Merge pull request #3663 from telepresenceio/thallgren/login-and-make…

…-test Empty commit
telepresenceio · Aug 14, 2024 · bebecb1 · bebecb1
2 parents dd6a286 + e266326
commit bebecb1
Show file tree

Hide file tree

Showing 5 changed files with 75 additions and 40 deletions.
diff --git a/.github/workflows/dev.yaml b/.github/workflows/dev.yaml
@@ -31,17 +31,29 @@ jobs:
         run: |
           v=$(go run build-aux/genversion/main.go ${{github.run_id}})
           echo "TELEPRESENCE_VERSION=$v" >> "$GITHUB_ENV"
+          echo "TELEPRESENCE_SEMVER=${v#v}" >> "$GITHUB_ENV"
           echo "version=$v" >> $GITHUB_OUTPUT
           echo "semver=${v#v}" >> $GITHUB_OUTPUT
-      - name: Log in to registry
-        run: echo "${{ secrets.GITHUB_TOKEN }}" | docker login ghcr.io -u ${{ github.actor }} --password-stdin
       - name: Setup docker buildx
         uses: docker/setup-buildx-action@v3
         with:
           platforms: linux/amd64,linux/arm64
-      - name: Build cluster images
+      - name: Build image dependencies
+        run: make images-deps
+      - name: Log in to registry
+        run: echo "${{ secrets.GITHUB_TOKEN }}" | docker login ghcr.io -u ${{ github.actor }} --password-stdin
+      - name: Push client image
         run: |
-          make push-images-x
+          docker buildx build --platform=linux/amd64,linux/arm64 --build-arg TELEPRESENCE_VERSION=${{env.TELEPRESENCE_SEMVER}} \
+          --push --tag ${{env.TELEPRESENCE_REGISTRY}}/telepresence:${{env.TELEPRESENCE_SEMVER}} -f build-aux/docker/images/Dockerfile.client .
+      - name: Push tel2 image
+        run: |
+          docker buildx build --platform=linux/amd64,linux/arm64 --build-arg TELEPRESENCE_VERSION=${{env.TELEPRESENCE_SEMVER}} \
+          --push --tag ${{env.TELEPRESENCE_REGISTRY}}/tel2:${{env.TELEPRESENCE_SEMVER}} -f build-aux/docker/images/Dockerfile.traffic .
+      - name: Log out from registry
+        if: always()
+        run: docker logout
+
   run_tests:
     if: github.event.label.name == 'ok to test'
     strategy:

diff --git a/.github/workflows/release.yaml b/.github/workflows/release.yaml
@@ -57,16 +57,29 @@ jobs:
         name: install dependencies
       - name: set version
         shell: bash
-        run: echo "TELEPRESENCE_VERSION=${{ github.ref_name }}" >> $GITHUB_ENV
+        run: |
+          v=${{ github.ref_name }}
+          echo "TELEPRESENCE_VERSION=$v" >> "$GITHUB_ENV"
+          echo "TELEPRESENCE_SEMVER=${v#v}" >> "$GITHUB_ENV"
       - name: Setup docker buildx
         uses: docker/setup-buildx-action@v3
         with:
           platforms: linux/amd64,linux/arm64
+      - name: Build image dependencies
+        run: make images-deps
       - name: Log in to registry
         run: echo "${{ secrets.GITHUB_TOKEN }}" | docker login ghcr.io -u ${{ github.actor }} --password-stdin
-      - name: Upload Docker images
+      - name: Push client image
+        run: |
+          docker buildx build --platform=linux/amd64,linux/arm64 --build-arg TELEPRESENCE_VERSION=${{env.TELEPRESENCE_SEMVER}} \
+          --push --tag ${{env.TELEPRESENCE_REGISTRY}}/telepresence:${{env.TELEPRESENCE_SEMVER}} -f build-aux/docker/images/Dockerfile.client .
+      - name: Push tel2 image
         run: |
-          make push-images-x
+          docker buildx build --platform=linux/amd64,linux/arm64 --build-arg TELEPRESENCE_VERSION=${{env.TELEPRESENCE_SEMVER}} \
+          --push --tag ${{env.TELEPRESENCE_REGISTRY}}/tel2:${{env.TELEPRESENCE_SEMVER}} -f build-aux/docker/images/Dockerfile.traffic .
+      - name: Log out from registry
+        if: always()
+        run: docker logout
 
