datadog 2.4.0 → 2.5.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ba1dc00d9afe70c54ea0918673e832793bb7c7a629618dc3d36e066fd121ea56
-  data.tar.gz: 9818bf0ba8dbd989451b7b0492efe848a9cdf97239acac3637aaa9d17937b541
+  metadata.gz: c1aa5227f68791f80fe6deb7f5a8b53e5adff15a7e443f78788f827888945c83
+  data.tar.gz: 7931209b3685744b1aa17f709766d9a5da91b5b34987a8ac975032ebee0a7caa
 SHA512:
-  metadata.gz: 419ac80497f45159d199d2b9d3b16c70c46903d65d828c6e5ab52029758c9b38e11b243e112d52e3e9d45f05d5e1acf814bcb3f7582496f1e5be120c13f0f685
-  data.tar.gz: ce8f36e974b194fbdedd40fdf634e49115c0bd6befb350092dfc4cb94185a173a7d3325b6adc9cf50e060b05a4968ca7858a52082748a4979090a7eb32aa96ba
+  metadata.gz: '081740cc882d48f62d18f8a823696339346ad574f4ac5c67cd64430ee37f347ec33d6c536e3790f3697ea3a6cb72f2275a6b5b14808e48a4246cc20786fcc20b'
+  data.tar.gz: c51018d1655343c234e45d72ab5dc9a27ebc465e253462eec483faf862e3c6fde0bbba6d4f3c1ca6656a6d7c33525a5942481a69beee7a7d889cbbe75c900ff7

data/CHANGELOG.md

@@ -2,6 +2,24 @@
 
 ## [Unreleased]
 
+## [2.5.0] - 2024-11-05
+
+### Added
+
+* Performance: Profiling: Add setting to lower heap profiling memory use/latency by cleaning up young objects after Ruby GC ([#4020][])
+
+### Changed
+
+* Core: Replace the `debase-ruby_core_source` gem with the `datadog-ruby_core_source` ([#4014][])
+* Core: Upgrade to `libdatadog` 13.1 ([#3997][])
+
+### Fixed
+
+* Fix `undefined method` error for Rails runner ([#3996][])
+* Apply version tag only to spans that use the global/default service name ([#4027][])
+* Ensure UDS takes precedence over HTTP when both Agent configurations defined ([#4024][])
+* Remove duplicate leading slash in resource name for Grape routes ([#4033][])
+
 ## [2.4.0] - 2024-10-11
 
 ### Added
@@ -2985,7 +3003,8 @@ Release notes: https://github.com/DataDog/dd-trace-rb/releases/tag/v0.3.1
 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/v2.4.0...master
+[Unreleased]: https://github.com/DataDog/dd-trace-rb/compare/v2.5.0...master
+[2.5.0]: https://github.com/DataDog/dd-trace-rb/compare/v2.4.0...v2.5.0
 [2.4.0]: https://github.com/DataDog/dd-trace-rb/compare/v2.3.0...v2.4.0
 [2.3.0]: https://github.com/DataDog/dd-trace-rb/compare/v2.2.0...v2.3.0
 [2.2.0]: https://github.com/DataDog/dd-trace-rb/compare/v2.1.0...v2.2.0
@@ -4419,6 +4438,13 @@ Git diff: https://github.com/DataDog/dd-trace-rb/compare/v0.3.0...v0.3.1
 [#3969]: https://github.com/DataDog/dd-trace-rb/issues/3969
 [#3970]: https://github.com/DataDog/dd-trace-rb/issues/3970
 [#3984]: https://github.com/DataDog/dd-trace-rb/issues/3984
+[#3996]: https://github.com/DataDog/dd-trace-rb/issues/3996
+[#3997]: https://github.com/DataDog/dd-trace-rb/issues/3997
+[#4014]: https://github.com/DataDog/dd-trace-rb/issues/4014
+[#4020]: https://github.com/DataDog/dd-trace-rb/issues/4020
+[#4024]: https://github.com/DataDog/dd-trace-rb/issues/4024
+[#4027]: https://github.com/DataDog/dd-trace-rb/issues/4027
+[#4033]: https://github.com/DataDog/dd-trace-rb/issues/4033
 [@AdrianLC]: https://github.com/AdrianLC
 [@Azure7111]: https://github.com/Azure7111
 [@BabyGroot]: https://github.com/BabyGroot
@@ -4570,4 +4596,4 @@ Git diff: https://github.com/DataDog/dd-trace-rb/compare/v0.3.0...v0.3.1
 [@y-yagi]: https://github.com/y-yagi
 [@yujideveloper]: https://github.com/yujideveloper
 [@yukimurasawa]: https://github.com/yukimurasawa
-[@zachmccormick]: https://github.com/zachmccormick
+[@zachmccormick]: https://github.com/zachmccormick

data/ext/datadog_profiling_native_extension/NativeExtensionDesign.md

@@ -68,7 +68,7 @@ internal types, structures and functions).
 Because these private header files are not included in regular Ruby installations, we have two different workarounds:
 
 1. for Ruby versions 2.6 to 3.2 we make use use the Ruby private MJIT header
-2. for Ruby versions < 2.6 and > 3.2 we make use of the `debase-ruby_core_source` gem
+2. for Ruby versions < 2.6 and > 3.2 we make use of the `datadog-ruby_core_source` gem
 
 Functions which make use of these headers are defined in the <private_vm_api_acccess.c> file.
 
