consul-container test: no splitting and on single runner (#17394)

.github/workflows/test-integrations.yml

@@ -343,51 +343,14 @@ jobs:
           DD_ENV: ci
         run: datadog-ci junit upload --service "$GITHUB_REPOSITORY" $TEST_RESULTS_DIR/results.xml
 
-  generate-compatibility-job-matrices:
-    needs: [setup]
-    runs-on: ${{ fromJSON(needs.setup.outputs.compute-small) }}
-    name: Generate Compatibility Job Matrices
-    outputs:
-      compatibility-matrix: ${{ steps.set-matrix.outputs.compatibility-matrix }}
-    steps:
-      - uses: actions/checkout@24cb9080177205b6e8c946b17badbe402adc938f # v3.4.0
-      - name: Generate Compatibility Job Matrix
-        id: set-matrix
-        env:
-          TOTAL_RUNNERS: 6
-          JQ_SLICER: '[ inputs ] | [_nwise(length / $runnercount | floor)]'
-        run: |
-          cd ./test/integration/consul-container
-          NUM_RUNNERS=$TOTAL_RUNNERS
-          NUM_DIRS=$(find ./test -mindepth 1 -maxdepth 2 -type d | wc -l)
-
-          if [ "$NUM_DIRS" -lt "$NUM_RUNNERS" ]; then
-            echo "TOTAL_RUNNERS is larger than the number of tests/packages to split."
-            NUM_RUNNERS=$((NUM_DIRS-1))
-          fi
-          # fix issue where test splitting calculation generates 1 more split than TOTAL_RUNNERS.
-          NUM_RUNNERS=$((NUM_RUNNERS-1))
-          {
-            echo -n "compatibility-matrix="
-            find ./test -maxdepth 2 -type d -print0 | xargs -0 -n 1 \
-              | grep -v util | grep -v upgrade \
-              | jq --raw-input --argjson runnercount "$NUM_RUNNERS" "$JQ_SLICER" \
-              | jq --compact-output 'map(join(" "))'
-          } >> "$GITHUB_OUTPUT"
-
   compatibility-integration-test:
     runs-on: ${{ fromJSON(needs.setup.outputs.compute-xl) }}
     needs:
       - setup
       - dev-build
-      - generate-compatibility-job-matrices
     permissions:
       id-token: write # NOTE: this permission is explicitly required for Vault auth.
       contents: read
-    strategy:
-      fail-fast: false
-      matrix:
-        test-cases: ${{ fromJSON(needs.generate-compatibility-job-matrices.outputs.compatibility-matrix) }}
     env:
       ENVOY_VERSION: "1.25.4"
     steps:
@@ -396,8 +359,10 @@ jobs:
         with:
           go-version-file: 'go.mod'
       - run: go env
-
-      # Build the consul:local image from the already built binary
+      - name: docker env
+        run: |
+          docker version
+          docker info
       - name: fetch binary
         uses: actions/download-artifact@9bc31d5ccc31df68ecc42ccf4149144866c47d8a # v3.0.2
         with:
@@ -405,7 +370,7 @@ jobs:
           path: .
       - name: restore mode+x
         run: chmod +x consul
-
+      # Build the consul:local image from the already built binary
       - name: Build consul:local image
         run: docker build -t ${{ env.CONSUL_LATEST_IMAGE_NAME }}:local -f ./build-support/docker/Consul-Dev.dockerfile .
       - name: Build consul-envoy:target-version image
@@ -426,21 +391,19 @@ jobs:
             mkdir -p "/tmp/test-results"
             cd ./test/integration/consul-container
             docker run --rm ${{ env.CONSUL_LATEST_IMAGE_NAME }}:local consul version
-            echo "Running $(sed 's,|, ,g' <<< "${{ matrix.test-cases }}" |wc -w) subtests"
-            # shellcheck disable=SC2001
-            sed 's, ,\n,g' <<< "${{ matrix.test-cases }}"
             go run gotest.tools/gotestsum@v${{env.GOTESTSUM_VERSION}} \
               --raw-command \
-              --format=short-verbose \
+              --format=standard-verbose \
               --debug \
               --rerun-fails=3 \
               -- \
               go test \
-              -p=4 \
+              -p=6 \
+              -parallel=4 \
               -tags "${{ env.GOTAGS }}" \
               -timeout=30m \
               -json \
-              ${{ matrix.test-cases }} \
+              `go list ./... | grep -v upgrade` \
               --target-image ${{ env.CONSUL_LATEST_IMAGE_NAME }} \
               --target-version local \
               --latest-image docker.mirror.hashicorp.services/${{ env.CONSUL_LATEST_IMAGE_NAME }} \
@@ -484,51 +447,15 @@ jobs:
           DD_ENV: ci
         run: datadog-ci junit upload --service "$GITHUB_REPOSITORY" $TEST_RESULTS_DIR/results.xml
   