   publish-release:
     runs-on: ubuntu-latest

diff --git a/build-aux/main.mk b/build-aux/main.mk
@@ -220,53 +220,33 @@ TELEPRESENCE_SEMVER=$(patsubst v%,%,$(TELEPRESENCE_VERSION))
 CLIENT_IMAGE_FQN=$(TELEPRESENCE_REGISTRY)/telepresence:$(TELEPRESENCE_SEMVER)
 TEL2_IMAGE_FQN=$(TELEPRESENCE_REGISTRY)/tel2:$(TELEPRESENCE_SEMVER)
 
+.PHONY: images-deps
+images-deps: build-deps setup-build-dir
+
 .PHONY: tel2-image
-tel2-image: build-deps setup-build-dir
+tel2-image: images-deps
 	$(eval PLATFORM_ARG := $(if $(TELEPRESENCE_TEL2_IMAGE_PLATFORM), --platform=$(TELEPRESENCE_TEL2_IMAGE_PLATFORM),))
 	docker build $(PLATFORM_ARG) --target tel2 --tag tel2 --tag $(TEL2_IMAGE_FQN) -f build-aux/docker/images/Dockerfile.traffic .
 
-.PHONY: tel2-image-x
-tel2-image-x: build-deps setup-build-dir
-	docker buildx build --platform=linux/amd64,linux/arm64 --build-arg TELEPRESENCE_VERSION=$(TELEPRESENCE_VERSION) --cache-to type=local,dest=$(BUILDDIR)/docker-cache -f build-aux/docker/images/Dockerfile.traffic .
-
 .PHONY: client-image
-client-image: build-deps setup-build-dir
+client-image: images-deps
 	docker build --target telepresence --tag telepresence --tag $(CLIENT_IMAGE_FQN) -f build-aux/docker/images/Dockerfile.client .
 
-.PHONY: client-image-x
-client-image-x: build-deps setup-build-dir
-	docker buildx build --platform=linux/amd64,linux/arm64 --build-arg TELEPRESENCE_VERSION=$(TELEPRESENCE_VERSION) --cache-to type=local,dest=$(BUILDDIR)/docker-cache -f build-aux/docker/images/Dockerfile.client .
-
 .PHONY: push-tel2-image
 push-tel2-image: tel2-image ## (Build) Push the manager/agent container image to $(TELEPRESENCE_REGISTRY)
 	docker push $(TEL2_IMAGE_FQN)
 
-.PHONY: push-tel2-image-x
-push-tel2-image-x: build-deps setup-build-dir
-	docker buildx build --platform=linux/amd64,linux/arm64 --build-arg TELEPRESENCE_VERSION=$(TELEPRESENCE_VERSION) --cache-from type=local,src=$(BUILDDIR)/docker-cache -f build-aux/docker/images/Dockerfile.traffic --push --tag $(TEL2_IMAGE_FQN) .
-
-.PHONY: push-client-image
-push-client-image: client-image ## (Build) Push the client container image to $(TELEPRESENCE_REGISTRY)
-	docker push $(CLIENT_IMAGE_FQN)
-
-.PHONY: push-client-image-x
-push-client-image-x: build-deps setup-build-dir
-	docker buildx build --platform=linux/amd64,linux/arm64 --build-arg TELEPRESENCE_VERSION=$(TELEPRESENCE_VERSION) --cache-from type=local,src=$(BUILDDIR)/docker-cache -f build-aux/docker/images/Dockerfile.client --push --tag $(CLIENT_IMAGE_FQN) .
-
 .PHONY: save-tel2-image
 save-tel2-image: tel2-image
 	docker save $(TEL2_IMAGE_FQN) > $(BUILDDIR)/tel2-image.tar
 
-.PHONY: save-client-image
-save-client-image: client-image
-	docker save $(CLIENT_IMAGE_FQN) > $(BUILDDIR)/telepresence-image.tar
+.PHONY: push-client-image
+push-client-image: client-image ## (Build) Push the client container image to $(TELEPRESENCE_REGISTRY)
+	docker push $(CLIENT_IMAGE_FQN)
 
