thu, Sep 10, 2026, at 5:07 PM, Joseph Myers wrote:
> On Thu, 10 Sep 2026, Zack Weinberg wrote:
>
>> While I do think this is the best way, don't you see that it also applies to
>> the changes on the table?  In particular, *starting* a hosting switch by
>> moving -- not mirroring -- the two most critical services (git and email),
>> which are also the two services most difficult to move, and most difficult
>> to roll back if something goes wrong, strikes me as incredibly unwise,
>> regardless of all other considerations.
>
> I've argued for a design involving a separate DNS name for each (project, 
> service pair), which makes subsequent moves / rollback much more 
> straightforward than with everything using the single name 
> "sourceware.org".

That makes a lot of sense as a goal, but I don't see why it can't or
shouldn't be rolled out incrementally starting with the non-critical services.

>> I feel like you and Alexandre are imagining different things.  Are you aware
>> of changes made on a "sure why not, sounds like a good idea to me" basis
>> *to critical production services specifically*, rather than more experimental
>> things like the patch tracker and the wiki?
>
> There was at least one such past change to Bugzilla configuration (from 
> memory, at a time when one of Sourceware Bugzilla and GCC Bugzilla had 
> fully restricted account creation and the other allowed account creation 
> except from a blacklist of forbidden domains, someone saw the difference, 
> guessed it was accidental and made both follow the more restrictive 
> configuration, when actually the less restrictive one had been a 
> deliberate choice).

I think I remember that incident myself; but bugzilla account *creation*
can hardly be said to be a critical service.  If the sloppy change management
had blacked out the entire bug database, say, that would be more "critical
service failure" in my book.

zw