-  generate-upgrade-job-matrices:
-    needs: [setup]
-    runs-on: ${{ fromJSON(needs.setup.outputs.compute-small) }}
-    name: Generate Upgrade Job Matrices
-    outputs:
-      upgrade-matrix: ${{ steps.set-matrix.outputs.upgrade-matrix }}
-    steps:
-      - uses: actions/checkout@24cb9080177205b6e8c946b17badbe402adc938f # v3.4.0
-      - uses: actions/setup-go@6edd4406fa81c3da01a34fa6f6343087c207a568 # v3.5.0
-        with:
-          go-version-file: 'go.mod'
-      - name: Generate Updgrade Job Matrix
-        id: set-matrix
-        env:
-          TOTAL_RUNNERS: 5
-          JQ_SLICER: '[ inputs ] | [_nwise(length / $runnercount | floor)]'
-        run: |
-          cd ./test/integration/consul-container/test/upgrade
-          NUM_RUNNERS=$TOTAL_RUNNERS
-          NUM_DIRS=$(go test ./...  -list=. -json | jq -r '.Output | select (. !=null) |  select(. | startswith("Test")) | gsub("[\\n\\t]"; "")' | wc -l)
-          
-          if [ "$NUM_DIRS" -lt "$NUM_RUNNERS" ]; then
-            echo "TOTAL_RUNNERS is larger than the number of tests/packages to split."
-            NUM_RUNNERS=$((NUM_DIRS-1))
-          fi
-          # fix issue where test splitting calculation generates 1 more split than TOTAL_RUNNERS.
-          NUM_RUNNERS=$((NUM_RUNNERS-1))
-          {
-            echo -n "upgrade-matrix="
-            go test ./...  -list=. -json | jq -r '.Output | select (. !=null) |  select(. | startswith("Test")) | gsub("[\\n\\t]"; "")' \
-            | jq --raw-input --argjson runnercount "$NUM_RUNNERS" "$JQ_SLICER" \
-            | jq --compact-output 'map(join("|"))'
-          } >> "$GITHUB_OUTPUT"
-  
   upgrade-integration-test:
-    runs-on: ${{ fromJSON(needs.setup.outputs.compute-large) }}
+    runs-on: ${{ fromJSON(needs.setup.outputs.compute-xl) }}
     needs:
       - setup
       - dev-build
-      - generate-upgrade-job-matrices
     strategy:
       fail-fast: false
       matrix:
         consul-version: [ "1.14", "1.15"]
-        test-cases: ${{ fromJSON(needs.generate-upgrade-job-matrices.outputs.upgrade-matrix) }}
     env:
       CONSUL_LATEST_VERSION: ${{ matrix.consul-version }}
       ENVOY_VERSION: "1.24.6"
@@ -576,9 +503,6 @@ jobs:
           mkdir -p "${{ env.TEST_RESULTS_DIR }}"
           cd ./test/integration/consul-container/test/upgrade
           docker run --rm ${{ env.CONSUL_LATEST_IMAGE_NAME }}:local consul version
-          echo "Running $(sed 's,|, ,g' <<< "${{ matrix.test-cases }}" |wc -w) subtests"
-          # shellcheck disable=SC2001
-          sed 's,|,\n,g' <<< "${{ matrix.test-cases }}"
           go run gotest.tools/gotestsum@v${{env.GOTESTSUM_VERSION}} \
             --raw-command \
             --format=short-verbose \
@@ -591,7 +515,6 @@ jobs:
             -tags "${{ env.GOTAGS }}" \
             -timeout=30m \
             -json ./... \
-            -run "${{ matrix.test-cases }}" \
             --target-image ${{ env.CONSUL_LATEST_IMAGE_NAME }} \
             --target-version local \
             --latest-image docker.mirror.hashicorp.services/${{ env.CONSUL_LATEST_IMAGE_NAME }} \
@@ -617,9 +540,7 @@ jobs:
     - vault-integration-test
     - generate-envoy-job-matrices
     - envoy-integration-test
-    - generate-compatibility-job-matrices
     - compatibility-integration-test
-    - generate-upgrade-job-matrices
     - upgrade-integration-test
     runs-on: ${{ fromJSON(needs.setup.outputs.compute-small) }}
     if: ${{ always() }}

