[pve-devel] [PATCH pve-container/qemu-server/pve-guest-common/pve-docs 0/1] Add pre/post-migrate hooks

[Thomas Lamprecht]

Am 27/09/2022 um 09:40 schrieb Stefan Hanreich:
> On 9/26/22 17:51, Thomas Lamprecht wrote:
>> Am 22/09/2022 um 16:13 schrieb Stefan Hanreich:
>>> I have decided to create distinct event types for source/target nodes, since
>>> otherwise the same script would run essentially twice on the source/target node.
>>> With distinct event types, the hooks should be more flexible in their usage.
>>
>> just make that a parameter, same flexibility but less cmd explosion and
>> complexity.
>>
>> Also, _iff_ (see reply  we keep the CLI entries for pct/qm it should just be
>> a single command there, any difference should be handled in the parameters;
>> it's internal after all and we want to avoid that there's more internal
>> commands then externals someday ;)
>>
>> Target and source should be part of the parameters on either call (pre/post,
>> src/target), it is relevant info and should be easily available. Some param
>> info like offline/online migration could be relevant too, but we can always
>> extend on that, so in that regard it can be fine to stop smaller, to avoid
>> going over board and having to keep all that info for backward compat. Any
>> parameter would need to be encoded in the example then.
>>
> 
> This is also an option I explored. One thing that I wasn't sure about was
> where the scripts run then? Does the pre event run on the source node and the
> post event on the target node? Dominik made an interesting point, that it
> might actually be desirable the other way around since you might want to do
> some setup code in the pre-hook, which would be nice on the target node. It
> might also be nice to run some cleanup code on the post-event which would be
> more suited to running on the source node.

IMO they both need to run on both, that's the point of a migration hookscript
prepare source & target for leaving/incomming guest and then cleanup source &
target after the migration happened (failed or not).

> 
> Do you think it would be smart to implement it as positional parameters to
> the script? Like 'qm pre-migrate <target> <source>' ? Since there are already
> ideas of adding additional contextual information, might it be smarter to
> expose all the additional info to the script in a dictionary? Not sure about
> this, but I could see us ending up with a situation where you have many
> additional variables only accessible by knowing their indexes. This has other
> downsides of course..

_if_ we stay with this approach I'd use as much non-positional params as
possible. This is an internal command so user facing UX doesn't matter, the
only thing that matters is having some flexibility for forward/backward
extendability/compat, and there fixed params are worse than none, iow. they
have no benefit for a internal, automatic called script.

> 
>> Some more general note, the example is better than nothing, but a nice
>> list/table directly in the docs would be really good to have. This could be
>> done upfront, before adding new hooks - best for now to duplicate it for
>> both CT and VM chapter (if sensible it can live in its own
>> guest-hook-list.adoc and just get included twice). Including the example
>> script as an appendix would be a nice touch too.
> 
> I looked at the documentation as well and found it a bit lacking, I thought
> it would be nice to overhaul this in an additional patch series, after all
> the hooks are merged. I figured it might be okay to silently add these
> features and document them afterwards in a subsequent patch. I will add a
> short documentation section for each hook to the documentation in the
> respective patches as well and then we can maybe overhaul/unify them
> afterwards.

IMO the status quo is best documented before extending it.