ddtrace 1.22.0 → 1.23.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 55da7739191e23ba6bb188ade9979c9dbd0a965bddfd2faedd6bf2c41b3a6e97
-  data.tar.gz: 98bfd95dc6dd156d3d57f2ac22deb081a8e6fb8d3221eadf6e682f45e358ef7c
+  metadata.gz: b5e8a3ccb932af75df7d30d6fe59edc8277e21ee34a9562f7c2c8bfba1ea7521
+  data.tar.gz: 1508bf7e56a24af598aad666665be6cd0a7b97c9cbf3bbcc95a3fda36b67c07a
 SHA512:
-  metadata.gz: 526f2826818f960ca18703a03cc0bde6313203a4e2dd03a696b8bd08f5a12cd963c370a8a9278d7387d81cf4445b849693fdbb8f91989f27f20d4f5bf55ad1a2
-  data.tar.gz: 9b8d42db31a997fd7f03e383e0e2671730577336f5bdb28bf36afd63e775639f584abb8d2005265835fc175c29ae275f5cdfea79ba9afe02eb16a3a826f47d70
+  metadata.gz: afccca8d070d1dfd472af2390da1b343dff3199c42a71e742f10e29e91ddb27a6fa709490adcd639e6914f792b8e8565f667ff5d916f9f0f6690d69b8d8b0b54
+  data.tar.gz: 581022bd8d146fe01e9dac70a309e58d503d48ed8420b845a6b91200796ce9a0b1c2bc29fb58382ddc546abd00785d2f6ee42863e1bc6df517052a566352604e

data/CHANGELOG.md

@@ -2,6 +2,17 @@
 
 ## [Unreleased]
 
+## [1.23.0] - 2024-05-09
+
+### Added
+
+* Profiling: Enable endpoint profiling for Sidekiq and similar background job processors ([#3619][])
+
+### Fixed
+
+* Fix no such file or directory issue when using single step instrumentation ([#3623][])
+* Fix error during telemetry debug logging attempt ([#3618][])
+
 ## [1.22.0] - 2024-04-16
 
 ### Added
@@ -2804,6 +2815,7 @@ Git diff: https://github.com/DataDog/dd-trace-rb/compare/v0.3.0...v0.3.1
 
 
 [Unreleased]: https://github.com/DataDog/dd-trace-rb/compare/v1.22.0...master
+[1.23.0]: https://github.com/DataDog/dd-trace-rb/compare/v1.22.0...v1.23.0
 [1.22.0]: https://github.com/DataDog/dd-trace-rb/compare/v1.21.1...v1.22.0
 [2.0.0.beta1]: https://github.com/DataDog/dd-trace-rb/compare/v1.21.1...v2.0.0.beta1
 [1.21.1]: https://github.com/DataDog/dd-trace-rb/compare/v1.21.0...v1.21.1
@@ -4111,6 +4123,9 @@ Git diff: https://github.com/DataDog/dd-trace-rb/compare/v0.3.0...v0.3.1
 [#3582]: https://github.com/DataDog/dd-trace-rb/issues/3582
 [#3585]: https://github.com/DataDog/dd-trace-rb/issues/3585
 [#3587]: https://github.com/DataDog/dd-trace-rb/issues/3587
+[#3618]: https://github.com/DataDog/dd-trace-rb/issues/3618
+[#3619]: https://github.com/DataDog/dd-trace-rb/issues/3619
+[#3623]: https://github.com/DataDog/dd-trace-rb/issues/3623
 [@AdrianLC]: https://github.com/AdrianLC
 [@Azure7111]: https://github.com/Azure7111
 [@BabyGroot]: https://github.com/BabyGroot

data/ext/datadog_profiling_native_extension/collectors_cpu_and_wall_time_worker.c

@@ -929,18 +929,6 @@ static VALUE _native_stats(DDTRACE_UNUSED VALUE self, VALUE instance) {
   struct cpu_and_wall_time_worker_state *state;
   TypedData_Get_Struct(instance, struct cpu_and_wall_time_worker_state, &cpu_and_wall_time_worker_typed_data, state);
 
-  VALUE pretty_cpu_sampling_time_ns_min = state->stats.cpu_sampling_time_ns_min == UINT64_MAX ? Qnil : ULL2NUM(state->stats.cpu_sampling_time_ns_min);
-  VALUE pretty_cpu_sampling_time_ns_max = state->stats.cpu_sampling_time_ns_max == 0 ? Qnil : ULL2NUM(state->stats.cpu_sampling_time_ns_max);
-  VALUE pretty_cpu_sampling_time_ns_total = state->stats.cpu_sampling_time_ns_total == 0 ? Qnil : ULL2NUM(state->stats.cpu_sampling_time_ns_total);
-  VALUE pretty_cpu_sampling_time_ns_avg =
-    state->stats.cpu_sampled == 0 ? Qnil : DBL2NUM(((double) state->stats.cpu_sampling_time_ns_total) / state->stats.cpu_sampled);
-
-  VALUE pretty_allocation_sampling_time_ns_min = state->stats.allocation_sampling_time_ns_min == UINT64_MAX ? Qnil : ULL2NUM(state->stats.allocation_sampling_time_ns_min);
-  VALUE pretty_allocation_sampling_time_ns_max = state->stats.allocation_sampling_time_ns_max == 0 ? Qnil : ULL2NUM(state->stats.allocation_sampling_time_ns_max);
-  VALUE pretty_allocation_sampling_time_ns_total = state->stats.allocation_sampling_time_ns_total == 0 ? Qnil : ULL2NUM(state->stats.allocation_sampling_time_ns_total);
-  VALUE pretty_allocation_sampling_time_ns_avg =
-    state->stats.allocation_sampled == 0 ? Qnil : DBL2NUM(((double) state->stats.allocation_sampling_time_ns_total) / state->stats.allocation_sampled);
-
   unsigned long total_cpu_samples_attempted = state->stats.cpu_sampled + state->stats.cpu_skipped;
   VALUE effective_cpu_sample_rate =
     total_cpu_samples_attempted == 0 ? Qnil : DBL2NUM(((double) state->stats.cpu_sampled) / total_cpu_samples_attempted);
@@ -968,19 +956,19 @@ static VALUE _native_stats(DDTRACE_UNUSED VALUE self, VALUE instance) {
     ID2SYM(rb_intern("cpu_sampled")),                /* => */ UINT2NUM(state->stats.cpu_sampled),
     ID2SYM(rb_intern("cpu_skipped")),                /* => */ UINT2NUM(state->stats.cpu_skipped),
     ID2SYM(rb_intern("cpu_effective_sample_rate")),  /* => */ effective_cpu_sample_rate,
-    ID2SYM(rb_intern("cpu_sampling_time_ns_min")),   /* => */ pretty_cpu_sampling_time_ns_min,
-    ID2SYM(rb_intern("cpu_sampling_time_ns_max")),   /* => */ pretty_cpu_sampling_time_ns_max,
-    ID2SYM(rb_intern("cpu_sampling_time_ns_total")), /* => */ pretty_cpu_sampling_time_ns_total,
-    ID2SYM(rb_intern("cpu_sampling_time_ns_avg")),   /* => */ pretty_cpu_sampling_time_ns_avg,
+    ID2SYM(rb_intern("cpu_sampling_time_ns_min")),   /* => */ RUBY_NUM_OR_NIL(state->stats.cpu_sampling_time_ns_min, != UINT64_MAX, ULL2NUM),
+    ID2SYM(rb_intern("cpu_sampling_time_ns_max")),   /* => */ RUBY_NUM_OR_NIL(state->stats.cpu_sampling_time_ns_max, > 0, ULL2NUM),
+    ID2SYM(rb_intern("cpu_sampling_time_ns_total")), /* => */ RUBY_NUM_OR_NIL(state->stats.cpu_sampling_time_ns_total, > 0, ULL2NUM),
+    ID2SYM(rb_intern("cpu_sampling_time_ns_avg")),   /* => */ RUBY_AVG_OR_NIL(state->stats.cpu_sampling_time_ns_total, state->stats.cpu_sampled),
 