test/integration/consul-container/libs/topology/peering_topology.go

@@ -32,6 +32,13 @@ type BuiltCluster struct {
 	Gateway   libservice.Service
 }
 
+type PeeringClusterSize struct {
+	AcceptingNumServers int
+	AcceptingNumClients int
+	DialingNumServers   int
+	DialingNumClients   int
+}
+
 // BasicPeeringTwoClustersSetup sets up a scenario for testing peering, which consists of
 //
 //   - an accepting cluster with 3 servers and 1 client agent. The client should be used to
@@ -46,11 +53,12 @@ func BasicPeeringTwoClustersSetup(
 	t *testing.T,
 	consulImage string,
 	consulVersion string,
+	pcs PeeringClusterSize,
 	peeringThroughMeshgateway bool,
 ) (*BuiltCluster, *BuiltCluster) {
 	acceptingCluster, acceptingCtx, acceptingClient := NewCluster(t, &ClusterConfig{
-		NumServers: 3,
-		NumClients: 1,
+		NumServers: pcs.AcceptingNumServers,
+		NumClients: pcs.AcceptingNumClients,
 		BuildOpts: &libcluster.BuildOptions{
 			Datacenter:           "dc1",
 			ConsulImageName:      consulImage,
@@ -61,8 +69,8 @@ func BasicPeeringTwoClustersSetup(
 	})
 
 	dialingCluster, dialingCtx, dialingClient := NewCluster(t, &ClusterConfig{
-		NumServers: 1,
-		NumClients: 1,
+		NumServers: pcs.DialingNumServers,
+		NumClients: pcs.DialingNumClients,
 		BuildOpts: &libcluster.BuildOptions{
 			Datacenter:           "dc2",
 			ConsulImageName:      consulImage,

test/integration/consul-container/test/gateways/http_route_test.go

@@ -502,8 +502,6 @@ func TestHTTPRouteParentRefChange(t *testing.T) {
 		t.Skip("too slow for testing.Short")
 	}
 
-	t.Parallel()
-
 	// infrastructure set up
 	address := "localhost"
 

test/integration/consul-container/test/peering/rotate_server_and_ca_then_fail_test.go

@@ -53,7 +53,14 @@ import (
 func TestPeering_RotateServerAndCAThenFail_(t *testing.T) {
 	t.Parallel()
 
-	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetTargetImageName(), utils.TargetVersion, false)
+	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetTargetImageName(), utils.TargetVersion,
+		libtopology.PeeringClusterSize{
+			AcceptingNumServers: 3,
+			AcceptingNumClients: 1,
+			DialingNumServers:   1,
+			DialingNumClients:   1,
+		},
+		false)
 	var (
 		acceptingCluster     = accepting.Cluster
 		dialingCluster       = dialing.Cluster

test/integration/consul-container/test/snapshot/snapshot_restore_test.go

@@ -82,7 +82,9 @@ func testSnapShotRestoreForLogStore(t *testing.T, logStore libcluster.LogStore)
 	libcluster.WaitForMembers(t, client2, 3)
 
