CD daemon: send SIGUSR1 to IMEX daemon on node updates

[jgehrcke]

Resolves #581.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[klueska]

Suggested change

	// TODO: review skipping this if the new set of IP addresses only
	// strictly removes addresses compared to the old set (then we don't
	// need to force the daemon to re-resolve & re-connect).
	if updated && !fresh {
	// Actively ask the IMEX daemon to re-read its config and to
	// re-connect to its peers (involving DNS name re-resolution).
	klog.Infof("updated DNS/IP mapping, old process: send SIGUSR1")
	if err := processManager.Signal(syscall.SIGUSR1); err != nil {
	// Only log (ignore this error for now: if the process went
	// away unexpectedly, the process manager will handle that.
	// If any other error resulted in bad signal delivery, we
	// may get away with it).
	klog.Errorf("failed to send SIGUSR1 to child process: %s", err)
	}
	}
	// TODO: review skipping this if the new set of IP addresses only
	// strictly removes addresses compared to the old set (then we don't
	// need to force the daemon to re-resolve & re-connect).
	if !updated || fresh {
	return nil
	}
	// Actively ask the IMEX daemon to re-read its config and to
	// re-connect to its peers (involving DNS name re-resolution).
	klog.Infof("updated DNS/IP mapping, old process: send SIGUSR1")
	if err := processManager.Signal(syscall.SIGUSR1); err != nil {
	// Only log (ignore this error for now: if the process went
	// away unexpectedly, the process manager will handle that.
	// If any other error resulted in bad signal delivery, we
	// may get away with it).
	klog.Errorf("failed to send SIGUSR1 to child process: %s", err)
	}

[jgehrcke]

Updated one more time to not do return nil but continue instead in the early exit.

I will also test this patch again and report back -- it's too easy to make mistakes during seemingly innocent changes, of course.

[jgehrcke]

Tested that last state, looking OK:

• ✔️ three nvbandwidth-based fault injection tests
• ✔️ make bats (see below)

tests.bats
 ✓ test VERSION_W_COMMIT, VERSION_GHCR_CHART, VERSION [65]
 ✓ confirm no kubelet plugin pods running [68]
 ✓ helm-install deployments/helm/nvidia-dra-driver-gpu/25.8.0-dev [5205]
 ✓ helm list: validate output [96]
 ✓ get crd computedomains.resource.nvidia.com [62]
 ✓ wait for kubelet plugin pods READY [264]
 ✓ wait for controller pod READY [162]
 ✓ validate CD controller container image spec [63]
 ✓ IMEX channel injection (single) [7104]
 ✓ IMEX channel injection (all) [9987]
 ✓ NodePrepareResources: catch unknown field in opaque cfg in ResourceClaim [3132]
 ✓ nickelpie (NCCL send/recv/broadcast, 2 pods, 2 nodes, small payload) [11552]
 ✓ nvbandwidth (2 nodes, 2 GPUs each) [19673]
 ✓ downgrade: current-dev -> last-stable [27903]
 ✓ upgrade: wipe-state, install-last-stable, upgrade-to-current-dev [29114]

15 tests, 0 failures in 115 seconds

Kevin reviewed and verbally approved -- merging now, to facilitate branch wrestling.