Allow more time for startup to fix jenkins failures (#2903)

Author: tballison

tika-pipes/tika-pipes-core/src/main/java/org/apache/tika/pipes/core/PipesClient.java

@@ -304,8 +304,10 @@ private void reconnect() throws InterruptedException, IOException, TimeoutExcept
         int port = serverManager.getPort();
         LOG.debug("pipesClientId={}: connecting to server", pipesClientId);
 
-        // Connect to server
-        Socket socket = serverManager.connect((int) pipesConfig.getSocketTimeoutMs());
+        // Connect to server. Use the generous startup timeout as the read SO_TIMEOUT so the
+        // server's post-connect initialization and READY handshake aren't bounded by the
+        // (possibly tight) per-request socketTimeoutMs.
+        Socket socket = serverManager.connect((int) pipesConfig.getStartupTimeoutMs());
 
         synchronized (connectionLock) {
             connectionTuple = new ConnectionTuple(socket,
@@ -314,6 +316,8 @@ private void reconnect() throws InterruptedException, IOException, TimeoutExcept
         }
 
         waitForStartup();
+        // Server is ready; subsequent reads use the normal per-request socket timeout.
+        socket.setSoTimeout((int) pipesConfig.getSocketTimeoutMs());
     }
 
     private void writeTask(FetchEmitTuple t) throws IOException {

tika-pipes/tika-pipes-core/src/main/java/org/apache/tika/pipes/core/PipesConfig.java

@@ -37,6 +37,8 @@ public class PipesConfig {
 
     public static final long DEFAULT_SOCKET_TIMEOUT_MS = 60000;
 
+    public static final long DEFAULT_STARTUP_TIMEOUT_MS = 60000;
+
     public static final long DEFAULT_HEARTBEAT_INTERVAL_MS = 1000;
 
     public static final boolean DEFAULT_USE_SHARED_SERVER = false;
@@ -55,6 +57,7 @@ public class PipesConfig {
     private boolean useSharedServer = DEFAULT_USE_SHARED_SERVER;
 
     private long socketTimeoutMs = DEFAULT_SOCKET_TIMEOUT_MS;
+    private long startupTimeoutMs = DEFAULT_STARTUP_TIMEOUT_MS;
     private long heartbeatIntervalMs = DEFAULT_HEARTBEAT_INTERVAL_MS;
 
     private long shutdownClientAfterMillis = DEFAULT_SHUTDOWN_CLIENT_AFTER_MILLS;
@@ -161,6 +164,21 @@ public void setSocketTimeoutMs(long socketTimeoutMs) {
         this.socketTimeoutMs = socketTimeoutMs;
     }
 
+    public long getStartupTimeoutMs() {
+        return startupTimeoutMs;
+    }
+
+    /**
+     * Timeout in milliseconds for the forked server to start up and send its READY handshake.
+     * Distinct from {@link #getSocketTimeoutMs()}: cold-starting the forked JVM (loading config,
+     * parsers and plugins) can take far longer than a normal per-read timeout, so the handshake
+     * gets its own generous budget. Once the server is ready, reads switch to {@code socketTimeoutMs}.
+     * @param startupTimeoutMs
+     */
+    public void setStartupTimeoutMs(long startupTimeoutMs) {
+        this.startupTimeoutMs = startupTimeoutMs;
+    }
+
     public long getHeartbeatIntervalMs() {
         return heartbeatIntervalMs;
     }

tika-pipes/tika-pipes-integration-tests/src/test/java/org/apache/tika/pipes/core/PipesClientTest.java

@@ -396,13 +396,13 @@ public void testSocketTimeout(@TempDir Path tmp) throws Exception {
             PipesResult pipesResult = pipesClient.process(tuple);
             long elapsed = System.currentTimeMillis() - startTime;
 
-            // Should timeout due to socket timeout (no heartbeats received within socketTimeoutMs)
-            // fails with FAILED_TO_INITIALIZE when using mchange-commons-java 0.5.0
+            // Should timeout due to socket timeout (no heartbeats received within socketTimeoutMs).
+            // Startup/handshake is bounded by startupTimeoutMs (not socketTimeoutMs), so a slow
+            // fork cold-start no longer misfires here as FAILED_TO_INITIALIZE.
             assertEquals(PipesResult.RESULT_STATUS.TIMEOUT, pipesResult.status(),
                     "Should timeout when socket times out");
 
-            // Socket timeout is 3 seconds; allow generous headroom for slow CI runners
-            // where the server may need multiple startup attempts before connecting.
+            // Socket timeout is 3 seconds; allow generous headroom for slow CI runners.
             assertTrue(elapsed < 60000,
                     "Socket timeout should occur within 60s (elapsed: " + elapsed + "ms)");