@@ -91,9 +91,9 @@ version. e.g. `rb_mjit_min_header-2.7.4.h`.
 
 This header was removed in Ruby 3.3.
 
-### Approach 2: Using the `debase-ruby_core_source` gem
+### Approach 2: Using the `datadog-ruby_core_source` gem
 
-The [`debase-ruby_core_source`](https://github.com/ruby-debug/debase-ruby_core_source) contains almost no code;
+The [`datadog-ruby_core_source`](https://github.com/DataDog/datadog-ruby_core_source) contains almost no code;
 instead, it just contains per-Ruby-version folders with the private VM headers (`.h`) files for that version.
 
 Thus, even though a regular Ruby installation does not include these files, we can access the copy inside this gem.

data/ext/datadog_profiling_native_extension/collectors_cpu_and_wall_time_worker.c

@@ -17,13 +17,6 @@
 #include "setup_signal_handler.h"
 #include "time_helpers.h"
 
-#define ERR_CLOCK_FAIL "failed to get clock time"
-
-// Maximum allowed value for an allocation weight. Attempts to use higher values will result in clamping.
-// See https://docs.google.com/document/d/1lWLB714wlLBBq6T4xZyAc4a5wtWhSmr4-hgiPKeErlA/edit#heading=h.ugp0zxcj5iqh
-// (Datadog-only link) for research backing the choice of this value.
-unsigned int MAX_ALLOC_WEIGHT = 10000;
-
 // Used to trigger the execution of Collectors::ThreadState, which implements all of the sampling logic
 // itself; this class only implements the "when to do it" part.
 //
@@ -83,6 +76,13 @@ unsigned int MAX_ALLOC_WEIGHT = 10000;
 //
 // ---
 
+#define ERR_CLOCK_FAIL "failed to get clock time"
+
+// Maximum allowed value for an allocation weight. Attempts to use higher values will result in clamping.
+// See https://docs.google.com/document/d/1lWLB714wlLBBq6T4xZyAc4a5wtWhSmr4-hgiPKeErlA/edit#heading=h.ugp0zxcj5iqh
+// (Datadog-only link) for research backing the choice of this value.
+unsigned int MAX_ALLOC_WEIGHT = 10000;
+
 #ifndef NO_POSTPONED_TRIGGER
   // Used to call the rb_postponed_job_trigger from Ruby 3.3+. These get initialized in
   // `collectors_cpu_and_wall_time_worker_init` below and always get reused after that.
@@ -180,6 +180,12 @@ struct cpu_and_wall_time_worker_state {
     // # GVL profiling stats
     // How many times we triggered the after_gvl_running sampling
     unsigned int after_gvl_running;
+    // How many times we skipped the after_gvl_running sampling
+    unsigned int gvl_dont_sample;
+    // Min/max/total wall-time spent on gvl sampling
+    uint64_t gvl_sampling_time_ns_min;
+    uint64_t gvl_sampling_time_ns_max;
+    uint64_t gvl_sampling_time_ns_total;
   } stats;
 };
 
@@ -230,6 +236,7 @@ static VALUE _native_resume_signals(DDTRACE_UNUSED VALUE self);
 static void on_gvl_event(rb_event_flag_t event_id, const rb_internal_thread_event_data_t *event_data, DDTRACE_UNUSED void *_unused);
 static void after_gvl_running_from_postponed_job(DDTRACE_UNUSED void *_unused);
 #endif
+static VALUE rescued_after_gvl_running_from_postponed_job(VALUE self_instance);
 static VALUE _native_gvl_profiling_hook_active(DDTRACE_UNUSED VALUE self, VALUE instance);
 
 // We're using `on_newobj_event` function with `rb_add_event_hook2`, which requires in its public signature a function
@@ -315,8 +322,6 @@ void collectors_cpu_and_wall_time_worker_init(VALUE profiling_module) {
   rb_define_singleton_method(testing_module, "_native_current_sigprof_signal_handler", _native_current_sigprof_signal_handler, 0);
   rb_define_singleton_method(collectors_cpu_and_wall_time_worker_class, "_native_hold_signals", _native_hold_signals, 0);
   rb_define_singleton_method(collectors_cpu_and_wall_time_worker_class, "_native_resume_signals", _native_resume_signals, 0);
-  // TODO: Remove `_native_is_running` from `testing_module` (should be in class) once `prof-correctness` has been updated to not need it
-  rb_define_singleton_method(testing_module, "_native_is_running?", _native_is_running, 1);
   rb_define_singleton_method(testing_module, "_native_install_testing_signal_handler", _native_install_testing_signal_handler, 0);
   rb_define_singleton_method(testing_module, "_native_remove_testing_signal_handler", _native_remove_testing_signal_handler, 0);
   rb_define_singleton_method(testing_module, "_native_trigger_sample", _native_trigger_sample, 0);
@@ -1042,7 +1047,12 @@ static VALUE _native_stats(DDTRACE_UNUSED VALUE self, VALUE instance) {
     ID2SYM(rb_intern("allocations_during_sample")),         /* => */ state->allocation_profiling_enabled ? UINT2NUM(state->stats.allocations_during_sample) : Qnil,
 
