Re: [pve-devel] [PATCH manager 1/2] ceph: remove global pg bits setting

[Thomas Lamprecht <t.lamprecht@proxmox.com>]

Am 20/09/2023 um 15:42 schrieb Maximiliano Sandoval:
> Thomas Lamprecht <t.lamprecht@proxmox.com> writes:
> 
>> @maximiliano: how does ceph react if oe sets this currently?
>> I.e., does it silently accepts unknown config keys, or does it break
>> something?
> 
> Running
> 
>     ceph config set global osd_pg_bits 42
> 
> results in
> 
>     Error EINVAL: unrecognized config option 'osd_pg_bits'
> 
> Having them set on /etc/pve/ceph.conf on the other hand does not throw
> any kind of warning as far as I can tell, but I suspect it does not do
> anything either.

Meh, as we do the latter it means that cephs will eat that value and
just ignore it, so this is a question about how strict we want to keep
the API "unbroken".

Dominik is def. right that just removing the parameter from our API
would be a breaking change in the technically correct sense.

But, AFAICT, this was never exposed via the pveceph CLI command nor the
web UI, so it's not that far-fetched to assume that it has close to zero
use in the wild, but we surely cannot be certain about that (some user
might have added it to their automatic provision script and us remove it
the param now would suddenly cause that to fail after the upgrade).

> Please tell if I am missing something or if I should do a new version of
> the patch.

While I'm not a favor of keeping things around that (Very Highly
Probably™) nobody will miss anyway, here just keeping the parameter is
simply way too cheap to not do that. So, let's not go for my normal
route of ripping it out and see if somebody complains once it hits
no-subscription, to then react accordingly.

In summary, yes, please send a new version of this patch that keeps a
(reduced) "pg_bits" API parameter, drops setting the value (as that is a
NO-OP anyway) and maybe even adds a warning like:
"the 'pg_bits' is deprecated and will be removed in the future"