test: use remote coverage (#945)

We have already implemented this in the csi-driver. With this solution 
we can test more paths (e.g., metrics) in the code and get a more
accurate test coverage.

Author: lukasmetzner

.github/workflows/test_e2e.yml

@@ -75,7 +75,12 @@ jobs:
       - name: Run tests
         run: |
           source dev/files/env.sh
-          go test ./tests/e2e -tags e2e -v -race -timeout 60m -coverprofile=coverage.txt
+          go test ./tests/e2e -tags e2e -v -race -timeout 60m
+
+      - name: Fetch coverage data
+        run: |
+          source dev/files/env.sh
+          bash scripts/get-coverage-from-k8s.sh
 
       - name: Upload coverage reports to Codecov
         uses: codecov/codecov-action@v5
@@ -84,6 +89,7 @@ jobs:
           !startsWith(github.head_ref, 'releaser-pleaser--')
         with:
           token: ${{ secrets.CODECOV_TOKEN }}
+          files: ./coverage/coverage.txt
 
       - name: Dump logs & events
         if: always()

.gitignore

@@ -7,3 +7,4 @@ deploy/gen/
 hcloud-cloud-controller-manager
 *.tgz
 hack/.*
+coverage/

codecov.yaml

@@ -0,0 +1,3 @@
+ignore:
+  - "test"
+  - "internal/mocks"

dev/Dockerfile

@@ -11,8 +11,9 @@ RUN ls -al
 ARG CGO_ENABLED=0
 # `skaffold debug` sets SKAFFOLD_GO_GCFLAGS to disable compiler optimizations
 ARG SKAFFOLD_GO_GCFLAGS
+ARG GO_BUILDFLAGS
 ENV GOCACHE=/cache
-RUN --mount=type=cache,target="/cache" go build -gcflags="$SKAFFOLD_GO_GCFLAGS" -o hcloud-cloud-controller-manager.bin github.com/hetznercloud/hcloud-cloud-controller-manager
+RUN --mount=type=cache,target="/cache" go build -gcflags="$SKAFFOLD_GO_GCFLAGS" $GO_BUILDFLAGS -o hcloud-cloud-controller-manager.bin github.com/hetznercloud/hcloud-cloud-controller-manager
 
 FROM alpine:3.22
 

main.go

@@ -22,6 +22,9 @@ package main
 
 import (
 	"os"
+	"os/signal"
+	"runtime/coverage"
+	"syscall"
 
 	"github.com/spf13/pflag"
 	"k8s.io/apimachinery/pkg/util/wait"
@@ -43,6 +46,8 @@ func main() {
 		klog.Fatalf("unable to initialize command options: %v", err)
 	}
 
+	setupCoverageSignalHandler()
+
 	fss := cliflag.NamedFlagSets{}
 	command := app.NewCloudControllerManagerCommand(ccmOptions, cloudInitializer, app.DefaultInitFuncConstructors, names.CCMControllerAliases(), fss, wait.NeverStop)
 
@@ -74,3 +79,24 @@ func cloudInitializer(config *config.CompletedConfig) cloudprovider.Interface {
 	}
 	return cloud
 }
+
+func setupCoverageSignalHandler() {
+	coverDir, exists := os.LookupEnv("GOCOVERDIR")
+	if !exists {
+		return
+	}
+
+	c := make(chan os.Signal, 1)
+	signal.Notify(c, syscall.SIGUSR1)
+
+	go func() {
+		for {
+			<-c
+			klog.Info("Writing coverage profile")
+
+			if err := coverage.WriteCountersDir(coverDir); err != nil {
+				klog.Warning("failed to write coverage profile", "err", err)
+			}
+		}
+	}()
+}

scripts/get-coverage-from-k8s.sh

@@ -0,0 +1,33 @@
+#!/usr/bin/env bash
+set -ueo pipefail
+
+SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}")"  &> /dev/null && pwd)"
+COVERDIR="$SCRIPT_DIR/../coverage"
+
+# Create/clean coverage directory
+if [ -d "$COVERDIR" ]; then
+  echo "$COVERDIR already exists; cleaning"
+  rm -r "$COVERDIR"
+fi
+mkdir -p "$COVERDIR"
+
+signal_coverage_write() {
+  for i in "${@:2}"; do
+    echo "Sending USR1 signal to $i"
+    kubectl -n kube-system exec -t "$i" -- kill -USR1 1
+    sleep 0.5
+    echo "Pulling coverage from $i into $1"
+    kubectl cp -n kube-system "$i:/coverage" "$1"
+  done
+
+  go tool covdata textfmt -i "$1" -o "$1/coverage.txt"
+}
+
+PODS=$(
+  kubectl -n kube-system get pods \
+    --no-headers -o custom-columns=":metadata.name" \
+    -l app.kubernetes.io/instance=hccm
+)
+
+# shellcheck disable=SC2086
+signal_coverage_write "$COVERDIR" $PODS

skaffold.yaml

@@ -7,6 +7,8 @@ build:
     - image: docker.io/hetznercloud/hcloud-cloud-controller-manager
       docker:
         dockerfile: dev/Dockerfile
+        buildArgs:
+          GO_BUILDFLAGS: -covermode=atomic -coverpkg=github.com/hetznercloud/hcloud-cloud-controller-manager/...
   local:
     useBuildkit: true
   insecureRegistries:
@@ -24,6 +26,11 @@ manifests:
           # We circumvent this by co-locating the registry & HCCM, so it's always a local pull.
           affinity.podAffinity.requiredDuringSchedulingIgnoredDuringExecution[0].topologyKey: "kubernetes.io/hostname"
           affinity.podAffinity.requiredDuringSchedulingIgnoredDuringExecution[0].labelSelector.matchLabels.app: docker-registry
+          env.GOCOVERDIR.value: "/coverage"
+          extraVolumes[0].name: coverage
+          extraVolumes[0].emptyDir: {}
+          extraVolumeMounts[0].name: coverage
+          extraVolumeMounts[0].mountPath: "/coverage"
 
 profiles:
   # Clusters with Robot Servers do not support the native Routing functionality right now.