     // GVL profiling stats
-    ID2SYM(rb_intern("after_gvl_running")), /* => */ UINT2NUM(state->stats.after_gvl_running),
+    ID2SYM(rb_intern("after_gvl_running")),          /* => */ UINT2NUM(state->stats.after_gvl_running),
+    ID2SYM(rb_intern("gvl_dont_sample")),            /* => */ UINT2NUM(state->stats.gvl_dont_sample),
+    ID2SYM(rb_intern("gvl_sampling_time_ns_min")),   /* => */ RUBY_NUM_OR_NIL(state->stats.gvl_sampling_time_ns_min, != UINT64_MAX, ULL2NUM),
+    ID2SYM(rb_intern("gvl_sampling_time_ns_max")),   /* => */ RUBY_NUM_OR_NIL(state->stats.gvl_sampling_time_ns_max, > 0, ULL2NUM),
+    ID2SYM(rb_intern("gvl_sampling_time_ns_total")), /* => */ RUBY_NUM_OR_NIL(state->stats.gvl_sampling_time_ns_total, > 0, ULL2NUM),
+    ID2SYM(rb_intern("gvl_sampling_time_ns_avg")),   /* => */ RUBY_AVG_OR_NIL(state->stats.gvl_sampling_time_ns_total, state->stats.after_gvl_running),
   };
   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;
@@ -1080,8 +1090,10 @@ static void reset_stats_not_thread_safe(struct cpu_and_wall_time_worker_state *s
   //       Given the expected infrequency of resetting (~once per 60s profile) and the auxiliary/non-critical nature of these stats
   //       this momentary loss of accuracy is deemed acceptable to keep overhead to a minimum.
   state->stats = (struct stats) {
-    .cpu_sampling_time_ns_min = UINT64_MAX, // Since we always take the min between existing and latest sample
-    .allocation_sampling_time_ns_min = UINT64_MAX, // Since we always take the min between existing and latest sample
+    // All these values are initialized to their highest value possible since we always take the min between existing and latest sample
+    .cpu_sampling_time_ns_min        = UINT64_MAX,
+    .allocation_sampling_time_ns_min = UINT64_MAX,
+    .gvl_sampling_time_ns_min        = UINT64_MAX,
   };
 }
 
@@ -1315,23 +1327,29 @@ static VALUE _native_resume_signals(DDTRACE_UNUSED VALUE self) {
     } else if (event_id == RUBY_INTERNAL_THREAD_EVENT_RESUMED) { /* running/runnable */
       // Interesting note: A RUBY_INTERNAL_THREAD_EVENT_RESUMED is guaranteed to be called with the GVL being acquired.
       // (And... I think target_thread will be == rb_thread_current()?)
+      //
       // But we're not sure if we're on the main Ractor yet. The thread context collector actually can actually help here:
       // it tags threads it's tracking, so if a thread is tagged then by definition we know that thread belongs to the main
-      // Ractor. Thus, if we really really wanted to access the state, we could do it after making sure we're on the correct Ractor.
+      // Ractor. Thus, if we get a ON_GVL_RUNNING_UNKNOWN result we shouldn't touch any state, but otherwise we're good to go.
 
       #ifdef USE_GVL_PROFILING_3_2_WORKAROUNDS
         target_thread = gvl_profiling_state_maybe_initialize();
       #endif
 
-      bool should_sample = thread_context_collector_on_gvl_running(target_thread);
+      on_gvl_running_result result = thread_context_collector_on_gvl_running(target_thread);
 
-      if (should_sample) {
-        // should_sample is only true if a thread belongs to the main Ractor, so we're good to go
+      if (result == ON_GVL_RUNNING_SAMPLE) {
         #ifndef NO_POSTPONED_TRIGGER
           rb_postponed_job_trigger(after_gvl_running_from_postponed_job_handle);
         #else
           rb_postponed_job_register_one(0, after_gvl_running_from_postponed_job, NULL);
         #endif
+      } else if (result == ON_GVL_RUNNING_DONT_SAMPLE) {
+        struct cpu_and_wall_time_worker_state *state = active_sampler_instance_state; // Read from global variable, see "sampler global state safety" note above
+
+        if (state == NULL) return; // This should not happen, but just in case...
+
+        state->stats.gvl_dont_sample++;
       }
     } else {
       // This is a very delicate time and it's hard for us to raise an exception so let's at least complain to stderr
@@ -1347,11 +1365,32 @@ static VALUE _native_resume_signals(DDTRACE_UNUSED VALUE self) {
 
     state->during_sample = true;
 
