refactor: adjust smoke tests to make them more robust in GHA (#318)

Signed-off-by: Matt Peterson <[email protected]>
Signed-off-by: Atanas Atanasov <[email protected]>
Co-authored-by: Atanas Atanasov <[email protected]>

.github/workflows/smoke-test.yaml

@@ -86,4 +86,9 @@ jobs:
         run: ./smoke-test.sh
 
       - name: Stop Smoke Tests Container
+        if: always()
         run: docker compose -p block-node stop
+
+      - name: Print Server Container Logs
+        if: always()
+        run: docker logs block-node-server

server/docker/docker-build.sh

@@ -5,12 +5,12 @@ if [[ $# -lt 1 ]]; then
   exit 1
 fi
 
-VERSION=${1}
+VERSION=$1
 
 echo "CREATING CONTAINER FOR VERSION ${VERSION}"
 echo "Using project directory: ${2}"
 echo
 
 # run docker build
 echo "Building container:"
-docker buildx build --load -t "block-node-server:${1}" --build-context distributions=../build/distributions --build-arg VERSION="${VERSION}" . || exit "${?}"
+docker buildx build --load -t "block-node-server:${VERSION}" --build-context distributions=../build/distributions --build-arg VERSION="${VERSION}" . || exit "${?}"

server/docker/docker-compose.yml

@@ -2,6 +2,7 @@ version: '3.8'
 
 services:
   block-node-server:
+    container_name: block-node-server
     image: block-node-server:${VERSION}
     env_file:
       - .env

server/src/test/resources/smoke-test.sh

@@ -2,7 +2,6 @@
 
 # Variable to track whether shutdown has been called
 SHUTDOWN_CALLED=0
-
 # Function to shut down the background processes
 shutdown() {
     if [[ $SHUTDOWN_CALLED -eq 1 ]]; then
@@ -42,20 +41,24 @@ shutdown() {
 # Trap any exit signal to ensure everything is cleaned up
 trap "shutdown; exit 1" ERR SIGINT SIGTERM
 
-# 1. Verify that the logs have the startup pattern (only if log file is provided)
-LOG_FILE=$1
-STARTUP_PATTERN="Block Node Server started at port"
+# 1. Call the endpoints /health/livez and /health/readyz
+SERVER_URL="http://localhost:8080"
+LIVENESS_ENDPOINT="/healthz/livez"
+READINESS_ENDPOINT="/healthz/readyz"
 
-if [[ -n "$LOG_FILE" ]]; then
-    if grep -q "$STARTUP_PATTERN" "$LOG_FILE"; then
-        echo "Startup pattern found in logs."
-    else
-        echo "Startup pattern not found in logs."
-        exit 1
-    fi
-else
-    echo "No log file provided, skipping startup pattern check."
+if ! curl -v --connect-timeout 5 --max-time 10 --retry 12 --retry-delay 5 --retry-max-time 100 --retry-all-errors $SERVER_URL$LIVENESS_ENDPOINT; then
+    echo "$LIVENESS_ENDPOINT failed."
+    shutdown
+    exit 1
 fi
+echo "$LIVENESS_ENDPOINT endpoint is healthy."
+
+if ! curl -v --connect-timeout 5 --max-time 10 --retry 12 --retry-delay 5 --retry-max-time 100 --retry-all-errors $SERVER_URL$READINESS_ENDPOINT; then
+    echo "$READINESS_ENDPOINT endpoint failed."
+    shutdown
+    exit 1
+fi
+echo "$READINESS_ENDPOINT endpoint is ready."
 
 # 2. Start the consumer script with parameters 1 1000 and save logs to consumer.log
 ./consumer.sh 1 1000 > consumer.log 2>&1 &
@@ -67,10 +70,10 @@ echo "Started consumer with PID $CONSUMER_PID, logging to consumer.log"
 PRODUCER_PID=$!
 echo "Started producer with PID $PRODUCER_PID, logging to producer.log"
 
-# Sleep time after starting the consumer
+# 4. Sleep time after starting the consumer and producer scripts
 sleep 5  # Adjust sleep time as needed
 
-# 4. Run the get-block script with parameter 1 and save logs to get-block.log
+# 5. Run the get-block script with parameter 1 and save logs to get-block.log
 if ! ./get-block.sh 1 > get-block.log 2>&1; then
     echo "get-block.sh failed."
     echo "get-block logs:"
@@ -80,25 +83,6 @@ if ! ./get-block.sh 1 > get-block.log 2>&1; then
 fi
 echo "get-block.sh executed successfully."
 
-# 5. Call the endpoints /health/livez and /health/readyz
-SERVER_URL="http://localhost:8080"
-LIVENESS_ENDPOINT="/healthz/livez"
-READINESS_ENDPOINT="/healthz/readyz"
-
-if ! curl -f $SERVER_URL$LIVENESS_ENDPOINT; then
-    echo "$LIVENESS_ENDPOINT failed."
-    shutdown
-    exit 1
-fi
-echo "$LIVENESS_ENDPOINT endpoint is healthy."
-
-if ! curl -f $SERVER_URL$READINESS_ENDPOINT; then
-    echo "$READINESS_ENDPOINT endpoint failed."
-    shutdown
-    exit 1
-fi
-echo "$READINESS_ENDPOINT endpoint is ready."
-
 # 6. Shut everything down
 shutdown