RPM: include criu dependencies #1256
Conversation
These dependencies were already included in the official Fedora package but were not included in rpm/crun.spec currently used by copr and soon to be used for the official Fedora package. Fixes: containers#1255 Signed-off-by: Lokesh Mandvekar <[email protected]>
|
/cc @Luap99 @martinpitt |
|
@martinpitt we can block this until you can verify containers/podman#19500 is green for checkpointing. |
|
@lsm5 I don't understand -- why do you want to block on this PR? It seems obviously correct to sync the upstream spec with what's in Fedora, as otherwise you have a regression on the next release? (That said, it seems that the Fedora gating tests -- which do seem to test checkpointing -- don't run upstream?) I would think that containers/podman#19500 can't become green until this lands and gets built in your COPR, can it? |
|
Interesting, it did become green. Hmm, that feels like your COPR receives both builds from PRs as well as builds from main branches -- isn't that horribly confusing and prone to conflicts and errors? Anyway, thanks for fixing! |
@martinpitt builds from PRs go to
The tmt tests pulled the packages built from |
Might be too late to change the copr names if that's still confusing. But let me know if I can update the description or something to make that difference clearer. We can take this conversation to matrix or something if necessary. |
|
Aah, makes sense, thanks @lsm5 for the explanation! I suppose that works because TF goes out of its way to install the precise NVR from a PR, out of this "pool" of (potentially) lots of different parallel PR builds. |
|
Something in the back of my head kept chewing on this.. A PR against podman should really not see this crun build from a PR, only the latest build from what landed on crun's main branch. That it did see it shows that a podman PR will run its tests against whichever crun PR happened to be opened/pushed last. In other words, sending a broken crun PR breaks all subsequent podman PRs. This is unexpected, unintuitive, unreproducible, and thus hard to debug or even realize. I strongly recommend you to reconfigure your COPRs to stop doing that. In other words, use packit's ephemeral coprs to build code from PRs, and configure it to build commits to main into your permanent COPR, for proper isolation of concurrent PRs. The proper way to do validate crun PRs is to run podman's tests in the crun PRs, in pretty much the same way as I proposed with running cockpit-podman's tests in podman PRs. (Just one level down in the dependency tree). (Happy to have a gmeet about this if you want to explore this more) |
@martinpitt good point. I think that should be doable. For now, apart from container-selinux, we are not running TMT tests anywhere apart from your cockpit tests PR for podman. And AFAICT, the copr change needs to be made in the PR itself (?) . Do you wanna try adding the copr change to your PR ? |
The problem is that both crun and podman currently do builds for PRs in the same rhcontainer/packit-builds COPR. So Runs for podman will necessarily see some (arbitrary) crun PR builds as well. That's what I meant with "use packit's ephemeral coprs to build code from PRs". So this needs to happen in both/all containers projects, the packit-builds COPR can be removed, and only the podman-next COPR remains. Does that make sense? |
These dependencies were already included in the official Fedora package but were not included in rpm/crun.spec currently used by copr and soon to be used for the official Fedora package.
Fixes: #1255