batou_ext.oci: Add support for podman
#204
Conversation
Ma27
force-pushed
the
FC-37959-podman
branch
from
d83ac63 to
db0e430
Compare
FC-37959 We mostly want this for healthchecks that must pass before the unit is actually active.
Ma27
force-pushed
the
FC-37959-podman
branch
from
db0e430 to
3476623
Compare
|
Considering how considerable the differences between podman and docker are I'd almost think about splitting it up into two Components 🤔 that might however introduce a bunch of copy-pasted code. This would also eliminate the need to specifiy the backend for every single container. |
|
Given that
I'd argue that it would be fine to keep it as-is.
Hmm... I'd have to check whether the current approach allows podman & docker in parallel in the first place now that I think of it. You set the backend globally in NixOS. |
Yeah, backend is set globally, so doing it per-container here doesn't make sense. |
FC-37959 This component is used to indicate that a host uses podman instead of docker to run containers. NixOS only allows to set this globally, so it doesn't make sense to configure it per-container in here.
…ealthy Otherwise the startup seems to run into a timeout. FC-37959
* I don't really know how to get 'container' to work, but it's probably irrelevant anyways (app in container is supposed to `sd_notify(3)` there) * Turn off lingering for conmon, this seems pretty broken. Also make sure the dependencies are set up correctly for strategy healthy.
|
@PhilTaken the backend setting is now done via another component. |
| users.users."{{ component.user }}".linger = true; | ||
| # {% endif %} | ||
| # {% if component.sd_notify == "conmon" %} | ||
| users.users."{{ component.user }}".linger = false; |
There was a problem hiding this comment.
This means btw we can't have conmon & healthy for containers running as the same user, unfortunately :(
|
I like the podmanruntime component here, that simplifies usage a bunch. I still think theres almost too much complexity in the ocicontainer component here but I cannot think of a good solution. Did you test this with a reasonably complex deployment? |
FC-37959
We mostly want this for healthchecks that must pass before the unit is actually active.