Repository navigation
ci(tests): give the two multi-peer qemu legs a bridge network - #5533
Merged
mudler merged 1 commit intoOct 20, 2026
Merged
Conversation
provider-decentralized-k8s and provider-qrcode-install are the only test-standard legs that need a second libp2p peer to be discovered: the first pairs the two VMs over edgevpn, the second pairs the runner with the VM over go-nodepair. The other three drive a single VM over the forwarded SSH port. QEMU user-mode networking does not let either pair see the other, so the rendezvous has to go out to the public DHT and come back. That works most of the time, which is why these two are the legs that redden intermittently on master while the rest stay green on the same run and the same image. reusable-qemu-test.yaml already carries everything needed to avoid it: a use_bridge_network input, a step that starts the libvirt default network and allows virbr0 in the QEMU bridge helper, and a suite that reads USE_BRIDGE_NETWORK and adds the second NIC. No caller ever set the input, so none of it has run. master.yaml and pr-full.yaml now set it for those two legs through a bridge key on the matrix entry. A bridge that cannot be set up is now an error rather than a logged warning. Before this the VM started on user-mode networking anyway and the spec failed ten minutes later at its own rendezvous timeout, naming discovery rather than the missing bridge. Fixes kairos-io#5494 Signed-off-by: Ettore Di Giacinto <mudler@kairos.io>
mudler
approved these changes
Oct 20, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes #5494
Summary
test-standard / provider-decentralized-k8sandtest-standard / provider-qrcode-installare the only two legs in thatmatrix that need a second libp2p peer to be discovered. The first pairs
the two VMs over edgevpn, the second pairs the runner with the VM over
go-nodepair. The other three legs drive a single VM from the runner over
the forwarded SSH port.
On QEMU user-mode networking neither pair can see the other, so the
rendezvous has to go out to the public DHT and come back. That works most
of the time, which is why these two are the legs that redden
intermittently while the three single-peer legs stay green on the same run
and the same image.
reusable-qemu-test.yamlalready carries the whole fix and nothing turnedit on:
use_bridge_networkinput;virbr0in the QEMU bridge helper and verifiesvirbr0exists;USE_BRIDGE_NETWORKto the suite, and the suite adds a secondNIC on
virbr0;libvirt-daemon-system,libvirt-clientsand
bridge-utilson every qemu job already, so nothing new is pulledin.
No caller in the repository ever set the input, so none of that has run.
master.yamlandpr-full.yamlnow set it for those two legs, through abridgekey on the matrix entry so the other three are untouched.The second change is the fail-open on the suite side. When
USE_BRIDGE_NETWORKwas set andensureBridgeNetworkfailed, the failurewas logged as a warning and the VM started on user-mode networking anyway.
The spec then failed ten minutes later at its own rendezvous timeout,
naming discovery rather than the missing bridge. It is now an error.
Closing the question #5494 was parked on
The issue asked for a re-run of
masterat4433f215to tell arendezvous flake apart from a regression in
ce57f7fc..4433f215, and Icould not trigger one. Four later
masterruns have answered it:0145342cce57f7fc4433f2155ac6129c581c17beTwo green runs at shas later than
4433f215rule the merge range out. The10-19 red has the same signature as the 10-15 one, a 600s
Eventuallywith a single peer for its whole length:
So it is the rendezvous, and it has been one all along.
How it was tested
A guard in
internal/ciguard/bridge_network_test.go, next to thepermissions guard, which reads the workflow YAML and never touches the
network. Three specs, each mutation-checked one at a time against this
branch:
bridge: truedropped fromprovider-decentralized-k8sinmaster.yamlpasses use_bridge_network to the reusable workflow, naming the leg and the fileuse_bridge_networkdropped from both callers'with:has at least one caller that sets it, which is the state this change foundUSE_BRIDGE_NETWORKto the suiteis an input reusable-qemu-test.yaml declares and consumesgo test ./internal/ciguard/...is green,golangci-lint runat theversion in
.golangci-versionreports 0 issues on that package and on thetestsmodule,go vet ./...intestsis clean, andactionlintisclean on both changed workflows.
What this does not prove
I cannot run a qemu leg here, so the bridge path itself has not been
exercised end to end.
It will not run on this pull request by itself either:
pr-full.yamlispull_request_targetonlabeled, and itsauthorizejob requires thelabeller to have write access, which
ci-robbotdoes not. So the firstreal exercise is either a maintainer adding
ci/fullhere, or the nextpush to
masterafter this merges.If
virbr0cannot be brought up on a hosted runner, theSetup bridge networkstep fails atip link show virbr0withinseconds, and the suite change makes the other failure mode loud as well,
so a wrong answer here costs one fast red rather than another 45-minute
timeout.
AI was used to write this. No human has read it yet.
π€ Generated with Claude Code