ddtrace 1.3.0 → 1.4.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: d512b6ba74cbbe3c0b879a8efd29a513680dfd1ce9864c785887717ababf7880
-  data.tar.gz: 637d702bc35996f24d28904f20d77e4c6dece600c98eec6bbc47ba72ba7a96e2
+  metadata.gz: '01987691359b14541248048bb2f720031b4f0f0ca0be8fade0f6d0daf7aef8cc'
+  data.tar.gz: b2cce425177f3d619d0064d88ffb52073515b8d390e054117a160eac21e939d5
 SHA512:
-  metadata.gz: bee0a447195f79d8a551f1c5076f283601c4ea14ec049f6dc077213ae578d0bdd5a0456f5b6d741a719d7543cc161b3da79970a5d696d0e2cd1e068166d9954d
-  data.tar.gz: da1801a3e6cdf5b559bca925126e68add53293b0a581666302020fdeb6dea3a27e7a4d70ead568f24a9158184f957cd25e6c3ae3b3c267b4d467ff96114e3c8a
+  metadata.gz: 7232b588ee13d8a47a1fccbe39a8753a19a4e9c3b0b2dee2cb278a64b183500ec9859cfecffda8c128971bf88c93b148e6f915c2d7906139d1df050b173df42c
+  data.tar.gz: 2605640bffc586a0fd8bb38139d49fa6dad9f04bfe91a0e672cdd04a6c8a8a36896be5322a4b78ecab37175cd600bdc144109646712265b49bad4e103cc1ab5e

data/CHANGELOG.md

@@ -2,6 +2,32 @@
 
 ## [Unreleased]
 
+## [1.4.1] - 2022-09-15
+
+### Fixed
+
+* Missing distributed traces when trace is dropped by priority sampling ([#2101][], [#2279][])
+* Profiling support when Ruby is compiled without a shared library ([#2250][])
+
+## [1.4.0] - 2022-08-25
+
+Release notes: https://github.com/DataDog/dd-trace-rb/releases/tag/v1.4.0
+
+Git diff: https://github.com/DataDog/dd-trace-rb/compare/v1.3.0...v1.4.0
+
+### Added
+
+* gRPC: tag `grpc.client.deadline` ([#2200][])
+* Implement telemetry, disable by default ([#2153][])
+
+### Changed
+
+* Bump `libdatadog` dependency version ([#2229][])
+
+### Fixed
+
+* Fix CI instrumentation configuration ([#2219][])
+
 ## [1.3.0] - 2022-08-04
 
