Re: [RFC PATCH datacenter-manager 1/4] server: connection: multi client: use correct client error for retrying

[Dominik Csapak <d.csapak@proxmox.com>]

On 6/3/26 1:48 PM, Thomas Lamprecht wrote:
> Am 01.06.26 um 09:19 schrieb Dominik Csapak:
>>
>> nice thanks, but after reading the commits wouldn't this now not also
>> retry the api call if it was a client one?
>>
>> it the first node returns a 'Client' error for any api call,
>> it will still retry on the second node, etc. ?
>>
>> the new code only does not retry the first node if it was a client
>> error?
> 
> yeah, you're right, this was still not completely right, oof tracing
> the error behavior of the client libs and these edge cases here are
> really not that trivial... I pushed another follow-up, would be glad
> if you could cross-check that too for sanity:
> 
> https://git.proxmox.com/?p=proxmox-datacenter-manager.git;a=commitdiff;h=c713876671f88d80bafdd3a1e7263644b4559da2
> 


I took a short look and it seem mostly OK FWICT

One thing that came to my mind though:

I think our API is not fully compliant with what the HTTP
spec defines as idempotent though:

from https://datatracker.ietf.org/doc/html/rfc7231#section-4.2.2

---
  Of the request methods
    defined by this specification, PUT, DELETE, and safe request methods
    are idempotent.
---

and

---
  Of the request methods defined by this specification, the GET, HEAD,
    OPTIONS, and TRACE methods are defined to be safe.
---

so of the methods we use in pve/pbs, only POST is not idempotent per
the spec, but e.g. we can use 'PUT' for making non-idempotent changes to 
a vm config (for example adding a disk to a VM, this will create
a disk and make any in the same slot unused)

I'm not super sure if we can get an client error here in a way
that would process the request with side effects on pve side and still
trigger a  retry on the pdm side, so I doubt this will make a difference
in practice. Still wanted to mention it.