tests/e2e: fix tests and failing github actions

.github/workflows/release.yaml

@@ -36,7 +36,13 @@ jobs:
         run: make test
       - name: Run E2E Tests
         timeout-minutes: 20
-        run: make e2e -e E2E_PARALLEL=3
+        # Quite often, tests were failing due to "too many open files" errors, so we're fixing this here
+        # Also, we want to see trace level logs if tests fail and the pipeline should exit on first error
+        run: |
+          sudo prlimit --pid $$ --nofile=1048576:1048576
+          sudo sysctl fs.inotify.max_user_instances=1280
+          sudo sysctl fs.inotify.max_user_watches=655360
+          make e2e -e E2E_LOG_LEVEL=trace -e E2E_FAIL_FAST=true
       # Builds
       - name: Test Platform Builds
         run: make build-cross

Makefile

@@ -55,6 +55,7 @@ E2E_KEEP ?=
 E2E_PARALLEL ?=
 E2E_DIND_VERSION ?=
 E2E_K3S_VERSION ?=
+E2E_FAIL_FAST ?=
 
 ########## Go Build Options ##########
 # Build targets
@@ -179,7 +180,7 @@ test:
 
 e2e:
 	@echo "Running e2e tests"
-	LOG_LEVEL="$(E2E_LOG_LEVEL)" E2E_INCLUDE="$(E2E_INCLUDE)" E2E_EXCLUDE="$(E2E_EXCLUDE)" E2E_EXTRA="$(E2E_EXTRA)" E2E_RUNNER_START_TIMEOUT=$(E2E_RUNNER_START_TIMEOUT) E2E_HELPER_IMAGE_TAG="$(E2E_HELPER_IMAGE_TAG)" E2E_KEEP="$(E2E_KEEP)" E2E_PARALLEL="$(E2E_PARALLEL)" E2E_DIND_VERSION="$(E2E_DIND_VERSION)" E2E_K3S_VERSION="$(E2E_K3S_VERSION)" tests/dind.sh "${K3D_IMAGE_TAG}"
+	LOG_LEVEL="$(E2E_LOG_LEVEL)" E2E_INCLUDE="$(E2E_INCLUDE)" E2E_EXCLUDE="$(E2E_EXCLUDE)" E2E_EXTRA="$(E2E_EXTRA)" E2E_RUNNER_START_TIMEOUT=$(E2E_RUNNER_START_TIMEOUT) E2E_HELPER_IMAGE_TAG="$(E2E_HELPER_IMAGE_TAG)" E2E_KEEP="$(E2E_KEEP)" E2E_PARALLEL="$(E2E_PARALLEL)" E2E_DIND_VERSION="$(E2E_DIND_VERSION)" E2E_K3S_VERSION="$(E2E_K3S_VERSION)" E2E_FAIL_FAST="$(E2E_FAIL_FAST)" tests/dind.sh "${K3D_IMAGE_TAG}"
 
 ci-tests: fmt check e2e
 

pkg/client/node.go

@@ -807,6 +807,7 @@ func NodeWaitForLogMessage(ctx context.Context, runtime runtimes.Runtime, node *
 			time.Sleep(500 * time.Millisecond)
 			continue
 		} else {
+			l.Log().Tracef("Non-fatal last log line in node %s: %s", node.Name, previousline)
 			// case 2: last log line we saw did not contain a fatal error, so we break the loop here and return a generic error
 			break
 		}

tests/common.sh

@@ -45,7 +45,10 @@ failed() {
   elif [[ -n "$LOG_FILE" ]]; then
     mv "$LOG_FILE" "$LOG_FILE.failed"
   fi
-  abort "test failed"
+  abort "$CURRENT_STAGE: test failed"
+  if [[ "$E2E_FAIL_FAST" == "true" ]]; then
+    exit 1
+  fi
 }
 