 Release notes: https://github.com/DataDog/dd-trace-rb/releases/tag/v1.3.0
@@ -2071,7 +2097,11 @@ 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/v1.1.0...master
+[Unreleased]: https://github.com/DataDog/dd-trace-rb/compare/v1.4.1...master
+[1.4.1]: https://github.com/DataDog/dd-trace-rb/compare/v1.4.0...v1.4.1
+[1.4.0]: https://github.com/DataDog/dd-trace-rb/compare/v1.3.0...v1.4.0
+[1.3.0]: https://github.com/DataDog/dd-trace-rb/compare/v1.2.0...v1.3.0
+[1.2.0]: https://github.com/DataDog/dd-trace-rb/compare/v1.1.0...v1.2.0
 [1.1.0]: https://github.com/DataDog/dd-trace-rb/compare/v1.0.0...v1.1.0
 [1.0.0]: https://github.com/DataDog/dd-trace-rb/compare/v1.0.0.beta2...v1.0.0
 [1.0.0.beta2]: https://github.com/DataDog/dd-trace-rb/compare/v1.0.0.beta1...v1.0.0.beta2
@@ -2931,6 +2961,7 @@ Git diff: https://github.com/DataDog/dd-trace-rb/compare/v0.3.0...v0.3.1
 [#2082]: https://github.com/DataDog/dd-trace-rb/issues/2082
 [#2096]: https://github.com/DataDog/dd-trace-rb/issues/2096
 [#2097]: https://github.com/DataDog/dd-trace-rb/issues/2097
+[#2101]: https://github.com/DataDog/dd-trace-rb/issues/2101
 [#2110]: https://github.com/DataDog/dd-trace-rb/issues/2110
 [#2113]: https://github.com/DataDog/dd-trace-rb/issues/2113
 [#2118]: https://github.com/DataDog/dd-trace-rb/issues/2118
@@ -2939,13 +2970,20 @@ Git diff: https://github.com/DataDog/dd-trace-rb/compare/v0.3.0...v0.3.1
 [#2138]: https://github.com/DataDog/dd-trace-rb/issues/2138
 [#2140]: https://github.com/DataDog/dd-trace-rb/issues/2140
 [#2150]: https://github.com/DataDog/dd-trace-rb/issues/2150
+[#2153]: https://github.com/DataDog/dd-trace-rb/issues/2153
 [#2158]: https://github.com/DataDog/dd-trace-rb/issues/2158
 [#2162]: https://github.com/DataDog/dd-trace-rb/issues/2162
 [#2163]: https://github.com/DataDog/dd-trace-rb/issues/2163
+[#2170]: https://github.com/DataDog/dd-trace-rb/issues/2170
 [#2173]: https://github.com/DataDog/dd-trace-rb/issues/2173
 [#2174]: https://github.com/DataDog/dd-trace-rb/issues/2174
 [#2180]: https://github.com/DataDog/dd-trace-rb/issues/2180
+[#2200]: https://github.com/DataDog/dd-trace-rb/issues/2200
 [#2201]: https://github.com/DataDog/dd-trace-rb/issues/2201
+[#2219]: https://github.com/DataDog/dd-trace-rb/issues/2219
+[#2229]: https://github.com/DataDog/dd-trace-rb/issues/2229
+[#2250]: https://github.com/DataDog/dd-trace-rb/issues/2250
+[#2279]: https://github.com/DataDog/dd-trace-rb/issues/2279
 [@AdrianLC]: https://github.com/AdrianLC
 [@Azure7111]: https://github.com/Azure7111
 [@BabyGroot]: https://github.com/BabyGroot
@@ -3087,4 +3125,4 @@ Git diff: https://github.com/DataDog/dd-trace-rb/compare/v0.3.0...v0.3.1
 [@walterking]: https://github.com/walterking
 [@y-yagi]: https://github.com/y-yagi
 [@yukimurasawa]: https://github.com/yukimurasawa
-[@zachmccormick]: https://github.com/zachmccormick
+[@zachmccormick]: https://github.com/zachmccormick

data/README.md

@@ -1,4 +1,4 @@
-***Version 1.0.0 has recently been released. Check out our [upgrade guide](https://github.com/DataDog/dd-trace-rb/blob/master/docs/UpgradeGuide.md#from-0x-to-10) for more details.***
+**We've recently released the 1.x version series. If you're upgrading from a 0.x version, check out our [upgrade guide](https://github.com/DataDog/dd-trace-rb/blob/master/docs/UpgradeGuide.md#from-0x-to-10).**
 
 # Datadog Trace Client
 

data/ext/ddtrace_profiling_loader/ddtrace_profiling_loader.c

@@ -85,7 +85,16 @@ static bool failed_to_load(void *handle, VALUE *failure_details) {
 static bool incompatible_library(void *handle, VALUE *failure_details) {
   // The library being loaded may be linked to a different libruby than the current executing Ruby.
   // We check if this is the case by checking if a well-known symbol resolves to a common address.
-  if (dlsym(handle, "ruby_xmalloc") != &ruby_xmalloc) {
+
+  void *xmalloc_from_library = dlsym(handle, "ruby_xmalloc");
+
+  if (xmalloc_from_library == NULL) {
+    // This happens when ruby is built without a `libruby.so` by using `--disable-shared` at compilation time.
+    // In this situation, no conflict between libruby version is possible.
+    return false;
+  }
+
+  if (xmalloc_from_library != &ruby_xmalloc) {
     *failure_details = rb_str_new_cstr("library was compiled and linked to a different Ruby version");
     unload_failed_library(handle);
     return true;

data/ext/ddtrace_profiling_native_extension/collectors_cpu_and_wall_time.c

@@ -153,8 +153,6 @@ static VALUE _native_new(VALUE klass) {
 }
 
 static VALUE _native_initialize(DDTRACE_UNUSED VALUE _self, VALUE collector_instance, VALUE recorder_instance, VALUE max_frames) {
-  enforce_recorder_instance(recorder_instance);
-
   struct cpu_and_wall_time_collector_state *state;
   TypedData_Get_Struct(collector_instance, struct cpu_and_wall_time_collector_state, &cpu_and_wall_time_collector_typed_data, state);
 
