Merge branch 'main' into release-3.1.2

Author: github-actions[bot]

template/v2/dirs/etc/sagemaker-ui/network_validation.sh

@@ -126,14 +126,18 @@ temp_dir=$(mktemp -d)
 # Launch all service API checks in parallel background jobs
 for service in "${!SERVICE_COMMANDS[@]}"; do
   {
-    # Run command with timeout, discard stdout/stderr
-    if timeout "${api_time_out_limit}s" bash -c "${SERVICE_COMMANDS[$service]}" > /dev/null 2>&1; then
+    output_file="$temp_dir/${service}_output"
+
+    # Run command with timeout
+    if timeout "${api_time_out_limit}s" bash -c "${SERVICE_COMMANDS[$service]}" > "$output_file" 2>&1; then
       # Success: write OK to temp file
       echo "OK" > "$temp_dir/$service"
     else
       # Get exit code to differentiate timeout or other errors
       exit_code=$?
-      if [ "$exit_code" -eq 124 ]; then
+      if grep -q "Could not connect to the endpoint URL" "$output_file"; then
+        echo "UNREACHABLE" > "$temp_dir/$service"
+      elif [ "$exit_code" -eq 124 ]; then
         # Timeout exit code
         echo "TIMEOUT" > "$temp_dir/$service"
       else
@@ -155,10 +159,13 @@ for service in "${!SERVICE_COMMANDS[@]}"; do
     if [[ "$result" == "TIMEOUT" ]]; then
       echo "$service API did NOT resolve within ${api_time_out_limit}s. Marking as unreachable."
       unreachable_services+=("$service")
+    elif [[ "$result" == "UNREACHABLE" ]]; then
+      echo "$service API failed to connect to the endpoint. Marking as unreachable."
+      unreachable_services+=("$service")
     elif [[ "$result" == "OK" ]]; then
       echo "$service API is reachable."
     else
-      echo "$service API returned an error (but not a timeout). Ignored for network check."
+      echo "$service API returned an error (but not a timeout or endpoint reachability failure). Ignored for network check."
     fi
   else
     echo "$service check did not produce a result file. Skipping."

template/v2/dirs/etc/sagemaker-ui/sagemaker_ui_post_startup.sh

@@ -241,7 +241,13 @@ if bash /etc/sagemaker-ui/network_validation.sh "$is_s3_storage_flag" "$network_
     # Read unreachable services from JSON file
     failed_services=$(jq -r '.UnreachableServices // empty' "$network_validation_file" || echo "")
     if [[ -n "$failed_services" ]]; then
-        error_message="$failed_services are unreachable. Please contact your admin."
+        # Count number of services by splitting on comma
+        IFS=',' read -ra failed_array <<< "$failed_services"
+        count=${#failed_array[@]}
+        verb="are"
+        [[ "$count" -eq 1 ]] && verb="is"
+
+        error_message="$failed_services $verb unreachable. Please contact your admin."
         # Example error message: Redshift Clusters, Athena, STS, Glue are unreachable. Please contact your admin.
         write_status_to_file "error" "$error_message"
         echo "$error_message"

template/v3/dirs/etc/sagemaker-ui/network_validation.sh

@@ -126,14 +126,18 @@ temp_dir=$(mktemp -d)
 # Launch all service API checks in parallel background jobs
 for service in "${!SERVICE_COMMANDS[@]}"; do
   {
-    # Run command with timeout, discard stdout/stderr
-    if timeout "${api_time_out_limit}s" bash -c "${SERVICE_COMMANDS[$service]}" > /dev/null 2>&1; then
+    output_file="$temp_dir/${service}_output"
+
+    # Run command with timeout
+    if timeout "${api_time_out_limit}s" bash -c "${SERVICE_COMMANDS[$service]}" > "$output_file" 2>&1; then
       # Success: write OK to temp file
       echo "OK" > "$temp_dir/$service"
     else
       # Get exit code to differentiate timeout or other errors
       exit_code=$?
-      if [ "$exit_code" -eq 124 ]; then
+      if grep -q "Could not connect to the endpoint URL" "$output_file"; then
+        echo "UNREACHABLE" > "$temp_dir/$service"
+      elif [ "$exit_code" -eq 124 ]; then
         # Timeout exit code
         echo "TIMEOUT" > "$temp_dir/$service"
       else
@@ -155,10 +159,13 @@ for service in "${!SERVICE_COMMANDS[@]}"; do
     if [[ "$result" == "TIMEOUT" ]]; then
       echo "$service API did NOT resolve within ${api_time_out_limit}s. Marking as unreachable."
       unreachable_services+=("$service")
+    elif [[ "$result" == "UNREACHABLE" ]]; then
+      echo "$service API failed to connect to the endpoint. Marking as unreachable."
+      unreachable_services+=("$service")
     elif [[ "$result" == "OK" ]]; then
       echo "$service API is reachable."
     else
-      echo "$service API returned an error (but not a timeout). Ignored for network check."
+      echo "$service API returned an error (but not a timeout or endpoint reachability failure). Ignored for network check."
     fi
   else
     echo "$service check did not produce a result file. Skipping."

template/v3/dirs/etc/sagemaker-ui/sagemaker_ui_post_startup.sh

@@ -241,7 +241,13 @@ if bash /etc/sagemaker-ui/network_validation.sh "$is_s3_storage_flag" "$network_
     # Read unreachable services from JSON file
     failed_services=$(jq -r '.UnreachableServices // empty' "$network_validation_file" || echo "")
     if [[ -n "$failed_services" ]]; then
-        error_message="$failed_services are unreachable. Please contact your admin."
+        # Count number of services by splitting on comma
+        IFS=',' read -ra failed_array <<< "$failed_services"
+        count=${#failed_array[@]}
+        verb="are"
+        [[ "$count" -eq 1 ]] && verb="is"
+
+        error_message="$failed_services $verb unreachable. Please contact your admin."
         # Example error message: Redshift Clusters, Athena, STS, Glue are unreachable. Please contact your admin.
         write_status_to_file "error" "$error_message"
         echo "$error_message"