     // Allocation stats
     ID2SYM(rb_intern("allocation_sampled")),                /* => */ state->allocation_profiling_enabled ? ULONG2NUM(state->stats.allocation_sampled) : Qnil,
     ID2SYM(rb_intern("allocation_skipped")),                /* => */ state->allocation_profiling_enabled ? ULONG2NUM(state->stats.allocation_skipped) : Qnil,
     ID2SYM(rb_intern("allocation_effective_sample_rate")),  /* => */ effective_allocation_sample_rate,
-    ID2SYM(rb_intern("allocation_sampling_time_ns_min")),   /* => */ pretty_allocation_sampling_time_ns_min,
-    ID2SYM(rb_intern("allocation_sampling_time_ns_max")),   /* => */ pretty_allocation_sampling_time_ns_max,
-    ID2SYM(rb_intern("allocation_sampling_time_ns_total")), /* => */ pretty_allocation_sampling_time_ns_total,
-    ID2SYM(rb_intern("allocation_sampling_time_ns_avg")),   /* => */ pretty_allocation_sampling_time_ns_avg,
+    ID2SYM(rb_intern("allocation_sampling_time_ns_min")),   /* => */ RUBY_NUM_OR_NIL(state->stats.allocation_sampling_time_ns_min, != UINT64_MAX, ULL2NUM),
+    ID2SYM(rb_intern("allocation_sampling_time_ns_max")),   /* => */ RUBY_NUM_OR_NIL(state->stats.allocation_sampling_time_ns_max, > 0, ULL2NUM),
+    ID2SYM(rb_intern("allocation_sampling_time_ns_total")), /* => */ RUBY_NUM_OR_NIL(state->stats.allocation_sampling_time_ns_total, > 0, ULL2NUM),
+    ID2SYM(rb_intern("allocation_sampling_time_ns_avg")),   /* => */ RUBY_AVG_OR_NIL(state->stats.allocation_sampling_time_ns_total, state->stats.allocation_sampled),
     ID2SYM(rb_intern("allocation_sampler_snapshot")),       /* => */ allocation_sampler_snapshot,
     ID2SYM(rb_intern("allocations_during_sample")),         /* => */ state->allocation_profiling_enabled ? UINT2NUM(state->stats.allocations_during_sample) : Qnil,
   };

data/ext/datadog_profiling_native_extension/collectors_thread_context.c

@@ -217,7 +217,7 @@ static long thread_id_for(VALUE thread);
 static VALUE _native_stats(VALUE self, VALUE collector_instance);
 static VALUE _native_gc_tracking(VALUE self, VALUE collector_instance);
 static void trace_identifiers_for(struct thread_context_collector_state *state, VALUE thread, struct trace_identifiers *trace_identifiers_result);
-static bool should_collect_resource(VALUE root_span_type);
+static bool should_collect_resource(VALUE root_span);
 static VALUE _native_reset_after_fork(DDTRACE_UNUSED VALUE self, VALUE collector_instance);
 static VALUE thread_list(struct thread_context_collector_state *state);
 static VALUE _native_sample_allocation(DDTRACE_UNUSED VALUE self, VALUE collector_instance, VALUE sample_weight, VALUE new_object);
@@ -1146,10 +1146,7 @@ static void trace_identifiers_for(struct thread_context_collector_state *state,
 
   trace_identifiers_result->valid = true;
 
-  if (!state->endpoint_collection_enabled) return;
-
-  VALUE root_span_type = rb_ivar_get(root_span, at_type_id /* @type */);
-  if (root_span_type == Qnil || !should_collect_resource(root_span_type)) return;
+  if (!state->endpoint_collection_enabled || !should_collect_resource(root_span)) return;
 
   VALUE trace_resource = rb_ivar_get(active_trace, at_resource_id /* @resource */);
   if (RB_TYPE_P(trace_resource, T_STRING)) {
@@ -1160,21 +1157,32 @@ static void trace_identifiers_for(struct thread_context_collector_state *state,
   }
 }
 
-// We only collect the resource for spans of types:
+// We opt-in to collecting the resource for spans of types:
 // * 'web', for web requests
-// * proxy', used by the rack integration with request_queuing: true (e.g. also represents a web request)
+// * 'proxy', used by the rack integration with request_queuing: true (e.g. also represents a web request)
+// * 'worker', used for sidekiq and similar background job processors
 //
-// NOTE: Currently we're only interested in HTTP service endpoints. Over time, this list may be expanded.
+// Over time, this list may be expanded.
 // Resources MUST NOT include personal identifiable information (PII); this should not be the case with
 // ddtrace integrations, but worth mentioning just in case :)
-static bool should_collect_resource(VALUE root_span_type) {
+static bool should_collect_resource(VALUE root_span) {
+  VALUE root_span_type = rb_ivar_get(root_span, at_type_id /* @type */);
+  if (root_span_type == Qnil) return false;
   ENFORCE_TYPE(root_span_type, T_STRING);
 
   int root_span_type_length = RSTRING_LEN(root_span_type);
   const char *root_span_type_value = StringValuePtr(root_span_type);
 
-  return (root_span_type_length == strlen("web") && (memcmp("web", root_span_type_value, strlen("web")) == 0)) ||
+  bool is_web_request =
+    (root_span_type_length == strlen("web") && (memcmp("web", root_span_type_value, strlen("web")) == 0)) ||
     (root_span_type_length == strlen("proxy") && (memcmp("proxy", root_span_type_value, strlen("proxy")) == 0));
+
+  if (is_web_request) return true;
+
+  bool is_worker_request =
+    (root_span_type_length == strlen("worker") && (memcmp("worker", root_span_type_value, strlen("worker")) == 0));
+
+  return is_worker_request;
 }
 
