Harden MCP Gateway startup health check against transient port-binding delays

actions/setup/js/start_mcp_gateway.cjs

@@ -31,6 +31,7 @@ const { spawn, execSync } = require("child_process");
 const fs = require("fs");
 const http = require("http");
 const path = require("path");
+const { withRetry } = require("./error_recovery.cjs");
 
 // ---------------------------------------------------------------------------
 // Timing helpers
@@ -370,43 +371,59 @@ async function main() {
 
   core.info(`Health endpoint: ${healthUrl}`);
   core.info(`(Note: MCP_GATEWAY_DOMAIN is '${gatewayDomain}' for container access)`);
-  core.info("Retrying up to 120 times with 1s delay (120s total timeout)");
+  core.info("Retrying up to 120 times with exponential backoff (250ms to 1s, ~120s total timeout)");
   core.info("");
 
-  const maxRetries = 120;
+  const maxTotalAttempts = 120;
+  // withRetry's maxRetries excludes the initial attempt.
+  const maxRetryCount = maxTotalAttempts - 1;
+  const initialRetryDelayMs = 250;
   let httpCode = 0;
   let healthBody = "";
   let succeeded = false;
+  let attemptsMade = 0;
 
   core.info("=== Health Check Progress ===");
-  for (let attempt = 1; attempt <= maxRetries; attempt++) {
-    const elapsedSec = Math.floor((nowMs() - healthCheckStart) / 1000);
-    if (attempt % 10 === 1 || attempt === 1) {
-      core.info(`Attempt ${attempt}/${maxRetries} (${elapsedSec}s elapsed)...`);
-    }
-
-    try {
-      const res = await httpGet(healthUrl, 2000);
-      httpCode = res.statusCode;
-      healthBody = res.body;
-      if (httpCode === 200 && healthBody) {
-        core.info(`✓ Health check succeeded on attempt ${attempt} (${elapsedSec}s elapsed)`);
-        succeeded = true;
-        break;
-      }
-    } catch {
-      // Connection refused / timeout – retry
-    }
+  try {
+    await withRetry(
+      async () => {
+        // Counts total health-check attempts, including the final successful attempt.
+        attemptsMade += 1;
+        const elapsedSec = Math.floor((nowMs() - healthCheckStart) / 1000);
+        if (attemptsMade % 10 === 1 || attemptsMade === 1) {
+          core.info(`Attempt ${attemptsMade}/${maxTotalAttempts} (${elapsedSec}s elapsed)...`);
+        }
 
-    if (attempt < maxRetries) {
-      await sleep(1000);
-    }
+        const res = await httpGet(healthUrl, 2000);
+        httpCode = res.statusCode;
+        healthBody = res.body;
+        if (httpCode === 200 && healthBody) {
+          core.info(`✓ Health check succeeded on attempt ${attemptsMade} (${elapsedSec}s elapsed)`);
+          succeeded = true;
+          return;
+        }
+        throw new Error(`Health endpoint not ready (HTTP ${httpCode || 0})`);
+      },
+      {
+        maxRetries: maxRetryCount,
+        initialDelayMs: initialRetryDelayMs,
+        maxDelayMs: 1000,
+        backoffMultiplier: 2,
+        jitterMs: 0,
+        // Preserve previous loop behavior: retry any health-check failure until attempts are exhausted.
+        shouldRetry: () => true,
+      },
+      "MCP gateway health check"
+    );
+  } catch {
+    // Retry exhaustion is handled below using existing diagnostics.
   }
+
   core.info("=== End Health Check Progress ===");
   core.info("");
 
   core.info(`Final HTTP code: ${httpCode}`);
-  core.info(`Total attempts: ${maxRetries}`);
+  core.info(`Total attempts: ${attemptsMade}`);
   if (healthBody) {
     core.info(`Health response body: ${healthBody}`);
   } else {

actions/setup/sh/start_mcp_gateway.sh

@@ -209,7 +209,7 @@ HEALTH_CHECK_START=$(date +%s%3N)
 HEALTH_CHECK_HOST="localhost"
 echo "Health endpoint: http://${HEALTH_CHECK_HOST}:${MCP_GATEWAY_PORT}/health"
 echo "(Note: MCP_GATEWAY_DOMAIN is '${MCP_GATEWAY_DOMAIN}' for container access)"
-echo "Retrying up to 120 times with 1s delay (120s total timeout)"
+echo "Retrying up to 120 times with exponential backoff (250ms to 1s, ~120s total timeout)"
 echo ""
 
 # Check health endpoint using localhost (since we're running on the host)
@@ -219,7 +219,6 @@ echo ""
 set +e
 
 MAX_RETRIES=120
-RETRY_DELAY=1
 RETRY_COUNT=0
 HTTP_CODE=""
 HEALTH_RESPONSE=""
@@ -234,7 +233,6 @@ while [ $RETRY_COUNT -lt $MAX_RETRIES ]; do
   ELAPSED_MS=$(($(date +%s%3N) - HEALTH_CHECK_START))
   ELAPSED_SEC=$((ELAPSED_MS / 1000))
 
-  # Show progress every 10 retries or on first attempt
   if [ $((RETRY_COUNT % 10)) -eq 1 ] || [ $RETRY_COUNT -eq 1 ]; then
     echo "Attempt $RETRY_COUNT/$MAX_RETRIES (${ELAPSED_SEC}s elapsed)..."
   fi
@@ -255,7 +253,16 @@ while [ $RETRY_COUNT -lt $MAX_RETRIES ]; do
 
   # If this is not the last attempt, wait before retrying
   if [ $RETRY_COUNT -lt $MAX_RETRIES ]; then
-    sleep $RETRY_DELAY
+    # Exponential backoff with 1s cap:
+    # attempt 1 -> 0.25s, attempt 2 -> 0.5s, attempt 3+ -> 1s
+    if [ $RETRY_COUNT -eq 1 ]; then
+      RETRY_DELAY="0.25"
+    elif [ $RETRY_COUNT -eq 2 ]; then
+      RETRY_DELAY="0.5"
+    else
+      RETRY_DELAY="1"
+    fi
+    sleep "$RETRY_DELAY"
   fi
 done
 echo "=== End Health Check Progress ==="

actions/setup/sh/start_mcp_gateway_test.sh

@@ -281,6 +281,21 @@ test_validation_functions_exist() {
   else
     print_result "--network host flag validation missing" "FAIL"
   fi
+
+  # Check for health check retry/backoff logic
+  if grep -q "RETRY_COUNT -eq 1" "$SCRIPT_PATH" &&
+    grep -q "RETRY_COUNT -eq 2" "$SCRIPT_PATH" &&
+    grep -q "elif \[ \$RETRY_COUNT -eq 2 \]" "$SCRIPT_PATH" &&
+    grep -q "else" "$SCRIPT_PATH" &&
+    grep -q "RETRY_DELAY=\"0.25\"" "$SCRIPT_PATH" &&
+    grep -q "RETRY_DELAY=\"0.5\"" "$SCRIPT_PATH" &&
+    grep -q "RETRY_DELAY=\"1\"" "$SCRIPT_PATH" &&
+    grep -q "attempt 3+ -> 1s" "$SCRIPT_PATH" &&
+    grep -q "sleep \"\$RETRY_DELAY\"" "$SCRIPT_PATH"; then
+    print_result "Health check exponential backoff configuration exists" "PASS"
+  else
+    print_result "Health check exponential backoff configuration missing" "FAIL"
+  fi
 }
 
 # Run all tests