 	// Restore the saved snapshot
-	require.NoError(t, client2.Snapshot().Restore(nil, snapshot))
+	retry.RunWith(libcluster.LongFailer(), t, func(r *retry.R) {
+		require.NoError(r, client2.Snapshot().Restore(nil, snapshot))
+	})
 
 	libcluster.WaitForLeader(t, cluster2, client2)
 

test/integration/consul-container/test/upgrade/basic/fullstopupgrade_test.go

@@ -32,8 +32,6 @@ var (
 
 // Test upgrade a cluster of latest version to the target version
 func TestStandardUpgradeToTarget_fromLatest(t *testing.T) {
-	t.Parallel()
-
 	tcs = append(tcs,
 		testcase{
 			// Use the case of "1.12.3" ==> "1.13.0" to verify the test can

test/integration/consul-container/test/upgrade/peering/peering_control_plane_mgw_test.go

@@ -27,7 +27,14 @@ import (
 func TestPeering_ControlPlaneMGW(t *testing.T) {
 	t.Parallel()
 
-	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetLatestImageName(), utils.LatestVersion, true)
+	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetLatestImageName(), utils.LatestVersion,
+		libtopology.PeeringClusterSize{
+			AcceptingNumServers: 1,
+			AcceptingNumClients: 1,
+			DialingNumServers:   1,
+			DialingNumClients:   1,
+		},
+		true)
 	var (
 		acceptingCluster = accepting.Cluster
 		dialingCluster   = dialing.Cluster

test/integration/consul-container/test/upgrade/peering/peering_http_test.go

@@ -20,14 +20,28 @@ import (
 
 func TestPeering_Basic(t *testing.T) {
 	t.Parallel()
-	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetLatestImageName(), utils.LatestVersion, false)
+	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetLatestImageName(), utils.LatestVersion,
+		libtopology.PeeringClusterSize{
+			AcceptingNumServers: 1,
+			AcceptingNumClients: 1,
+			DialingNumServers:   1,
+			DialingNumClients:   1,
+		},
+		false)
 	peeringUpgrade(t, accepting, dialing, utils.TargetVersion)
 	peeringPostUpgradeValidation(t, dialing)
 }
 
 func TestPeering_HTTPRouter(t *testing.T) {
 	t.Parallel()
-	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetLatestImageName(), utils.LatestVersion, false)
+	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetLatestImageName(), utils.LatestVersion,
+		libtopology.PeeringClusterSize{
+			AcceptingNumServers: 1,
+			AcceptingNumClients: 1,
+			DialingNumServers:   1,
+			DialingNumClients:   1,
+		},
+		false)
 	acceptingCluster := accepting.Cluster
 
 	// Create a second static-server at the client agent of accepting cluster and
@@ -90,7 +104,14 @@ func TestPeering_HTTPRouter(t *testing.T) {
 func TestPeering_HTTPResolverAndFailover(t *testing.T) {
 	t.Parallel()
 
-	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetLatestImageName(), utils.LatestVersion, false)
+	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetLatestImageName(), utils.LatestVersion,
+		libtopology.PeeringClusterSize{
+			AcceptingNumServers: 1,
+			AcceptingNumClients: 1,
+			DialingNumServers:   1,
+			DialingNumClients:   1,
+		},
+		false)
 	dialingCluster := dialing.Cluster
 
 	require.NoError(t, dialingCluster.ConfigEntryWrite(&api.ProxyConfigEntry{
@@ -183,7 +204,14 @@ func TestPeering_HTTPResolverAndFailover(t *testing.T) {
 func TestPeering_HTTPResolverAndSplitter(t *testing.T) {
 	t.Parallel()
 
-	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetLatestImageName(), utils.LatestVersion, false)
+	accepting, dialing := libtopology.BasicPeeringTwoClustersSetup(t, utils.GetLatestImageName(), utils.LatestVersion,
+		libtopology.PeeringClusterSize{
+			AcceptingNumServers: 1,
+			AcceptingNumClients: 1,
+			DialingNumServers:   1,
+			DialingNumClients:   1,
+		},
+		false)
 	dialingCluster := dialing.Cluster
 
 	require.NoError(t, dialingCluster.ConfigEntryWrite(&api.ProxyConfigEntry{