Bound reference-chain pass rate and root-enum frequency (PROF-16384)

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

@@ -291,11 +291,8 @@
   X(REFERENCE_CHAIN_CANDIDATE_COUNT, "reference_chain_candidate_count") \
   X(REFERENCE_CHAIN_CANDIDATES_FOUND, "reference_chain_candidates_found") \
   /* hasLeakSignal() gate evaluations refused while the OOM urgency latch is \
-   * set and selectLeakCandidates() found zero candidates - a heap-wide leak \
-   * no single klass explains, so no reference-chain search may start. \
-   * One increment per refused evaluation, not per urgency episode: evaluation \
-   * runs only after canAffordNewSearch()'s safepoint budget check and follows \
-   * the ramped cadence. */ \
+   * set and selectLeakCandidates() found zero candidates (a heap-wide leak no \
+   * single klass explains). Counts evaluations (~1/s while latched), not episodes. */ \
   X(REFERENCE_CHAIN_URGENT_NO_CANDIDATE, "reference_chain_urgent_no_candidate") \
   /* Heap-floor seqlock: samples DROPPED when a writer claims a slot whose \
    * seq is still the sentinel - only possible if the ring lapped an \

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

@@ -1473,6 +1473,75 @@ Java_com_datadoghq_profiler_JavaProfiler_isUrgentForTest0(JNIEnv *env,
   return tracker->isUrgentForTest() ? JNI_TRUE : JNI_FALSE;
 }
 
+// Places the previous pass a controlled distance in the past and sets the effective cadence
+// shouldRunPass()'s minimum-spacing gate enforces - see setPassSpacingForTest().
+extern "C" DLLEXPORT void JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_setReferenceChainPassSpacingForTest0(
+    JNIEnv *env, jclass unused, jlong ms_since_last_pass, jlong cadence_ms) {
+  ProfiledThread::initCurrentThreadSignalSafe();
+  ReferenceChainTracker::instance()->setPassSpacingForTest(
+      (u64)ms_since_last_pass, (u64)cadence_ms);
+}
+
+// Sets the root-enumeration scheduling state (last attempt's time and measured in-safepoint
+// cost) - see setRootEnumStateForTest().
+extern "C" DLLEXPORT void JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_setReferenceChainRootEnumStateForTest0(
+    JNIEnv *env, jclass unused, jlong ms_since_last_enum,
+    jlong last_enum_cost_ms) {
+  ProfiledThread::initCurrentThreadSignalSafe();
+  ReferenceChainTracker::instance()->setRootEnumStateForTest(
+      (long)ms_since_last_enum, (u64)last_enum_cost_ms);
+}
+
+// Evaluates shouldRunRootEnum() with a synthetic clock so a test can place the last attempt a
+// controlled distance in the past.
+extern "C" DLLEXPORT jboolean JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_shouldRunRootEnumForTest0(
+    JNIEnv *env, jclass unused, jboolean manual_first_pass) {
+  ProfiledThread::initCurrentThreadSignalSafe();
+  return ReferenceChainTracker::instance()->shouldRunRootEnumForTest(
+             OS::nanotime(), manual_first_pass == JNI_TRUE)
+             ? JNI_TRUE
+             : JNI_FALSE;
+}
+
+// Forces a started, RUNNING search with fresh pain budgets so shouldRunPassForTest0() evaluates
+// the running-search gates deterministically - see forceRunningSearchForTest().
+extern "C" DLLEXPORT void JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_forceRunningReferenceChainSearchForTest0(
+    JNIEnv *env, jclass unused) {
+  ProfiledThread::initCurrentThreadSignalSafe();
+  ReferenceChainTracker::instance()->forceRunningSearchForTest();
+}
+
+// Sets the canary chase's candidate count and found-bits mask so shouldRunPassForTest0()
+// evaluates the canary lane - see setCanaryChaseStateForTest().
+extern "C" DLLEXPORT void JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_setReferenceChainCanaryStateForTest0(
+    JNIEnv *env, jclass unused, jint candidate_count, jlong found_bits) {
+  ProfiledThread::initCurrentThreadSignalSafe();
+  ReferenceChainTracker::instance()->setCanaryChaseStateForTest(
+      (int)candidate_count, (u64)found_bits);
+}
+
+// Current adaptive root-enumeration interval floor - see rootEnumIntervalFloorNs().
+extern "C" DLLEXPORT jlong JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_rootEnumIntervalFloorNsForTest0(
+    JNIEnv *env, jclass unused) {
+  ProfiledThread::initCurrentThreadSignalSafe();
+  return (jlong)ReferenceChainTracker::instance()
+      ->rootEnumIntervalFloorNsForTest();
+}
+
+// Live GC-finish epoch - see gcFinishEpoch() (relaxed atomic load, no lock needed).
+extern "C" DLLEXPORT jlong JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_referenceChainGcFinishEpochForTest0(
+    JNIEnv *env, jclass unused) {
+  ProfiledThread::initCurrentThreadSignalSafe();
+  return (jlong)ReferenceChainTracker::instance()->gcFinishEpoch();
+}
+
 #else // !DEBUG
 
 // Release builds: no-op stubs so the Java natives still link.
@@ -1595,6 +1664,41 @@ Java_com_datadoghq_profiler_JavaProfiler_backdateLeakSignalActivationForTest0(
   return 0;
 }
 
+extern "C" DLLEXPORT void JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_setReferenceChainPassSpacingForTest0(
+    JNIEnv *env, jclass unused, jlong ms_since_last_pass, jlong cadence_ms) {}
+
+extern "C" DLLEXPORT void JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_setReferenceChainRootEnumStateForTest0(
+    JNIEnv *env, jclass unused, jlong ms_since_last_enum,
+    jlong last_enum_cost_ms) {}
+
+extern "C" DLLEXPORT jboolean JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_shouldRunRootEnumForTest0(
+    JNIEnv *env, jclass unused, jboolean manual_first_pass) {
+  return JNI_FALSE;
+}
+
+extern "C" DLLEXPORT void JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_forceRunningReferenceChainSearchForTest0(
+    JNIEnv *env, jclass unused) {}
+
+extern "C" DLLEXPORT void JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_setReferenceChainCanaryStateForTest0(
+    JNIEnv *env, jclass unused, jint candidate_count, jlong found_bits) {}
+
+extern "C" DLLEXPORT jlong JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_rootEnumIntervalFloorNsForTest0(
+    JNIEnv *env, jclass unused) {
+  return 0;
+}
+
+extern "C" DLLEXPORT jlong JNICALL
+Java_com_datadoghq_profiler_JavaProfiler_referenceChainGcFinishEpochForTest0(
+    JNIEnv *env, jclass unused) {
+  return 0;
+}
+
 extern "C" DLLEXPORT void JNICALL
 Java_com_datadoghq_profiler_JavaProfiler_setLeakSignalActivatedNsForTest0(
     JNIEnv *env, jclass unused, jlong activatedNs) {}

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

@@ -47,6 +47,9 @@ constexpr u8 CANARY_STUCK = 3;
 // starts from a brand-new-search shape (release-then-restart, fresh TTL clock, no urgency
 // carry-over).
 constexpr u8 RECORDING_END = 4;
+// The root enumerator returned a JVMTI error on MAX_CONSECUTIVE_ENUM_ERRORS consecutive attempts -
+// the root picture is unusable and retrying is not making progress.
+constexpr u8 ROOT_ENUM_ERROR = 5;
 } // namespace SearchAbandonReason
 
 // Metadata for one tagged frontier object.