Add walkVM flow counters and stacktrace debug output

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: jbachorik

ddprof-lib/src/main/cpp/counters.h

@@ -73,7 +73,12 @@
   X(WALKVM_ANCHOR_NULL, "walkvm_anchor_null")                                  \
   X(WALKVM_CACHED_NOT_JAVA, "walkvm_cached_not_java")                          \
   X(WALKVM_NO_VMTHREAD, "walkvm_no_vmthread")                                  \
-  X(WALKVM_VMTHREAD_OK, "walkvm_vmthread_ok")
+  X(WALKVM_VMTHREAD_OK, "walkvm_vmthread_ok")                                  \
+  X(WALKVM_ANCHOR_USED_INLINE, "walkvm_anchor_used_inline")                    \
+  X(WALKVM_ANCHOR_FALLBACK, "walkvm_anchor_fallback")                          \
+  X(WALKVM_CODEH_NO_VM, "walkvm_codeh_no_vm")                                  \
+  X(WALKVM_DEPTH_ZERO, "walkvm_depth_zero")                                    \
+  X(WALKVM_HIT_CODEHEAP, "walkvm_hit_codeheap")
 #define X_ENUM(a, b) a,
 typedef enum CounterId : int {
   DD_COUNTER_TABLE(X_ENUM) DD_NUM_COUNTERS

ddprof-lib/src/main/cpp/stackWalker.cpp

@@ -329,6 +329,7 @@ __attribute__((no_sanitize("address"))) int StackWalker::walkVM(void* ucontext,
     // Walk until the bottom of the stack or until the first Java frame
     while (depth < actual_max_depth) {
         if (CodeHeap::contains(pc)) {
+            Counters::increment(WALKVM_HIT_CODEHEAP);
             // If we're in JVM-generated code but don't have a VMThread, we cannot safely
             // walk the Java stack because crash protection is not set up.
             //
@@ -343,6 +344,7 @@ __attribute__((no_sanitize("address"))) int StackWalker::walkVM(void* ucontext,
             //
             // The missing VMThread is a timing issue during thread lifecycle.
             if (vm_thread == NULL) {
+                Counters::increment(WALKVM_CODEH_NO_VM);
                 fillFrame(frames[depth++], BCI_ERROR, "break_no_vmthread");
                 break;
             }
@@ -560,6 +562,7 @@ __attribute__((no_sanitize("address"))) int StackWalker::walkVM(void* ucontext,
                 // These workarounds will minimize the number of unknown frames for 'vm'
                 // We want to keep the 'raw' data in 'vmx', though
                 if (anchor) {
+                    Counters::increment(WALKVM_ANCHOR_USED_INLINE);
                     uintptr_t prev_sp = sp;
                     sp = anchor->lastJavaSP();
                     fp = anchor->lastJavaFP();
@@ -673,6 +676,7 @@ __attribute__((no_sanitize("address"))) int StackWalker::walkVM(void* ucontext,
     // If we did not meet Java frame but current thread has JavaFrameAnchor set,
     // retry stack walking from the anchor
     if (anchor != NULL && anchor->getFrame(pc, sp, fp)) {
+        Counters::increment(WALKVM_ANCHOR_FALLBACK);
         anchor = NULL;
         while (depth > 0 && frames[depth - 1].method_id == NULL) depth--;  // pop unknown frames
         goto unwind_loop;
@@ -689,6 +693,10 @@ __attribute__((no_sanitize("address"))) int StackWalker::walkVM(void* ucontext,
     // aggregation by lumping unrelated samples under a single "unknown" entry
     depth = StackWalkValidation::dropUnknownLeaf(frames, depth);
 
+    if (depth == 0) {
+        Counters::increment(WALKVM_DEPTH_ZERO);
+    }
+
     if (truncated) {
         if (depth > max_depth) {
             *truncated = true;

ddprof-test/src/test/java/com/datadoghq/profiler/context/TagContextTest.java

@@ -59,6 +59,9 @@ public void test() throws InterruptedException {
                     jfrCounters.put(nameAccessor.getMember(item), countAccessor.getMember(item).longValue());
                 }
             }
+            int debugPrinted = 0;
+            int nullTraceCount = 0;
+            int emptyTraceCount = 0;
             for (IItemIterable wallclockSamples : events) {
                 IMemberAccessor<IQuantity, IItem> weightAccessor = WEIGHT.getAccessor(wallclockSamples.getType());
                 // this will become more generic in the future
@@ -69,6 +72,12 @@ public void test() throws InterruptedException {
                 IMemberAccessor<String, IItem> stacktraceAccessor = JdkAttributes.STACK_TRACE_STRING.getAccessor(wallclockSamples.getType());
                 for (IItem sample : wallclockSamples) {
                     String stacktrace = stacktraceAccessor.getMember(sample);
+                    if (stacktrace == null) { nullTraceCount++; continue; }
+                    if (stacktrace.isEmpty()) { emptyTraceCount++; continue; }
+                    if (debugPrinted < 5) {
+                        System.out.println("[DEBUG-TRACE-" + debugPrinted + "] " + stacktrace.substring(0, Math.min(500, stacktrace.length())));
+                        debugPrinted++;
+                    }
                     if (!stacktrace.contains("sleep")) {
                         // we don't know the context has been set for sure until the sleep has started
                         continue;
@@ -134,13 +143,19 @@ public void test() throws InterruptedException {
             System.out.printf("Sample statistics: %d total (%d dropped, %.2f%%), weight %d total (%d dropped, %.2f%%)%n",
                     totalSamplesCount, droppedSamplesCount, dropRate,
                     totalSamplesWeight, droppedSamplesWeight, dropWeightRate);
+            System.out.println("nullTraces=" + nullTraceCount + " emptyTraces=" + emptyTraceCount);
             // Print walkvm diagnostic counters
             System.out.println("walkvm_vmthread_ok=" + jfrCounters.getOrDefault("walkvm_vmthread_ok", 0L)
                     + " walkvm_no_vmthread=" + jfrCounters.getOrDefault("walkvm_no_vmthread", 0L)
                     + " walkvm_thread_inaccessible=" + jfrCounters.getOrDefault("walkvm_thread_inaccessible", 0L)
                     + " walkvm_anchor_null=" + jfrCounters.getOrDefault("walkvm_anchor_null", 0L)
                     + " walkvm_cached_not_java=" + jfrCounters.getOrDefault("walkvm_cached_not_java", 0L)
                     + " thread_entry_mark_detections=" + jfrCounters.getOrDefault("thread_entry_mark_detections", 0L));
+            System.out.println("walkvm_hit_codeheap=" + jfrCounters.getOrDefault("walkvm_hit_codeheap", 0L)
+                    + " walkvm_codeh_no_vm=" + jfrCounters.getOrDefault("walkvm_codeh_no_vm", 0L)
+                    + " walkvm_anchor_used_inline=" + jfrCounters.getOrDefault("walkvm_anchor_used_inline", 0L)
+                    + " walkvm_anchor_fallback=" + jfrCounters.getOrDefault("walkvm_anchor_fallback", 0L)
+                    + " walkvm_depth_zero=" + jfrCounters.getOrDefault("walkvm_depth_zero", 0L));
         }
     }