 // After the Ruby VM forks, this method gets called in the child process to clean up any leftover state from the parent.

data/ext/datadog_profiling_native_extension/heap_recorder.c

@@ -158,6 +158,13 @@ struct heap_recorder {
 
   // Sampling state
   uint num_recordings_skipped;
+
+  struct stats_last_update {
+    size_t objects_alive;
+    size_t objects_dead;
+    size_t objects_skipped;
+    size_t objects_frozen;
+  } stats_last_update;
 };
 static heap_record* get_or_create_heap_record(heap_recorder*, ddog_prof_Slice_Location);
 static void cleanup_heap_record_if_unused(heap_recorder*, heap_record*);
@@ -372,6 +379,9 @@ void heap_recorder_prepare_iteration(heap_recorder *heap_recorder) {
     rb_raise(rb_eRuntimeError, "New heap recorder iteration prepared without the previous one having been finished.");
   }
 
+  // Reset last update stats, we'll be building them from scratch during the st_foreach call below
+  heap_recorder->stats_last_update = (struct stats_last_update) {};
+
   st_foreach(heap_recorder->object_records, st_object_record_update, (st_data_t) heap_recorder);
 
   heap_recorder->object_records_snapshot = st_copy(heap_recorder->object_records);
@@ -427,6 +437,22 @@ bool heap_recorder_for_each_live_object(
   return true;
 }
 
+VALUE heap_recorder_state_snapshot(heap_recorder *heap_recorder) {
+  VALUE arguments[] = {
+    ID2SYM(rb_intern("num_object_records")), /* => */ LONG2NUM(heap_recorder->object_records->num_entries),
+    ID2SYM(rb_intern("num_heap_records")),   /* => */ LONG2NUM(heap_recorder->heap_records->num_entries),
+
+    // Stats as of last update
+    ID2SYM(rb_intern("last_update_objects_alive")), /* => */ LONG2NUM(heap_recorder->stats_last_update.objects_alive),
+    ID2SYM(rb_intern("last_update_objects_dead")), /* => */ LONG2NUM(heap_recorder->stats_last_update.objects_dead),
+    ID2SYM(rb_intern("last_update_objects_skipped")), /* => */ LONG2NUM(heap_recorder->stats_last_update.objects_skipped),
+    ID2SYM(rb_intern("last_update_objects_frozen")), /* => */ LONG2NUM(heap_recorder->stats_last_update.objects_frozen),
+  };
+  VALUE hash = rb_hash_new();
+  for (long unsigned int i = 0; i < VALUE_COUNT(arguments); i += 2) rb_hash_aset(hash, arguments[i], arguments[i+1]);
+  return hash;
+}
+
 void heap_recorder_testonly_assert_hash_matches(ddog_prof_Slice_Location locations) {
   heap_stack *stack = heap_stack_new(locations);
   heap_record_key stack_based_key = (heap_record_key) {
@@ -497,12 +523,14 @@ static int st_object_record_update(st_data_t key, st_data_t value, st_data_t ext
     // no point checking for liveness or updating its size, so exit early.
     // NOTE: This means that there should be an equivalent check during actual
     //       iteration otherwise we'd iterate/expose stale object data.
+    recorder->stats_last_update.objects_skipped++;
     return ST_CONTINUE;
   }
 
   if (!ruby_ref_from_id(LONG2NUM(obj_id), &ref)) {
     // Id no longer associated with a valid ref. Need to delete this object record!
     on_committed_object_record_cleanup(recorder, record);
+    recorder->stats_last_update.objects_dead++;
     return ST_DELETE;
   }
 
@@ -537,6 +565,7 @@ static int st_object_record_update(st_data_t key, st_data_t value, st_data_t ext
       RB_FL_SET(ref, RUBY_FL_SEEN_OBJ_ID);
 
       on_committed_object_record_cleanup(recorder, record);
+      recorder->stats_last_update.objects_dead++;
       return ST_DELETE;
     }
 
@@ -550,6 +579,11 @@ static int st_object_record_update(st_data_t key, st_data_t value, st_data_t ext
     record->object_data.is_frozen = RB_OBJ_FROZEN(ref);
   }
 
+  recorder->stats_last_update.objects_alive++;
+  if (record->object_data.is_frozen) {
+    recorder->stats_last_update.objects_frozen++;
+  }
+
   return ST_CONTINUE;
 }
 
@@ -767,9 +801,10 @@ void object_record_free(object_record *record) {
 
 VALUE object_record_inspect(object_record *record) {
   heap_frame top_frame = record->heap_record->stack->frames[0];
-  VALUE inspect = rb_sprintf("obj_id=%ld weight=%d size=%zu location=%s:%d alloc_gen=%zu gen_age=%zu ",
-      record->obj_id, record->object_data.weight, record->object_data.size, top_frame.filename,
-      (int) top_frame.line, record->object_data.alloc_gen, record->object_data.gen_age);
+  live_object_data object_data = record->object_data;
+  VALUE inspect = rb_sprintf("obj_id=%ld weight=%d size=%zu location=%s:%d alloc_gen=%zu gen_age=%zu frozen=%d ",
+      record->obj_id, object_data.weight, object_data.size, top_frame.filename,
+      (int) top_frame.line, object_data.alloc_gen, object_data.gen_age, object_data.is_frozen);
 
   const char *class = record->object_data.class;
   if (class != NULL) {

data/ext/datadog_profiling_native_extension/heap_recorder.h

@@ -150,6 +150,11 @@ bool heap_recorder_for_each_live_object(
     bool (*for_each_callback)(heap_recorder_iteration_data data, void* extra_arg),
     void *for_each_callback_extra_arg);
 
+// Return a Ruby hash containing a snapshot of this recorder's interesting state at calling time.
+// WARN: This allocates in the Ruby VM and therefore should not be called without the
+//       VM lock or during GC.
+VALUE heap_recorder_state_snapshot(heap_recorder *heap_recorder);
+
 // v--- TEST-ONLY APIs ---v
 
 // Assert internal hashing logic is valid for the provided locations and its

data/ext/datadog_profiling_native_extension/ruby_helpers.h

@@ -82,6 +82,9 @@ NORETURN(
 #define ENFORCE_SUCCESS_HELPER(expression, have_gvl) \
   { int result_syserr_errno = expression; if (RB_UNLIKELY(result_syserr_errno)) raise_syserr(result_syserr_errno, have_gvl, ADD_QUOTES(expression), __FILE__, __LINE__, __func__); }
 
+#define RUBY_NUM_OR_NIL(val, condition, conv) ((val condition) ? conv(val) : Qnil)
+#define RUBY_AVG_OR_NIL(total, count) ((count == 0) ? Qnil : DBL2NUM(((double) total) / count))
+
 // Called by ENFORCE_SUCCESS_HELPER; should not be used directly
 NORETURN(void raise_syserr(
   int syserr_errno,

data/ext/datadog_profiling_native_extension/stack_recorder.c

@@ -169,28 +169,51 @@ static const uint8_t all_value_types_positions[] =
 
 #define ALL_VALUE_TYPES_COUNT (sizeof(all_value_types) / sizeof(ddog_prof_ValueType))
 
