Attempt to debug CI flaky timeout. DO NOT MERGE!

.gitlab/collect_reports.sh

@@ -62,6 +62,7 @@ function process_reports () {
       cp -r workspace/$project_to_save/build/reports/* $report_path/ 2>/dev/null || true
       cp workspace/$project_to_save/build/hs_err_pid*.log $report_path/ 2>/dev/null || true
       cp workspace/$project_to_save/build/javacore*.txt $report_path/ 2>/dev/null || true
+      cp workspace/$project_to_save/build/thread-dump-*.log $report_path/ 2>/dev/null || true
     fi
 }
 

dd-java-agent/instrumentation/java-concurrent/java-concurrent-21/src/previewTest/groovy/StructuredConcurrencyTest.groovy

@@ -2,13 +2,61 @@ import datadog.trace.agent.test.AgentTestRunner
 import datadog.trace.api.Trace
 
 import java.util.concurrent.Callable
+import java.util.concurrent.Executors
+import java.util.concurrent.ScheduledExecutorService
+import java.util.concurrent.ScheduledFuture
 import java.util.concurrent.StructuredTaskScope
+import java.util.concurrent.TimeUnit
 
 import static datadog.trace.agent.test.utils.TraceUtils.runUnderTrace
 import static datadog.trace.agent.test.utils.TraceUtils.runnableUnderTrace
 import static java.time.Instant.now
 
 class StructuredConcurrencyTest extends AgentTestRunner {
+  ScheduledExecutorService scheduler
+  ScheduledFuture<?> threadDumpTask
+
+  def setup() {
+    scheduler = Executors.newSingleThreadScheduledExecutor()
+
+    threadDumpTask = scheduler.scheduleAtFixedRate({
+      File reportDir = new File("build")
+      println("DEBUG: " + reportDir.absolutePath)
+
+      if (!reportDir.exists()) {
+        println("build folder not found")
+        return
+      }
+
+      // Define the file path
+      File reportFile = new File(reportDir, String.format("thread-dump-%d.log", System.currentTimeMillis()))
+
+      // Write to the file
+      try (FileWriter writer = new FileWriter(reportFile)) {
+        def s = "=== Thread Dump Triggered at ${new Date()} ===\n"
+        println(s)
+        writer.write(s)
+        Thread.getAllStackTraces().each { thread, stack ->
+          def t = "Thread: ${thread.name}, daemon: ${thread.daemon}\n"
+          println(t)
+          writer.write(t)
+          stack.each {
+            def st = "\tat ${it}\n"
+            println(st)
+            writer.write(st)
+          }
+        }
+        writer.write("==============================================\n")
+      }
+    }, 10, 60_000, TimeUnit.MILLISECONDS)
+  }
+
+  def cleanup() {
+    threadDumpTask?.cancel(false)
+    scheduler?.shutdownNow()
+  }
+
+
   /**
    * Tests the structured task scope with a single task.
    */
@@ -18,14 +66,15 @@ class StructuredConcurrencyTest extends AgentTestRunner {
     def result = false
 
     when:
+    Thread.sleep(100)
     runUnderTrace("parent") {
       def task = taskScope.fork(new Callable<Boolean>() {
-          @Trace(operationName = "child")
-          @Override
-          Boolean call() throws Exception {
-            return true
-          }
-        })
+        @Trace(operationName = "child")
+        @Override
+        Boolean call() throws Exception {
+          return true
+        }
+      })
       taskScope.joinUntil(now() + 10) // Wait for 10 seconds at maximum
       result = task.get()
     }

dd-java-agent/instrumentation/lettuce-4/src/test/groovy/Lettuce4ClientTestBase.groovy

@@ -8,6 +8,11 @@ import datadog.trace.agent.test.utils.PortUtils
 import redis.embedded.RedisServer
 import spock.lang.Shared
 
+import java.util.concurrent.Executors
+import java.util.concurrent.ScheduledExecutorService
+import java.util.concurrent.ScheduledFuture
+import java.util.concurrent.TimeUnit
+
 import static datadog.trace.agent.test.utils.TraceUtils.runUnderTrace
 
 abstract class Lettuce4ClientTestBase extends VersionedNamingTestBase {
@@ -32,6 +37,9 @@ abstract class Lettuce4ClientTestBase extends VersionedNamingTestBase {
   @Shared
   RedisServer redisServer
 
+  ScheduledExecutorService scheduler
+  ScheduledFuture<?> threadDumpTask
+
   @Shared
   Map<String, String> testHashMap = [
     firstname: "John",
@@ -53,14 +61,47 @@ abstract class Lettuce4ClientTestBase extends VersionedNamingTestBase {
     embeddedDbUri = "redis://" + dbAddr
 
     redisServer = RedisServer.newRedisServer()
-      // bind to localhost to avoid firewall popup
-      .setting("bind " + HOST)
-      // set max memory to avoid problems in CI
-      .setting("maxmemory 128M")
-      .port(port).build()
+    // bind to localhost to avoid firewall popup
+    .setting("bind " + HOST)
+    // set max memory to avoid problems in CI
+    .setting("maxmemory 128M")
+    .port(port).build()
   }
 
   def setup() {
+    scheduler = Executors.newSingleThreadScheduledExecutor()
+
+    threadDumpTask = scheduler.scheduleAtFixedRate({
+      File reportDir = new File("build")
+      println("DEBUG: " + reportDir.absolutePath)
+
+      if (!reportDir.exists()) {
+        println("build folder not found")
+        return
+      }
+
+      // Define the file path
+      File reportFile = new File(reportDir, String.format("thread-dump-%d.log", System.currentTimeMillis()))
+
+      // Write to the file
+      try (FileWriter writer = new FileWriter(reportFile)) {
+        def s = "=== Thread Dump Triggered at ${new Date()} ===\n"
+        println(s)
+        writer.write(s)
+        Thread.getAllStackTraces().each { thread, stack ->
+          def t = "Thread: ${thread.name}, daemon: ${thread.daemon}\n"
+          println(t)
+          writer.write(t)
+          stack.each {
+            def st = "\tat ${it}\n"
+            println(st)
+            writer.write(st)
+          }
+        }
+        writer.write("==============================================\n")
+      }
+    }, 10, 60_000, TimeUnit.MILLISECONDS)
+
     redisServer.start()
 
     redisClient = RedisClient.create(embeddedDbUri)
@@ -79,6 +120,9 @@ abstract class Lettuce4ClientTestBase extends VersionedNamingTestBase {
   }
 
   def cleanup() {
+    threadDumpTask?.cancel(false)
+    scheduler?.shutdownNow()
+
     connection.close()
     redisClient.shutdown()
     redisServer.stop()