-    safely_call(thread_context_collector_sample_after_gvl_running, state->thread_context_collector_instance, state->self_instance);
+    // Rescue against any exceptions that happen during sampling
+    safely_call(rescued_after_gvl_running_from_postponed_job, state->self_instance, state->self_instance);
+
+    state->during_sample = false;
+  }
+
+  static VALUE rescued_after_gvl_running_from_postponed_job(VALUE self_instance) {
+    struct cpu_and_wall_time_worker_state *state;
+    TypedData_Get_Struct(self_instance, struct cpu_and_wall_time_worker_state, &cpu_and_wall_time_worker_typed_data, state);
+
+    long wall_time_ns_before_sample = monotonic_wall_time_now_ns(RAISE_ON_FAILURE);
+    thread_context_collector_sample_after_gvl_running(state->thread_context_collector_instance, rb_thread_current(), wall_time_ns_before_sample);
+    long wall_time_ns_after_sample = monotonic_wall_time_now_ns(RAISE_ON_FAILURE);
+
+    long delta_ns = wall_time_ns_after_sample - wall_time_ns_before_sample;
+
+    // Guard against wall-time going backwards, see https://github.com/DataDog/dd-trace-rb/pull/2336 for discussion.
+    uint64_t sampling_time_ns = delta_ns < 0 ? 0 : delta_ns;
+
+    state->stats.gvl_sampling_time_ns_min = uint64_min_of(sampling_time_ns, state->stats.gvl_sampling_time_ns_min);
+    state->stats.gvl_sampling_time_ns_max = uint64_max_of(sampling_time_ns, state->stats.gvl_sampling_time_ns_max);
+    state->stats.gvl_sampling_time_ns_total += sampling_time_ns;
 
     state->stats.after_gvl_running++;
 
-    state->during_sample = false;
+    return Qnil;
   }
 
   static VALUE _native_gvl_profiling_hook_active(DDTRACE_UNUSED VALUE self, VALUE instance) {

data/ext/datadog_profiling_native_extension/collectors_thread_context.c

@@ -109,7 +109,7 @@ static ID otel_context_storage_id; // id of :__opentelemetry_context_storage__ i
 // and that'll be the one that last wrote this setting.
 static uint32_t global_waiting_for_gvl_threshold_ns = MILLIS_AS_NS(10);
 
-enum otel_context_enabled {otel_context_enabled_false, otel_context_enabled_only, otel_context_enabled_both};
+typedef enum { OTEL_CONTEXT_ENABLED_FALSE, OTEL_CONTEXT_ENABLED_ONLY, OTEL_CONTEXT_ENABLED_BOTH } otel_context_enabled;
 
 // Contains state for a single ThreadContext instance
 struct thread_context_collector_state {
@@ -138,9 +138,7 @@ struct thread_context_collector_state {
   // Used to omit timestamps / timeline events from collected data
   bool timeline_enabled;
   // Used to control context collection
-  enum otel_context_enabled otel_context_enabled;
-  // Used to omit class information from collected allocation data
-  bool allocation_type_enabled;
+  otel_context_enabled otel_context_enabled;
   // Used when calling monotonic_to_system_epoch_ns
   monotonic_to_system_epoch_state time_converter_state;
   // Used to identify the main thread, to give it a fallback name
@@ -204,18 +202,7 @@ static void thread_context_collector_typed_data_free(void *state_ptr);
 static int hash_map_per_thread_context_mark(st_data_t key_thread, st_data_t _value, st_data_t _argument);
 static int hash_map_per_thread_context_free_values(st_data_t _thread, st_data_t value_per_thread_context, st_data_t _argument);
 static VALUE _native_new(VALUE klass);
-static VALUE _native_initialize(
-  VALUE self,
-  VALUE collector_instance,
-  VALUE recorder_instance,
-  VALUE max_frames,
-  VALUE tracer_context_key,
-  VALUE endpoint_collection_enabled,
-  VALUE timeline_enabled,
-  VALUE waiting_for_gvl_threshold_ns,
-  VALUE otel_context_enabled,
-  VALUE allocation_type_enabled
-);
+static VALUE _native_initialize(int argc, VALUE *argv, DDTRACE_UNUSED VALUE _self);
 static VALUE _native_sample(VALUE self, VALUE collector_instance, VALUE profiler_overhead_stack_thread);
 static VALUE _native_on_gc_start(VALUE self, VALUE collector_instance);
 static VALUE _native_on_gc_finish(VALUE self, VALUE collector_instance);
@@ -312,7 +299,7 @@ void collectors_thread_context_init(VALUE profiling_module) {
   // https://bugs.ruby-lang.org/issues/18007 for a discussion around this.
   rb_define_alloc_func(collectors_thread_context_class, _native_new);
 