+// Struct for storing stats related to a profile in a particular slot.
+// These stats will share the same lifetime as the data in that profile slot.
+typedef struct slot_stats {
+  // How many individual samples were recorded into this slot (un-weighted)
+  uint64_t recorded_samples;
+} stats_slot;
+
+typedef struct profile_slot {
+  ddog_prof_Profile profile;
+  stats_slot stats;
+} profile_slot;
+
 // Contains native state for each instance
 struct stack_recorder_state {
   // Heap recorder instance
   heap_recorder *heap_recorder;
 
-  pthread_mutex_t slot_one_mutex;
-  ddog_prof_Profile slot_one_profile;
-
-  pthread_mutex_t slot_two_mutex;
-  ddog_prof_Profile slot_two_profile;
+  pthread_mutex_t mutex_slot_one;
+  profile_slot profile_slot_one;
+  pthread_mutex_t mutex_slot_two;
+  profile_slot profile_slot_two;
 
   short active_slot; // MUST NEVER BE ACCESSED FROM record_sample; this is NOT for the sampler thread to use.
 
   uint8_t position_for[ALL_VALUE_TYPES_COUNT];
   uint8_t enabled_values_count;
+
+  // Struct for storing stats related to behaviour of a stack recorder instance during its entire lifetime.
+  struct lifetime_stats {
+    // How many profiles have we serialized successfully so far
+    uint64_t serialization_successes;
+    // How many profiles have we serialized unsuccessfully so far
+    uint64_t serialization_failures;
+    // Stats on profile serialization time
+    long serialization_time_ns_min;
+    long serialization_time_ns_max;
+    uint64_t serialization_time_ns_total;
+  } stats_lifetime;
 };
 
-// Used to return a pair of values from sampler_lock_active_profile()
-struct active_slot_pair {
+// Used to group mutex and the corresponding profile slot for easy unlocking after work is done.
+typedef struct locked_profile_slot {
   pthread_mutex_t *mutex;
-  ddog_prof_Profile *profile;
-};
+  profile_slot *data;
+} locked_profile_slot;
 
 struct call_serialize_without_gvl_arguments {
   // Set by caller
@@ -198,8 +221,10 @@ struct call_serialize_without_gvl_arguments {
   ddog_Timespec finish_timestamp;
 
   // Set by callee
-  ddog_prof_Profile *profile;
+  profile_slot *slot;
   ddog_prof_Profile_SerializeResult result;
+  long heap_profile_build_time_ns;
+  long serialize_no_gvl_time_ns;
 
   // Set by both
   bool serialize_ran;
@@ -222,9 +247,9 @@ static VALUE _native_initialize(
 static VALUE _native_serialize(VALUE self, VALUE recorder_instance);
 static VALUE ruby_time_from(ddog_Timespec ddprof_time);
 static void *call_serialize_without_gvl(void *call_args);
-static struct active_slot_pair sampler_lock_active_profile(struct stack_recorder_state *state);
-static void sampler_unlock_active_profile(struct active_slot_pair active_slot);
-static ddog_prof_Profile *serializer_flip_active_and_inactive_slots(struct stack_recorder_state *state);
+static locked_profile_slot sampler_lock_active_profile(struct stack_recorder_state *state);
+static void sampler_unlock_active_profile(locked_profile_slot active_slot);
+static profile_slot* serializer_flip_active_and_inactive_slots(struct stack_recorder_state *state);
 static VALUE _native_active_slot(DDTRACE_UNUSED VALUE _self, VALUE recorder_instance);
 static VALUE _native_is_slot_one_mutex_locked(DDTRACE_UNUSED VALUE _self, VALUE recorder_instance);
 static VALUE _native_is_slot_two_mutex_locked(DDTRACE_UNUSED VALUE _self, VALUE recorder_instance);
@@ -233,7 +258,7 @@ static ddog_Timespec system_epoch_now_timespec(void);
 static VALUE _native_reset_after_fork(DDTRACE_UNUSED VALUE self, VALUE recorder_instance);
 static void serializer_set_start_timestamp_for_next_profile(struct stack_recorder_state *state, ddog_Timespec start_time);
 static VALUE _native_record_endpoint(DDTRACE_UNUSED VALUE _self, VALUE recorder_instance, VALUE local_root_span_id, VALUE endpoint);
-static void reset_profile(ddog_prof_Profile *profile, ddog_Timespec *start_time /* Can be null */);
+static void reset_profile_slot(profile_slot *slot, ddog_Timespec *start_time /* Can be null */);
 static VALUE _native_track_object(DDTRACE_UNUSED VALUE _self, VALUE recorder_instance, VALUE new_obj, VALUE weight, VALUE alloc_class);
 static VALUE _native_check_heap_hashes(DDTRACE_UNUSED VALUE _self, VALUE locations);
 static VALUE _native_start_fake_slow_heap_serialization(DDTRACE_UNUSED VALUE _self, VALUE recorder_instance);
@@ -241,6 +266,8 @@ static VALUE _native_end_fake_slow_heap_serialization(DDTRACE_UNUSED VALUE _self
 static VALUE _native_debug_heap_recorder(DDTRACE_UNUSED VALUE _self, VALUE recorder_instance);
 static VALUE _native_gc_force_recycle(DDTRACE_UNUSED VALUE _self, VALUE obj);
 static VALUE _native_has_seen_id_flag(DDTRACE_UNUSED VALUE _self, VALUE obj);
+static VALUE _native_stats(DDTRACE_UNUSED VALUE self, VALUE instance);
+static VALUE build_profile_stats(profile_slot *slot, long serialization_time_ns, long heap_iteration_prep_time_ns, long heap_profile_build_time_ns);
 
 
 void stack_recorder_init(VALUE profiling_module) {
@@ -261,6 +288,7 @@ void stack_recorder_init(VALUE profiling_module) {
   rb_define_singleton_method(stack_recorder_class, "_native_initialize", _native_initialize, 7);
   rb_define_singleton_method(stack_recorder_class, "_native_serialize",  _native_serialize, 1);
   rb_define_singleton_method(stack_recorder_class, "_native_reset_after_fork", _native_reset_after_fork, 1);
+  rb_define_singleton_method(stack_recorder_class, "_native_stats", _native_stats, 1);
   rb_define_singleton_method(testing_module, "_native_active_slot", _native_active_slot, 1);
   rb_define_singleton_method(testing_module, "_native_slot_one_mutex_locked?", _native_is_slot_one_mutex_locked, 1);
   rb_define_singleton_method(testing_module, "_native_slot_two_mutex_locked?", _native_is_slot_two_mutex_locked, 1);
@@ -305,6 +333,9 @@ static VALUE _native_new(VALUE klass) {
   initialize_slot_concurrency_control(state);
   for (uint8_t i = 0; i < ALL_VALUE_TYPES_COUNT; i++) { state->position_for[i] = all_value_types_positions[i]; }
   state->enabled_values_count = ALL_VALUE_TYPES_COUNT;
+  state->stats_lifetime = (struct lifetime_stats) {
+    .serialization_time_ns_min = INT64_MAX,
+  };
 