 passed() {
@@ -73,11 +76,16 @@ check_url() {
 check_clusters() {
   [ -n "$EXE" ] || abort "EXE is not defined"
   for c in "$@" ; do
-    $EXE kubeconfig merge "$c" --kubeconfig-switch-context || return 1
-    if kubectl cluster-info ; then
-      passed "cluster $c is reachable"
+    if $EXE kubeconfig merge "$c" --kubeconfig-switch-context; then
+      if kubectl cluster-info ; then
+        passed "cluster $c is reachable"
+      else
+        warn "could not obtain cluster info for $c. Kubeconfig:\n$(kubectl config view)"
+        docker ps -a
+        return 1
+      fi
     else
-      warn "could not obtain cluster info for $c. Kubeconfig:\n$(kubectl config view)"
+      warn "could not merge kubeconfig for $c."
       docker ps -a
       return 1
     fi

tests/dind.sh

@@ -15,6 +15,9 @@ RUNNER_START_TIMEOUT=${E2E_RUNNER_START_TIMEOUT:-10}
 # Override Docker-in-Docker version
 E2E_DIND_VERSION=${E2E_DIND_VERSION:-}
 
+# Fail on first error instead of waiting until all tests are done. Useful in CI.
+E2E_FAIL_FAST=${E2E_FAIL_FAST:-}
+
 ####################################################################################
 
 CURR_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" >/dev/null 2>&1 && pwd)"
@@ -50,6 +53,7 @@ k3de2e=$(docker run -d \
           -e EXE="$K3D_EXE" \
           -e CI="true" \
           -e LOG_LEVEL="$LOG_LEVEL" \
+          -e E2E_FAIL_FAST="$E2E_FAIL_FAST" \
           -e E2E_INCLUDE="$E2E_INCLUDE" \
           -e E2E_EXCLUDE="$E2E_EXCLUDE" \
           -e E2E_PARALLEL="$E2E_PARALLEL" \

tests/test_config_file_from_stdin.sh

@@ -28,22 +28,18 @@ fi
 
 export CURRENT_STAGE="Test | config-file-stdin | $K3S_IMAGE"
 
-configfileoriginal="$CURR_DIR/assets/config_test_simple.yaml"
-configfile="/tmp/config_test_simple-tmp_$(date -u +'%Y%m%dT%H%M%SZ').yaml"
 clustername="configteststdin"
 
-sed -E "s/^  name:.+/  name: $clustername/g" < "$configfileoriginal" > "$configfile" # replace cluster name in config file so we can use it in this script without running into override issues
-cat "$configfile"
 highlight "[START] ConfigTest $EXTRA_TITLE"
 
 info "Creating cluster $clustername..."
 
-cat <<EOF | $EXE cluster create "$clustername" --config=-
+cat <<EOF | $EXE cluster create --config=-
 apiVersion: k3d.io/v1alpha4
 kind: Simple
 metadata:
-  name: test
-servers: 3
+  name: $clustername
+servers: 1
 agents: 2
 #image: rancher/k3s:latest
 volumes:
@@ -99,8 +95,8 @@ sleep 5
 info "Checking that we have access to the cluster..."
 check_clusters "$clustername" || failed "error checking cluster"
 
-info "Checking that we have 5 nodes online..."
-check_multi_node "$clustername" 5 || failed "failed to verify number of nodes"
+info "Checking that we have 3 nodes online..."
+check_multi_node "$clustername" 3 || failed "failed to verify number of nodes"
 
 # 2. check some config settings
 
@@ -128,10 +124,16 @@ exec_in_node "k3d-$clustername-server-0" "cat /etc/rancher/k3s/registries.yaml"
 
 # Cleanup
 
-info "Deleting cluster $clustername (using config file)..."
-$EXE cluster delete --config "$configfile" --trace || failed "could not delete the cluster $clustername"
+info "Deleting cluster $clustername (using config file from stdin)..."
+cat <<EOF | $EXE cluster delete --config=-
+apiVersion: k3d.io/v1alpha4
+kind: Simple
+metadata:
+  name: $clustername
+EOF
+
+test $? -eq 0 || failed "could not delete cluster $clustername from stdin config"
 
-rm "$configfile"
 
 highlight "[DONE] ConfigTest $EXTRA_TITLE"