-  rb_define_singleton_method(collectors_thread_context_class, "_native_initialize", _native_initialize, 9);
+  rb_define_singleton_method(collectors_thread_context_class, "_native_initialize", _native_initialize, -1);
   rb_define_singleton_method(collectors_thread_context_class, "_native_inspect", _native_inspect, 1);
   rb_define_singleton_method(collectors_thread_context_class, "_native_reset_after_fork", _native_reset_after_fork, 1);
   rb_define_singleton_method(testing_module, "_native_sample", _native_sample, 2);
@@ -435,8 +422,7 @@ static VALUE _native_new(VALUE klass) {
   state->thread_list_buffer = thread_list_buffer;
   state->endpoint_collection_enabled = true;
   state->timeline_enabled = true;
-  state->otel_context_enabled = otel_context_enabled_false;
-  state->allocation_type_enabled = true;
+  state->otel_context_enabled = OTEL_CONTEXT_ENABLED_FALSE;
   state->time_converter_state = (monotonic_to_system_epoch_state) MONOTONIC_TO_SYSTEM_EPOCH_INITIALIZER;
   VALUE main_thread = rb_thread_main();
   state->main_thread = main_thread;
@@ -456,26 +442,27 @@ static VALUE _native_new(VALUE klass) {
   return instance;
 }
 
-// TODO: Convert this to use options like CpuAndWallTimeWorker
-static VALUE _native_initialize(
-  DDTRACE_UNUSED VALUE _self,
-  VALUE collector_instance,
-  VALUE recorder_instance,
-  VALUE max_frames,
-  VALUE tracer_context_key,
-  VALUE endpoint_collection_enabled,
-  VALUE timeline_enabled,
-  VALUE waiting_for_gvl_threshold_ns,
-  VALUE otel_context_enabled,
-  VALUE allocation_type_enabled
-) {
+static VALUE _native_initialize(int argc, VALUE *argv, DDTRACE_UNUSED VALUE _self) {
+  VALUE options;
+  rb_scan_args(argc, argv, "0:", &options);
+  if (options == Qnil) options = rb_hash_new();
+
+  VALUE self_instance = rb_hash_fetch(options, ID2SYM(rb_intern("self_instance")));
+  VALUE recorder_instance = rb_hash_fetch(options, ID2SYM(rb_intern("recorder")));
+  VALUE max_frames = rb_hash_fetch(options, ID2SYM(rb_intern("max_frames")));
+  VALUE tracer_context_key = rb_hash_fetch(options, ID2SYM(rb_intern("tracer_context_key")));
+  VALUE endpoint_collection_enabled = rb_hash_fetch(options, ID2SYM(rb_intern("endpoint_collection_enabled")));
+  VALUE timeline_enabled = rb_hash_fetch(options, ID2SYM(rb_intern("timeline_enabled")));
+  VALUE waiting_for_gvl_threshold_ns = rb_hash_fetch(options, ID2SYM(rb_intern("waiting_for_gvl_threshold_ns")));
+  VALUE otel_context_enabled = rb_hash_fetch(options, ID2SYM(rb_intern("otel_context_enabled")));
+
+  ENFORCE_TYPE(max_frames, T_FIXNUM);
   ENFORCE_BOOLEAN(endpoint_collection_enabled);
   ENFORCE_BOOLEAN(timeline_enabled);
   ENFORCE_TYPE(waiting_for_gvl_threshold_ns, T_FIXNUM);
-  ENFORCE_BOOLEAN(allocation_type_enabled);
 
   struct thread_context_collector_state *state;
-  TypedData_Get_Struct(collector_instance, struct thread_context_collector_state, &thread_context_collector_typed_data, state);
+  TypedData_Get_Struct(self_instance, struct thread_context_collector_state, &thread_context_collector_typed_data, state);
 
   // Update this when modifying state struct
   state->max_frames = sampling_buffer_check_max_frames(NUM2INT(max_frames));
@@ -485,15 +472,14 @@ static VALUE _native_initialize(
   state->endpoint_collection_enabled = (endpoint_collection_enabled == Qtrue);
   state->timeline_enabled = (timeline_enabled == Qtrue);
   if (otel_context_enabled == Qfalse || otel_context_enabled == Qnil) {
-    state->otel_context_enabled = otel_context_enabled_false;
+    state->otel_context_enabled = OTEL_CONTEXT_ENABLED_FALSE;
   } else if (otel_context_enabled == ID2SYM(rb_intern("only"))) {
-    state->otel_context_enabled = otel_context_enabled_only;
+    state->otel_context_enabled = OTEL_CONTEXT_ENABLED_ONLY;
   } else if (otel_context_enabled == ID2SYM(rb_intern("both"))) {
-    state->otel_context_enabled = otel_context_enabled_both;
+    state->otel_context_enabled = OTEL_CONTEXT_ENABLED_BOTH;
   } else {
     rb_raise(rb_eArgError, "Unexpected value for otel_context_enabled: %+" PRIsVALUE, otel_context_enabled);
   }
-  state->allocation_type_enabled = (allocation_type_enabled == Qtrue);
 
   global_waiting_for_gvl_threshold_ns = NUM2UINT(waiting_for_gvl_threshold_ns);
 