   // Note: At this point, slot_one_profile and slot_two_profile contain null pointers. Libdatadog validates pointers
   // before using them so it's ok for us to go ahead and create the StackRecorder object.
@@ -325,11 +356,11 @@ static VALUE _native_new(VALUE klass) {
 }
 
 static void initialize_slot_concurrency_control(struct stack_recorder_state *state) {
-  state->slot_one_mutex = (pthread_mutex_t) PTHREAD_MUTEX_INITIALIZER;
-  state->slot_two_mutex = (pthread_mutex_t) PTHREAD_MUTEX_INITIALIZER;
+  state->mutex_slot_one = (pthread_mutex_t) PTHREAD_MUTEX_INITIALIZER;
+  state->mutex_slot_two = (pthread_mutex_t) PTHREAD_MUTEX_INITIALIZER;
 
   // A newly-created StackRecorder starts with slot one being active for samples, so let's lock slot two
-  ENFORCE_SUCCESS_GVL(pthread_mutex_lock(&state->slot_two_mutex));
+  ENFORCE_SUCCESS_GVL(pthread_mutex_lock(&state->mutex_slot_two));
 
   state->active_slot = 1;
 }
@@ -352,18 +383,22 @@ static void initialize_profiles(struct stack_recorder_state *state, ddog_prof_Sl
     rb_raise(rb_eRuntimeError, "Failed to initialize slot two profile: %"PRIsVALUE, get_error_details_and_drop(&slot_two_profile_result.err));
   }
 
-  state->slot_one_profile = slot_one_profile_result.ok;
-  state->slot_two_profile = slot_two_profile_result.ok;
+  state->profile_slot_one = (profile_slot) {
+    .profile = slot_one_profile_result.ok,
+  };
+  state->profile_slot_two = (profile_slot) {
+    .profile = slot_two_profile_result.ok,
+  };
 }
 
 static void stack_recorder_typed_data_free(void *state_ptr) {
   struct stack_recorder_state *state = (struct stack_recorder_state *) state_ptr;
 
-  pthread_mutex_destroy(&state->slot_one_mutex);
-  ddog_prof_Profile_drop(&state->slot_one_profile);
+  pthread_mutex_destroy(&state->mutex_slot_one);
+  ddog_prof_Profile_drop(&state->profile_slot_one.profile);
 
-  pthread_mutex_destroy(&state->slot_two_mutex);
-  ddog_prof_Profile_drop(&state->slot_two_profile);
+  pthread_mutex_destroy(&state->mutex_slot_two);
+  ddog_prof_Profile_drop(&state->profile_slot_two.profile);
 
   heap_recorder_free(state->heap_recorder);
 
@@ -462,8 +497,8 @@ static VALUE _native_initialize(
     state->position_for[TIMELINE_VALUE_ID] = next_disabled_pos++;
   }
 
-  ddog_prof_Profile_drop(&state->slot_one_profile);
-  ddog_prof_Profile_drop(&state->slot_two_profile);
+  ddog_prof_Profile_drop(&state->profile_slot_one.profile);
+  ddog_prof_Profile_drop(&state->profile_slot_two.profile);
 
   ddog_prof_Slice_ValueType sample_types = {.ptr = enabled_value_types, .len = state->enabled_values_count};
   initialize_profiles(state, sample_types);
@@ -479,9 +514,11 @@ static VALUE _native_serialize(DDTRACE_UNUSED VALUE _self, VALUE recorder_instan
   // Need to do this while still holding on to the Global VM Lock; see comments on method for why
   serializer_set_start_timestamp_for_next_profile(state, finish_timestamp);
 
+  long heap_iteration_prep_start_time_ns = monotonic_wall_time_now_ns(DO_NOT_RAISE_ON_FAILURE);
   // Prepare the iteration on heap recorder we'll be doing outside the GVL. The preparation needs to
   // happen while holding on to the GVL.
   heap_recorder_prepare_iteration(state->heap_recorder);
+  long heap_iteration_prep_time_ns = monotonic_wall_time_now_ns(DO_NOT_RAISE_ON_FAILURE) - heap_iteration_prep_start_time_ns;
 
   // We'll release the Global VM Lock while we're calling serialize, so that the Ruby VM can continue to work while this
   // is pending
@@ -508,12 +545,27 @@ static VALUE _native_serialize(DDTRACE_UNUSED VALUE _self, VALUE recorder_instan
   // Cleanup after heap recorder iteration. This needs to happen while holding on to the GVL.
   heap_recorder_finish_iteration(state->heap_recorder);
 
+  // NOTE: We are focusing on the serialization time outside of the GVL in this stat here. This doesn't
+  //       really cover the full serialization process but it gives a more useful number since it bypasses
+  //       the noise of acquiring GVLs and dealing with interruptions which is highly specific to runtime
+  //       conditions and over which we really have no control about.
+  long serialization_time_ns = args.serialize_no_gvl_time_ns;
+  if (serialization_time_ns >= 0) {
+    // Only update stats if our serialization time is valid.
+    state->stats_lifetime.serialization_time_ns_max = long_max_of(state->stats_lifetime.serialization_time_ns_max, serialization_time_ns);
+    state->stats_lifetime.serialization_time_ns_min = long_min_of(state->stats_lifetime.serialization_time_ns_min, serialization_time_ns);
+    state->stats_lifetime.serialization_time_ns_total += serialization_time_ns;
+  }
+
   ddog_prof_Profile_SerializeResult serialized_profile = args.result;
 
   if (serialized_profile.tag == DDOG_PROF_PROFILE_SERIALIZE_RESULT_ERR) {
+    state->stats_lifetime.serialization_failures++;
     return rb_ary_new_from_args(2, error_symbol, get_error_details_and_drop(&serialized_profile.err));
   }
 
+  state->stats_lifetime.serialization_successes++;
+
   VALUE encoded_pprof = ruby_string_from_vec_u8(serialized_profile.ok.buffer);
 
   ddog_Timespec ddprof_start = serialized_profile.ok.start;
@@ -523,8 +575,9 @@ static VALUE _native_serialize(DDTRACE_UNUSED VALUE _self, VALUE recorder_instan
 
   VALUE start = ruby_time_from(ddprof_start);
   VALUE finish = ruby_time_from(ddprof_finish);
+  VALUE profile_stats = build_profile_stats(args.slot, serialization_time_ns, heap_iteration_prep_time_ns, args.heap_profile_build_time_ns);
 
