Repository navigation
Support runs in toolbox - #1
Merged
Merged
Conversation
A control node that is itself a container (toolbox, distrobox, a dev container) cannot run nested containers: any podman command would talk to the outer environment's store, and there is no root for dnf install of the host tools the qemu backend needs (qemu-img, an ISO tool). freeipa_env.container detects that situation (containerenv/dockerenv markers, $container, PID 1 environment) and offers the two helpers the later commits build on: prepare_tools(), which fetches missing Fedora RPMs with 'dnf download' (rootless), extracts them with rpm2archive|tar into a cache directory and exposes wrapper scripts setting LD_LIBRARY_PATH on PATH; and scratch_dir(), which keeps large temporary files (image pulls) off tmpfs /tmp. Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com>
The private NIC between a run's VMs is a qemu UDP multicast socket on 230.0.0.1. On a containerized control node (and on hosts that route multicast out a real interface without local delivery) no packet ever reaches the peer qemu: the VMs sit on 192.168.100.0/24 with ARP INCOMPLETE, and every multi-VM environment fails far from the cause. Probe local multicast delivery once at import (IP_MULTICAST_LOOP on a socketpair in the group); when it does not work, build the same L2 segment as a chain of TCP socket netdevs on loopback: one listening netdev per VM, each later VM connecting to the previous one. Same semantics, no multicast, no privileges needed. Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com>
_check_host refuses to start VMs when qemu-img or an ISO tool is not on PATH. Inside a container there is no root to install them, but freeipa_env.container can fetch the Fedora RPMs rootlessly and wrap them. Ask it before the tool check so the check passes on a container control node. Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com>
Inside a container podman must not run (it would use the outer environment's container store), so a containerdisk: image was unresolvable there. Pull the same image with skopeo into a temporary OCI directory instead and take /disk/disk.qcow2 out of its single gzipped layer. The cache file name stays the image id (the config digest, the same value podman reports), so a cache filled by either path is reused by the other. An unreadable outer /run/containers/<uid>/auth.json is worked around by pointing REGISTRY_AUTH_FILE at an empty file in the scratch dir: the public registries hosting containerdisks need no credentials. containers-storage: and oci-archive: refs keep using podman: only it can reach those stores. Also seed the fetched qemu-img wrapper when in a container, so the qcow2 validation step finds it. Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com>
Running podman on a control node that is itself a container would talk to (or try to create) a store belonging to the outer environment and can break it. Refuse the podman provider there with a message pointing at the nested qemu backend; FREEIPA_CI_ALLOW_PODMAN=1 opts back in when a podman socket is deliberately forwarded into the container. main() also calls container.prepare_tools() so every subcommand finds the fetched qemu-img/ISO wrappers on PATH. Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com>
The needs_podman skip only checked that a podman binary is on PATH, but inside a container the binary exists and cannot run (its store belongs to the outer environment): the tests failed in podman build instead of skipping. Probe podman info when in a container and skip on failure, with a reason pointing at freeipa_env.container. Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com>
Marker-based detection (containerenv file, $container, PID 1 environment), the rootless RPM fetch that produces a working and idempotent qemu-img wrapper, and the skopeo containerdisk resolution that must not run podman at all when the control node is a container, including the cache-hit and the bad-layer rejection. Signed-off-by: Alexander Bokovoy <abokovoy@redhat.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Handle VM deployments initiated inside toolbx environment.
Toolbx runs podman containers with $HOME mounted in. When podman tool runs inside this environment, it damages host's
/run/user/...podman configuration to the point that the host's podman stops recognizing running containers. Detect this situation and avoid running podman inside toolbx.Also, support different mechanism for communicating between VMs if multicast is not available.
Assisted-by: GLM-5.3