@@ -829,6 +815,9 @@ VALUE thread_context_collector_sample_after_gc(VALUE self_instance) {
 
   state->stats.gc_samples++;
 
+  // Let recorder do any cleanup/updates it requires after a GC step.
+  recorder_after_gc_step(state->recorder_instance);
+
   // Return a VALUE to make it easier to call this function from Ruby APIs that expect a return value (such as rb_rescue2)
   return Qnil;
 }
@@ -885,7 +874,7 @@ static void trigger_sample_for_thread(
   struct trace_identifiers trace_identifiers_result = {.valid = false, .trace_endpoint = Qnil};
   trace_identifiers_for(state, thread, &trace_identifiers_result);
 
-  if (!trace_identifiers_result.valid && state->otel_context_enabled != otel_context_enabled_false) {
+  if (!trace_identifiers_result.valid && state->otel_context_enabled != OTEL_CONTEXT_ENABLED_FALSE) {
     // If we couldn't get something with ddtrace, let's see if we can get some trace identifiers from opentelemetry directly
     otel_without_ddtrace_trace_identifiers_for(state, thread, &trace_identifiers_result);
   }
@@ -1111,7 +1100,6 @@ static VALUE _native_inspect(DDTRACE_UNUSED VALUE _self, VALUE collector_instanc
   rb_str_concat(result, rb_sprintf(" endpoint_collection_enabled=%"PRIsVALUE, state->endpoint_collection_enabled ? Qtrue : Qfalse));
   rb_str_concat(result, rb_sprintf(" timeline_enabled=%"PRIsVALUE, state->timeline_enabled ? Qtrue : Qfalse));
   rb_str_concat(result, rb_sprintf(" otel_context_enabled=%d", state->otel_context_enabled));
-  rb_str_concat(result, rb_sprintf(" allocation_type_enabled=%"PRIsVALUE, state->allocation_type_enabled ? Qtrue : Qfalse));
   rb_str_concat(result, rb_sprintf(
     " time_converter_state={.system_epoch_ns_reference=%ld, .delta_to_epoch_ns=%ld}",
     state->time_converter_state.system_epoch_ns_reference,
@@ -1302,7 +1290,7 @@ static VALUE _native_gc_tracking(DDTRACE_UNUSED VALUE _self, VALUE collector_ins
 
 // Assumption 1: This function is called in a thread that is holding the Global VM Lock. Caller is responsible for enforcing this.
 static void trace_identifiers_for(struct thread_context_collector_state *state, VALUE thread, struct trace_identifiers *trace_identifiers_result) {
-  if (state->otel_context_enabled == otel_context_enabled_only) return;
+  if (state->otel_context_enabled == OTEL_CONTEXT_ENABLED_ONLY) return;
   if (state->tracer_context_key == MISSING_TRACER_CONTEXT_KEY) return;
 
   VALUE current_context = rb_thread_local_aref(thread, state->tracer_context_key);
@@ -1415,62 +1403,61 @@ void thread_context_collector_sample_allocation(VALUE self_instance, unsigned in
   ddog_CharSlice *optional_class_name = NULL;
   char imemo_type[100];
 