-  return rb_ary_new_from_args(2, ok_symbol, rb_ary_new_from_args(3, start, finish, encoded_pprof));
+  return rb_ary_new_from_args(2, ok_symbol, rb_ary_new_from_args(4, start, finish, encoded_pprof, profile_stats));
 }
 
 static VALUE ruby_time_from(ddog_Timespec ddprof_time) {
@@ -537,7 +590,7 @@ void record_sample(VALUE recorder_instance, ddog_prof_Slice_Location locations,
   struct stack_recorder_state *state;
   TypedData_Get_Struct(recorder_instance, struct stack_recorder_state, &stack_recorder_typed_data, state);
 
-  struct active_slot_pair active_slot = sampler_lock_active_profile(state);
+  locked_profile_slot active_slot = sampler_lock_active_profile(state);
 
   // Note: We initialize this array to have ALL_VALUE_TYPES_COUNT but only tell libdatadog to use the first
   // state->enabled_values_count values. This simplifies handling disabled value types -- we still put them on the
@@ -561,7 +614,7 @@ void record_sample(VALUE recorder_instance, ddog_prof_Slice_Location locations,
   }
 
   ddog_prof_Profile_Result result = ddog_prof_Profile_add(
-    active_slot.profile,
+    &active_slot.data->profile,
     (ddog_prof_Sample) {
       .locations = locations,
       .values = (ddog_Slice_I64) {.ptr = metric_values, .len = state->enabled_values_count},
@@ -570,6 +623,8 @@ void record_sample(VALUE recorder_instance, ddog_prof_Slice_Location locations,
     labels.end_timestamp_ns
   );
 
+  active_slot.data->stats.recorded_samples++;
+
   sampler_unlock_active_profile(active_slot);
 
   if (result.tag == DDOG_PROF_PROFILE_RESULT_ERR) {
@@ -590,9 +645,9 @@ void record_endpoint(VALUE recorder_instance, uint64_t local_root_span_id, ddog_
   struct stack_recorder_state *state;
   TypedData_Get_Struct(recorder_instance, struct stack_recorder_state, &stack_recorder_typed_data, state);
 
-  struct active_slot_pair active_slot = sampler_lock_active_profile(state);
+  locked_profile_slot active_slot = sampler_lock_active_profile(state);
 
-  ddog_prof_Profile_Result result = ddog_prof_Profile_set_endpoint(active_slot.profile, local_root_span_id, endpoint);
+  ddog_prof_Profile_Result result = ddog_prof_Profile_set_endpoint(&active_slot.data->profile, local_root_span_id, endpoint);
 
   sampler_unlock_active_profile(active_slot);
 
@@ -607,7 +662,7 @@ void record_endpoint(VALUE recorder_instance, uint64_t local_root_span_id, ddog_
 // during iteration of heap recorder live objects.
 typedef struct heap_recorder_iteration_context {
   struct stack_recorder_state *state;
-  ddog_prof_Profile *profile;
+  profile_slot *slot;
 
   bool error;
   char error_msg[MAX_LEN_HEAP_ITERATION_ERROR_MSG];
@@ -643,7 +698,7 @@ static bool add_heap_sample_to_active_profile_without_gvl(heap_recorder_iteratio
   };
 
   ddog_prof_Profile_Result result = ddog_prof_Profile_add(
-    context->profile,
+    &context->slot->profile,
     (ddog_prof_Sample) {
       .locations = iteration_data.locations,
       .values = (ddog_Slice_I64) {.ptr = metric_values, .len = context->state->enabled_values_count},
@@ -655,6 +710,8 @@ static bool add_heap_sample_to_active_profile_without_gvl(heap_recorder_iteratio
     0
   );
 
+  context->slot->stats.recorded_samples++;
+
   if (result.tag == DDOG_PROF_PROFILE_RESULT_ERR) {
     read_ddogerr_string_and_drop(&result.err, context->error_msg, MAX_LEN_HEAP_ITERATION_ERROR_MSG);
     context->error = true;
@@ -666,10 +723,10 @@ static bool add_heap_sample_to_active_profile_without_gvl(heap_recorder_iteratio
   return true;
 }
 
-static void build_heap_profile_without_gvl(struct stack_recorder_state *state, ddog_prof_Profile *profile) {
+static void build_heap_profile_without_gvl(struct stack_recorder_state *state, profile_slot *slot) {
   heap_recorder_iteration_context iteration_context = {
     .state = state,
-    .profile = profile,
+    .slot = slot,
     .error = false,
     .error_msg = {0},
   };
@@ -689,15 +746,21 @@ static void build_heap_profile_without_gvl(struct stack_recorder_state *state, d
 static void *call_serialize_without_gvl(void *call_args) {
   struct call_serialize_without_gvl_arguments *args = (struct call_serialize_without_gvl_arguments *) call_args;
 
-  args->profile = serializer_flip_active_and_inactive_slots(args->state);
+  long serialize_no_gvl_start_time_ns = monotonic_wall_time_now_ns(DO_NOT_RAISE_ON_FAILURE);
+
+  profile_slot *slot_now_inactive = serializer_flip_active_and_inactive_slots(args->state);
+
+  args->slot = slot_now_inactive;
 
   // Now that we have the inactive profile with all but heap samples, lets fill it with heap data
   // without needing to race with the active sampler
-  build_heap_profile_without_gvl(args->state, args->profile);
+  build_heap_profile_without_gvl(args->state, args->slot);
+  args->heap_profile_build_time_ns = monotonic_wall_time_now_ns(DO_NOT_RAISE_ON_FAILURE) - serialize_no_gvl_start_time_ns;
 
   // Note: The profile gets reset by the serialize call
-  args->result = ddog_prof_Profile_serialize(args->profile, &args->finish_timestamp, NULL /* duration_nanos is optional */, NULL /* start_time is optional */);
+  args->result = ddog_prof_Profile_serialize(&args->slot->profile, &args->finish_timestamp, NULL /* duration_nanos is optional */, NULL /* start_time is optional */);
   args->serialize_ran = true;
+  args->serialize_no_gvl_time_ns = monotonic_wall_time_now_ns(DO_NOT_RAISE_ON_FAILURE) - serialize_no_gvl_start_time_ns;
 
   return NULL; // Unused
 }
@@ -707,42 +770,42 @@ VALUE enforce_recorder_instance(VALUE object) {
   return object;
 }
 
-static struct active_slot_pair sampler_lock_active_profile(struct stack_recorder_state *state) {
+static locked_profile_slot sampler_lock_active_profile(struct stack_recorder_state *state) {
   int error;
 
   for (int attempts = 0; attempts < 2; attempts++) {
-    error = pthread_mutex_trylock(&state->slot_one_mutex);
+    error = pthread_mutex_trylock(&state->mutex_slot_one);
     if (error && error != EBUSY) ENFORCE_SUCCESS_GVL(error);
 