 .PHONY: push-images
 push-images: push-tel2-image push-client-image
 
-.PHONY: push-images-x
-push-images-x: push-tel2-image-x push-client-image-x
-
 .PHONY: clobber
 clobber: ## (Build) Remove all build artifacts and tools
 	rm -rf $(BUILDDIR)

diff --git a/cmd/traffic/cmd/manager/state/intercept.go b/cmd/traffic/cmd/manager/state/intercept.go
@@ -392,6 +392,8 @@ func (s *state) waitForAgent(ctx context.Context, name, namespace string, failed
 		return agent.Name == name && agent.Namespace == namespace
 	})
 	failedContainerRx := regexp.MustCompile(`restarting failed container (\S+) in pod ([0-9A-Za-z_-]+)_` + namespace)
+	mm := mutator.GetMap(ctx)
+
 	// fes collects events from the failedCreatedCh and is included in the error message in case
 	// the waitForAgent call times out.
 	var fes []*events.Event
@@ -451,8 +453,10 @@ func (s *state) waitForAgent(ctx context.Context, name, namespace string, failed
 				return status.Error(codes.Canceled, fmt.Sprintf("channel closed while waiting for agent %s.%s to arrive", name, namespace))
 			}
 			for _, a := range snapshot.State {
-				dlog.Debugf(ctx, "Agent %s.%s is ready", a.Name, a.Namespace)
-				return nil
+				if !mm.IsBlacklisted(a.PodName, a.Namespace) {
+					dlog.Debugf(ctx, "Agent %s.%s is ready", a.Name, a.Namespace)
+					return nil
+				}
 			}
 			dlog.Debugf(ctx, "Got empty snapshot while waiting for agent %s.%s", name, namespace)
 		case <-ctx.Done():

diff --git a/integration_test/itest/cluster.go b/integration_test/itest/cluster.go
@@ -19,6 +19,7 @@ import (
 	"time"
 	"unicode/utf8"
 
+	"github.com/cenkalti/backoff/v4"
 	"github.com/sirupsen/logrus"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -1014,11 +1015,22 @@ func Kubectl(ctx context.Context, namespace string, args ...string) error {
 	} else {
 		ks = args
 	}
-	return Run(ctx, "kubectl", ks...)
+	b := backoff.NewExponentialBackOff(
+		backoff.WithInitialInterval(2*time.Second),
+		backoff.WithMaxInterval(7*time.Second),
+		backoff.WithMaxElapsedTime(30*time.Second),
+	)
+	return backoff.Retry(func() error {
+		err := Run(ctx, "kubectl", ks...)
+		if err != nil && !strings.Contains(err.Error(), "(ServiceUnavailable)") {
+			err = backoff.Permanent(err)
+		}
+		return err
+	}, b)
 }
 
 // KubectlOut runs kubectl with the default context and the application namespace and returns its combined output.
-func KubectlOut(ctx context.Context, namespace string, args ...string) (string, error) {
+func KubectlOut(ctx context.Context, namespace string, args ...string) (out string, err error) {
 	getT(ctx).Helper()
 	var ks []string
 	if namespace != "" {
@@ -1027,7 +1039,20 @@ func KubectlOut(ctx context.Context, namespace string, args ...string) (string,
 	} else {
 		ks = args
 	}
-	return Output(ctx, "kubectl", ks...)
+	b := backoff.NewExponentialBackOff(
+		backoff.WithInitialInterval(2*time.Second),
+		backoff.WithMaxInterval(7*time.Second),
+		backoff.WithMaxElapsedTime(30*time.Second),
+	)
+	err = backoff.Retry(func() error {
+		var bErr error
+		out, bErr = Output(ctx, "kubectl", ks...)
+		if bErr != nil && !strings.Contains(bErr.Error(), "(ServiceUnavailable)") {
+			bErr = backoff.Permanent(bErr)
+		}
+		return bErr
+	}, b)
+	return out, err
 }
 
 func CreateNamespaces(ctx context.Context, namespaces ...string) {
@@ -1108,6 +1133,7 @@ func PingInterceptedEchoServer(ctx context.Context, svc, svcPort string, headers
 			dlog.Info(ctx, err)
 			return false
 		}
+		ips = iputil.UniqueSorted(ips)
 		if len(ips) != 1 {
 			dlog.Infof(ctx, "Lookup for %s returned %v", svc, ips)
 			return false