-  if (state->allocation_type_enabled) {
-    optional_class_name = &class_name;
-
-    if (
-      type == RUBY_T_OBJECT   ||
-      type == RUBY_T_CLASS    ||
-      type == RUBY_T_MODULE   ||
-      type == RUBY_T_FLOAT    ||
-      type == RUBY_T_STRING   ||
-      type == RUBY_T_REGEXP   ||
-      type == RUBY_T_ARRAY    ||
-      type == RUBY_T_HASH     ||
-      type == RUBY_T_STRUCT   ||
-      type == RUBY_T_BIGNUM   ||
-      type == RUBY_T_FILE     ||
-      type == RUBY_T_DATA     ||
-      type == RUBY_T_MATCH    ||
-      type == RUBY_T_COMPLEX  ||
-      type == RUBY_T_RATIONAL ||
-      type == RUBY_T_NIL      ||
-      type == RUBY_T_TRUE     ||
-      type == RUBY_T_FALSE    ||
-      type == RUBY_T_SYMBOL   ||
-      type == RUBY_T_FIXNUM
-    ) {
-      VALUE klass = rb_class_of(new_object);
-
-      // Ruby sometimes plays a bit fast and loose with some of its internal objects, e.g.
-      // `rb_str_tmp_frozen_acquire` allocates a string with no class (klass=0).
-      // Thus, we need to make sure there's actually a class before getting its name.
-
-      if (klass != 0) {
-        const char *name = rb_class2name(klass);
-        size_t name_length = name != NULL ? strlen(name) : 0;
-
-        if (name_length > 0) {
-          class_name = (ddog_CharSlice) {.ptr = name, .len = name_length};
-        } else {
-          // @ivoanjo: I'm not sure this can ever happen, but just-in-case
-          class_name = ruby_value_type_to_class_name(type);
-        }
+  optional_class_name = &class_name;
+
+  if (
+    type == RUBY_T_OBJECT   ||
+    type == RUBY_T_CLASS    ||
+    type == RUBY_T_MODULE   ||
+    type == RUBY_T_FLOAT    ||
+    type == RUBY_T_STRING   ||
+    type == RUBY_T_REGEXP   ||
+    type == RUBY_T_ARRAY    ||
+    type == RUBY_T_HASH     ||
+    type == RUBY_T_STRUCT   ||
+    type == RUBY_T_BIGNUM   ||
+    type == RUBY_T_FILE     ||
+    type == RUBY_T_DATA     ||
+    type == RUBY_T_MATCH    ||
+    type == RUBY_T_COMPLEX  ||
+    type == RUBY_T_RATIONAL ||
+    type == RUBY_T_NIL      ||
+    type == RUBY_T_TRUE     ||
+    type == RUBY_T_FALSE    ||
+    type == RUBY_T_SYMBOL   ||
+    type == RUBY_T_FIXNUM
+  ) {
+    VALUE klass = rb_class_of(new_object);
+
+    // Ruby sometimes plays a bit fast and loose with some of its internal objects, e.g.
+    // `rb_str_tmp_frozen_acquire` allocates a string with no class (klass=0).
+    // Thus, we need to make sure there's actually a class before getting its name.
+
+    if (klass != 0) {
+      const char *name = rb_class2name(klass);
+      size_t name_length = name != NULL ? strlen(name) : 0;
+
+      if (name_length > 0) {
+        class_name = (ddog_CharSlice) {.ptr = name, .len = name_length};
       } else {
-        // Fallback for objects with no class
+        // @ivoanjo: I'm not sure this can ever happen, but just-in-case
         class_name = ruby_value_type_to_class_name(type);
       }
-    } else if (type == RUBY_T_IMEMO) {
-      const char *imemo_string = imemo_kind(new_object);
-      if (imemo_string != NULL) {
-        snprintf(imemo_type, 100, "(VM Internal, T_IMEMO, %s)", imemo_string);
-        class_name = (ddog_CharSlice) {.ptr = imemo_type, .len = strlen(imemo_type)};
-      } else { // Ruby < 3
-        class_name = DDOG_CHARSLICE_C("(VM Internal, T_IMEMO)");
-      }
     } else {
-      class_name = ruby_vm_type; // For other weird internal things we just use the VM type
+      // Fallback for objects with no class. Objects with no class are a way for the Ruby VM to mark them
+      // as internal objects; see rb_objspace_internal_object_p for details.
+      class_name = ruby_value_type_to_class_name(type);
     }
+  } else if (type == RUBY_T_IMEMO) {
+    const char *imemo_string = imemo_kind(new_object);
+    if (imemo_string != NULL) {
+      snprintf(imemo_type, 100, "(VM Internal, T_IMEMO, %s)", imemo_string);
+      class_name = (ddog_CharSlice) {.ptr = imemo_type, .len = strlen(imemo_type)};
+    } else { // Ruby < 3
+      class_name = DDOG_CHARSLICE_C("(VM Internal, T_IMEMO)");
+    }
+  } else {
+    class_name = ruby_vm_type; // For other weird internal things we just use the VM type
   }
 
   track_object(state->recorder_instance, new_object, sample_weight, optional_class_name);
@@ -1763,14 +1750,14 @@ static uint64_t otel_span_id_to_uint(VALUE otel_span_id) {
 
   // This function can get called from outside the GVL and even on non-main Ractors
   __attribute__((warn_unused_result))
-  bool thread_context_collector_on_gvl_running_with_threshold(gvl_profiling_thread thread, uint32_t waiting_for_gvl_threshold_ns) {
+  on_gvl_running_result thread_context_collector_on_gvl_running_with_threshold(gvl_profiling_thread thread, uint32_t waiting_for_gvl_threshold_ns) {
     intptr_t gvl_waiting_at = gvl_profiling_state_get(thread);
 
     // Thread was not being profiled / not waiting on gvl
-    if (gvl_waiting_at == 0 || gvl_waiting_at == GVL_WAITING_ENABLED_EMPTY) return false;
+    if (gvl_waiting_at == 0 || gvl_waiting_at == GVL_WAITING_ENABLED_EMPTY) return ON_GVL_RUNNING_UNKNOWN;
 
     // @ivoanjo: I'm not sure if this can happen -- It means we should've sampled already but haven't gotten the chance yet?
-    if (gvl_waiting_at < 0) return true;
+    if (gvl_waiting_at < 0) return ON_GVL_RUNNING_SAMPLE;
 
     long waiting_for_gvl_duration_ns = monotonic_wall_time_now_ns(DO_NOT_RAISE_ON_FAILURE) - gvl_waiting_at;
 