     // Slot one is active
-    if (!error) return (struct active_slot_pair) {.mutex = &state->slot_one_mutex, .profile = &state->slot_one_profile};
+    if (!error) return (locked_profile_slot) {.mutex = &state->mutex_slot_one, .data = &state->profile_slot_one};
 
     // If we got here, slot one was not active, let's try slot two
 
-    error = pthread_mutex_trylock(&state->slot_two_mutex);
+    error = pthread_mutex_trylock(&state->mutex_slot_two);
     if (error && error != EBUSY) ENFORCE_SUCCESS_GVL(error);
 
     // Slot two is active
-    if (!error) return (struct active_slot_pair) {.mutex = &state->slot_two_mutex, .profile = &state->slot_two_profile};
+    if (!error) return (locked_profile_slot) {.mutex = &state->mutex_slot_two, .data = &state->profile_slot_two};
   }
 
   // We already tried both multiple times, and we did not succeed. This is not expected to happen. Let's stop sampling.
   rb_raise(rb_eRuntimeError, "Failed to grab either mutex in sampler_lock_active_profile");
 }
 
-static void sampler_unlock_active_profile(struct active_slot_pair active_slot) {
+static void sampler_unlock_active_profile(locked_profile_slot active_slot) {
   ENFORCE_SUCCESS_GVL(pthread_mutex_unlock(active_slot.mutex));
 }
 
-static ddog_prof_Profile *serializer_flip_active_and_inactive_slots(struct stack_recorder_state *state) {
+static profile_slot* serializer_flip_active_and_inactive_slots(struct stack_recorder_state *state) {
   int previously_active_slot = state->active_slot;
 
   if (previously_active_slot != 1 && previously_active_slot != 2) {
     grab_gvl_and_raise(rb_eRuntimeError, "Unexpected active_slot state %d in serializer_flip_active_and_inactive_slots", previously_active_slot);
   }
 
-  pthread_mutex_t *previously_active = (previously_active_slot == 1) ? &state->slot_one_mutex : &state->slot_two_mutex;
-  pthread_mutex_t *previously_inactive = (previously_active_slot == 1) ? &state->slot_two_mutex : &state->slot_one_mutex;
+  pthread_mutex_t *previously_active = (previously_active_slot == 1) ? &state->mutex_slot_one : &state->mutex_slot_two;
+  pthread_mutex_t *previously_inactive = (previously_active_slot == 1) ? &state->mutex_slot_two : &state->mutex_slot_one;
 
   // Release the lock, thus making this slot active
   ENFORCE_SUCCESS_NO_GVL(pthread_mutex_unlock(previously_inactive));
@@ -753,8 +816,8 @@ static ddog_prof_Profile *serializer_flip_active_and_inactive_slots(struct stack
   // Update active_slot
   state->active_slot = (previously_active_slot == 1) ? 2 : 1;
 
-  // Return profile for previously active slot (now inactive)
-  return (previously_active_slot == 1) ? &state->slot_one_profile : &state->slot_two_profile;
+  // Return pointer to previously active slot (now inactive)
+  return (previously_active_slot == 1) ? &state->profile_slot_one : &state->profile_slot_two;
 }
 
 // This method exists only to enable testing Datadog::Profiling::StackRecorder behavior using RSpec.
@@ -778,7 +841,7 @@ static VALUE test_slot_mutex_state(VALUE recorder_instance, int slot) {
   struct stack_recorder_state *state;
   TypedData_Get_Struct(recorder_instance, struct stack_recorder_state, &stack_recorder_typed_data, state);
 
-  pthread_mutex_t *slot_mutex = (slot == 1) ? &state->slot_one_mutex : &state->slot_two_mutex;
+  pthread_mutex_t *slot_mutex = (slot == 1) ? &state->mutex_slot_one : &state->mutex_slot_two;
 
   // Like Heisenberg's uncertainty principle, we can't observe without affecting...
   int error = pthread_mutex_trylock(slot_mutex);
@@ -813,8 +876,8 @@ static VALUE _native_reset_after_fork(DDTRACE_UNUSED VALUE self, VALUE recorder_
   // resulting state is inconsistent, we make sure to reset it back to the initial state.
   initialize_slot_concurrency_control(state);
 
-  reset_profile(&state->slot_one_profile, /* start_time: */ NULL);
-  reset_profile(&state->slot_two_profile, /* start_time: */ NULL);
+  reset_profile_slot(&state->profile_slot_one, /* start_time: */ NULL);
+  reset_profile_slot(&state->profile_slot_two, /* start_time: */ NULL);
 
   heap_recorder_after_fork(state->heap_recorder);
 
@@ -825,8 +888,8 @@ static VALUE _native_reset_after_fork(DDTRACE_UNUSED VALUE self, VALUE recorder_
 // not be interrupted part-way through by a VM fork.
 static void serializer_set_start_timestamp_for_next_profile(struct stack_recorder_state *state, ddog_Timespec start_time) {
   // Before making this profile active, we reset it so that it uses the correct start_time for its start
-  ddog_prof_Profile *next_profile = (state->active_slot == 1) ? &state->slot_two_profile : &state->slot_one_profile;
-  reset_profile(next_profile, &start_time);
+  profile_slot *next_profile_slot = (state->active_slot == 1) ? &state->profile_slot_two : &state->profile_slot_one;
+  reset_profile_slot(next_profile_slot, &start_time);
 }
 
 static VALUE _native_record_endpoint(DDTRACE_UNUSED VALUE _self, VALUE recorder_instance, VALUE local_root_span_id, VALUE endpoint) {
@@ -872,11 +935,12 @@ static VALUE _native_check_heap_hashes(DDTRACE_UNUSED VALUE _self, VALUE locatio
   return Qnil;
 }
 
-static void reset_profile(ddog_prof_Profile *profile, ddog_Timespec *start_time /* Can be null */) {
-  ddog_prof_Profile_Result reset_result = ddog_prof_Profile_reset(profile, start_time);
+static void reset_profile_slot(profile_slot *slot, ddog_Timespec *start_time /* Can be null */) {
+  ddog_prof_Profile_Result reset_result = ddog_prof_Profile_reset(&slot->profile, start_time);
   if (reset_result.tag == DDOG_PROF_PROFILE_RESULT_ERR) {
     rb_raise(rb_eRuntimeError, "Failed to reset profile: %"PRIsVALUE, get_error_details_and_drop(&reset_result.err));
   }
+  slot->stats = (stats_slot) {};
 }
 
 // This method exists only to enable testing Datadog::Profiling::StackRecorder behavior using RSpec.
@@ -937,3 +1001,40 @@ static VALUE _native_has_seen_id_flag(DDTRACE_UNUSED VALUE _self, VALUE obj) {
     return Qfalse;
   #endif
 }
+
+static VALUE _native_stats(DDTRACE_UNUSED VALUE self, VALUE recorder_instance) {
+  struct stack_recorder_state *state;
+  TypedData_Get_Struct(recorder_instance, struct stack_recorder_state, &stack_recorder_typed_data, state);
+
+  uint64_t total_serializations = state->stats_lifetime.serialization_successes + state->stats_lifetime.serialization_failures;
+
+  VALUE heap_recorder_snapshot = state->heap_recorder ?
+    heap_recorder_state_snapshot(state->heap_recorder) : Qnil;
+
+  VALUE stats_as_hash = rb_hash_new();
+  VALUE arguments[] = {
+    ID2SYM(rb_intern("serialization_successes")), /* => */ ULL2NUM(state->stats_lifetime.serialization_successes),
+    ID2SYM(rb_intern("serialization_failures")),  /* => */ ULL2NUM(state->stats_lifetime.serialization_failures),
+
+    ID2SYM(rb_intern("serialization_time_ns_min")),   /* => */ RUBY_NUM_OR_NIL(state->stats_lifetime.serialization_time_ns_min, != INT64_MAX, LONG2NUM),
+    ID2SYM(rb_intern("serialization_time_ns_max")),   /* => */ RUBY_NUM_OR_NIL(state->stats_lifetime.serialization_time_ns_max, > 0, LONG2NUM),
+    ID2SYM(rb_intern("serialization_time_ns_total")), /* => */ RUBY_NUM_OR_NIL(state->stats_lifetime.serialization_time_ns_total, > 0, LONG2NUM),
+    ID2SYM(rb_intern("serialization_time_ns_avg")),   /* => */ RUBY_AVG_OR_NIL(state->stats_lifetime.serialization_time_ns_total, total_serializations),
+
+    ID2SYM(rb_intern("heap_recorder_snapshot")), /* => */ heap_recorder_snapshot,
+  };
+  for (long unsigned int i = 0; i < VALUE_COUNT(arguments); i += 2) rb_hash_aset(stats_as_hash, arguments[i], arguments[i+1]);
+  return stats_as_hash;
+}
+
+static VALUE build_profile_stats(profile_slot *slot, long serialization_time_ns, long heap_iteration_prep_time_ns, long heap_profile_build_time_ns) {
+  VALUE stats_as_hash = rb_hash_new();
+  VALUE arguments[] = {
+    ID2SYM(rb_intern("recorded_samples")), /* => */ ULL2NUM(slot->stats.recorded_samples),
+    ID2SYM(rb_intern("serialization_time_ns")), /* => */ LONG2NUM(serialization_time_ns),
+    ID2SYM(rb_intern("heap_iteration_prep_time_ns")), /* => */ LONG2NUM(heap_iteration_prep_time_ns),
+    ID2SYM(rb_intern("heap_profile_build_time_ns")), /* => */ LONG2NUM(heap_profile_build_time_ns),
+  };
+  for (long unsigned int i = 0; i < VALUE_COUNT(arguments); i += 2) rb_hash_aset(stats_as_hash, arguments[i], arguments[i+1]);
+  return stats_as_hash;
+}

data/lib/datadog/core/telemetry/emitter.rb

@@ -24,7 +24,7 @@ module Datadog
             seq_id = self.class.sequence.next
             payload = Request.build_payload(event, seq_id)
             res = @http_transport.request(request_type: event.type, payload: payload.to_json)
-            Datadog.logger.debug { "Telemetry sent for event `#{event.type}` (status code: #{res.code})" }
+            Datadog.logger.debug { "Telemetry sent for event `#{event.type}` (code: #{res.code.inspect})" }
             res
           rescue => e
             Datadog.logger.debug("Unable to send telemetry request for event `#{event.type rescue 'unknown'}`: #{e}")

data/lib/datadog/core/telemetry/http/response.rb

@@ -32,6 +32,10 @@ module Datadog
             nil
           end
 
+          def code
+            nil
+          end
+
           def inspect
             "#{self.class} ok?:#{ok?} unsupported?:#{unsupported?}, " \
             "not_found?:#{not_found?}, client_error?:#{client_error?}, " \

data/lib/datadog/profiling/exporter.rb

@@ -52,10 +52,11 @@ module Datadog
 
       def flush
         worker_stats = @worker.stats_and_reset_not_thread_safe
-        start, finish, compressed_pprof = pprof_recorder.serialize
-        @last_flush_finish_at = finish
+        serialization_result = pprof_recorder.serialize
+        return if serialization_result.nil?
 
-        return if compressed_pprof.nil? # We don't want to report empty profiles
+        start, finish, compressed_pprof, profile_stats = serialization_result
+        @last_flush_finish_at = finish
 
         if duration_below_threshold?(start, finish)
           Datadog.logger.debug('Skipped exporting profiling events as profile duration is below minimum')
@@ -75,6 +76,8 @@ module Datadog
           internal_metadata: internal_metadata.merge(
             {
               worker_stats: worker_stats,
+              profile_stats: profile_stats,
+              recorder_stats: pprof_recorder.stats,
               gc: GC.stat,
             }
           ),

data/lib/datadog/profiling/stack_recorder.rb

@@ -31,11 +31,11 @@ module Datadog
         status, result = @no_concurrent_synchronize_mutex.synchronize { self.class._native_serialize(self) }
 
         if status == :ok
-          start, finish, encoded_pprof = result
+          start, finish, encoded_pprof, profile_stats = result
 
           Datadog.logger.debug { "Encoded profile covering #{start.iso8601} to #{finish.iso8601}" }
 
-          [start, finish, encoded_pprof]
+          [start, finish, encoded_pprof, profile_stats]
         else
           error_message = result
 
@@ -62,6 +62,10 @@ module Datadog
       def reset_after_fork
         self.class._native_reset_after_fork(self)
       end
+
+      def stats
+        self.class._native_stats(self)
+      end
     end
   end
 end

data/lib/ddtrace/version.rb

@@ -3,7 +3,7 @@
 module DDTrace
   module VERSION
     MAJOR = 1
-    MINOR = 22
+    MINOR = 23
     PATCH = 0
     PRE = nil
     BUILD = nil

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: ddtrace
 version: !ruby/object:Gem::Version
-  version: 1.22.0
+  version: 1.23.0
 platform: ruby
 authors:
 - Datadog, Inc.
-autorequire:
+autorequire: 
 bindir: bin
 cert_chain: []
-date: 2024-04-16 00:00:00.000000000 Z
+date: 2024-05-09 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: msgpack
@@ -879,7 +879,7 @@ licenses:
 metadata:
   allowed_push_host: https://rubygems.org
   changelog_uri: https://github.com/DataDog/dd-trace-rb/blob/master/CHANGELOG.md
-post_install_message:
+post_install_message: 
 rdoc_options: []
 require_paths:
 - lib
@@ -897,8 +897,8 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: 2.0.0
 requirements: []
-rubygems_version: 3.5.3
-signing_key:
+rubygems_version: 3.4.10
+signing_key: 
 specification_version: 4
 summary: Datadog tracing code for your Ruby applications
 test_files: []