@@ -1786,11 +1773,11 @@ static uint64_t otel_span_id_to_uint(VALUE otel_span_id) {
       gvl_profiling_state_set(thread, GVL_WAITING_ENABLED_EMPTY);
     }
 
-    return should_sample;
+    return should_sample ? ON_GVL_RUNNING_SAMPLE : ON_GVL_RUNNING_DONT_SAMPLE;
   }
 
   __attribute__((warn_unused_result))
-  bool thread_context_collector_on_gvl_running(gvl_profiling_thread thread) {
+  on_gvl_running_result thread_context_collector_on_gvl_running(gvl_profiling_thread thread) {
     return thread_context_collector_on_gvl_running_with_threshold(thread, global_waiting_for_gvl_threshold_ns);
   }
 
@@ -1822,7 +1809,7 @@ static uint64_t otel_span_id_to_uint(VALUE otel_span_id) {
   //
   // NOTE: In normal use, current_thread is expected to be == rb_thread_current(); the `current_thread` parameter only
   // exists to enable testing.
-  VALUE thread_context_collector_sample_after_gvl_running_with_thread(VALUE self_instance, VALUE current_thread) {
+  VALUE thread_context_collector_sample_after_gvl_running(VALUE self_instance, VALUE current_thread, long current_monotonic_wall_time_ns) {
     struct thread_context_collector_state *state;
     TypedData_Get_Struct(self_instance, struct thread_context_collector_state, &thread_context_collector_typed_data, state);
 
@@ -1854,14 +1841,10 @@ static uint64_t otel_span_id_to_uint(VALUE otel_span_id) {
       thread_context,
       thread_context->sampling_buffer,
       cpu_time_for_thread,
-      monotonic_wall_time_now_ns(RAISE_ON_FAILURE)
+      current_monotonic_wall_time_ns
     );
 
-    return Qtrue; // To allow this to be called from rb_rescue2
-  }
-
-  VALUE thread_context_collector_sample_after_gvl_running(VALUE self_instance) {
-    return thread_context_collector_sample_after_gvl_running_with_thread(self_instance, rb_thread_current());
+    return Qtrue;
   }
 
   // This method is intended to be called from update_metrics_and_sample. It exists to handle extra sampling steps we
@@ -1980,13 +1963,17 @@ static uint64_t otel_span_id_to_uint(VALUE otel_span_id) {
   static VALUE _native_on_gvl_running(DDTRACE_UNUSED VALUE self, VALUE thread) {
     ENFORCE_THREAD(thread);
 
-    return thread_context_collector_on_gvl_running(thread_from_thread_object(thread)) ? Qtrue : Qfalse;
+    return thread_context_collector_on_gvl_running(thread_from_thread_object(thread)) == ON_GVL_RUNNING_SAMPLE ? Qtrue : Qfalse;
   }
 
   static VALUE _native_sample_after_gvl_running(DDTRACE_UNUSED VALUE self, VALUE collector_instance, VALUE thread) {
     ENFORCE_THREAD(thread);
 
-    return thread_context_collector_sample_after_gvl_running_with_thread(collector_instance, thread);
+    return thread_context_collector_sample_after_gvl_running(
+      collector_instance,
+      thread,
+      monotonic_wall_time_now_ns(RAISE_ON_FAILURE)
+    );
   }
 
   static VALUE _native_apply_delta_to_cpu_time_at_previous_sample_ns(DDTRACE_UNUSED VALUE self, VALUE collector_instance, VALUE thread, VALUE delta_ns) {

data/ext/datadog_profiling_native_extension/collectors_thread_context.h

@@ -18,7 +18,13 @@ __attribute__((warn_unused_result)) bool thread_context_collector_on_gc_finish(V
 VALUE enforce_thread_context_collector_instance(VALUE object);
 
 #ifndef NO_GVL_INSTRUMENTATION
+  typedef enum {
+    ON_GVL_RUNNING_UNKNOWN, // Thread is not known, it may not even be from the current Ractor
+    ON_GVL_RUNNING_DONT_SAMPLE, // Thread is known, but "Waiting for GVL" period was too small to be sampled
+    ON_GVL_RUNNING_SAMPLE, // Thread is known, and "Waiting for GVL" period should be sampled
+  } on_gvl_running_result;
+
   void thread_context_collector_on_gvl_waiting(gvl_profiling_thread thread);
-  __attribute__((warn_unused_result)) bool thread_context_collector_on_gvl_running(gvl_profiling_thread thread);
-  VALUE thread_context_collector_sample_after_gvl_running(VALUE self_instance);
+  __attribute__((warn_unused_result)) on_gvl_running_result thread_context_collector_on_gvl_running(gvl_profiling_thread thread);
+  VALUE thread_context_collector_sample_after_gvl_running(VALUE self_instance, VALUE current_thread, long current_monotonic_wall_time_ns);
 #endif