@@ -164,7 +162,7 @@ static VALUE _native_initialize(DDTRACE_UNUSED VALUE _self, VALUE collector_inst
   // Update this when modifying state struct
   state->sampling_buffer = sampling_buffer_new(max_frames_requested);
   // hash_map_per_thread_context is already initialized, nothing to do here
-  state->recorder_instance = recorder_instance;
+  state->recorder_instance = enforce_recorder_instance(recorder_instance);
 
   return Qtrue;
 }
@@ -180,6 +178,8 @@ static VALUE _native_sample(DDTRACE_UNUSED VALUE _self, VALUE collector_instance
 //
 // Assumption 1: This function is called in a thread that is holding the Global VM Lock. Caller is responsible for enforcing this.
 // Assumption 2: This function is allowed to raise exceptions. Caller is responsible for handling them, if needed.
+// Assumption 3: This function IS NOT called from a signal handler. This function is not async-signal-safe.
+// Assumption 4: This function IS NOT called in a reentrant way.
 VALUE cpu_and_wall_time_collector_sample(VALUE self_instance) {
   struct cpu_and_wall_time_collector_state *state;
   TypedData_Get_Struct(self_instance, struct cpu_and_wall_time_collector_state, &cpu_and_wall_time_collector_typed_data, state);
@@ -384,6 +384,7 @@ static long thread_id_for(VALUE thread) {
   return FIXNUM_P(object_id) ? FIX2LONG(object_id) : -1;
 }
 
-void enforce_cpu_and_wall_time_collector_instance(VALUE object) {
+VALUE enforce_cpu_and_wall_time_collector_instance(VALUE object) {
   Check_TypedStruct(object, &cpu_and_wall_time_collector_typed_data);
+  return object;
 }

data/ext/ddtrace_profiling_native_extension/collectors_cpu_and_wall_time.h

@@ -3,4 +3,4 @@
 #include <ruby.h>
 
 VALUE cpu_and_wall_time_collector_sample(VALUE self_instance);
-void enforce_cpu_and_wall_time_collector_instance(VALUE object);
+VALUE enforce_cpu_and_wall_time_collector_instance(VALUE object);

data/ext/ddtrace_profiling_native_extension/collectors_cpu_and_wall_time_worker.c

@@ -0,0 +1,391 @@
+#include <ruby.h>
+#include <ruby/thread.h>
+#include <ruby/thread_native.h>
+#include <ruby/debug.h>
+#include <stdbool.h>
+#include <signal.h>
+#include "helpers.h"
+#include "ruby_helpers.h"
+#include "collectors_cpu_and_wall_time.h"
+#include "private_vm_api_access.h"
+
+// Used to trigger the periodic execution of Collectors::CpuAndWallTime, which implements all of the sampling logic
+// itself; this class only implements the "doing it periodically" part.
+//
+// This file implements the native bits of the Datadog::Profiling::Collectors::CpuAndWallTimeWorker class
+
+// ---
+// Here be dragons: This component is quite fiddly and probably one of the more complex in the profiler as it deals with
+// multiple threads, signal handlers, global state, etc.
+//
+// ## Design notes for this class:
+//
+// ### Constraints
+//
+// Currently, sampling Ruby threads requires calling Ruby VM APIs that are only safe to call while holding on to the
+// global VM lock (and are not async-signal safe -- cannot be called from a signal handler).
+//
+// @ivoanjo: As a note, I don't think we should think of this constraint as set in stone. Since can reach into the Ruby
+// internals, we may be able to figure out a way of overcoming it. But it's definitely going to be hard so for now
+// we're considering it as a given.
+//
+// ### Flow for triggering samples
+//
+// The flow for triggering samples is as follows:
+//
+// 1. Inside the `run_sampling_trigger_loop` function (running in the `CpuAndWallTimeWorker` background thread),
+// a `SIGPROF` signal gets sent to the current process.
+//
+// 2. The `handle_sampling_signal` signal handler function gets called to handle the `SIGPROF` signal.
+//
+//   Which thread the signal handler function gets called on by the operating system is quite important. We need to perform
+// an operation -- calling the `rb_postponed_job_register_one` API -- that can only be called from the thread that
+// is holding on to the global VM lock. So this is the thread we're "hoping" our signal lands on.
+//
+//   The signal never lands on the `CpuAndWallTimeWorker` background thread because we explicitly block it off from that
+// thread in `block_sigprof_signal_handler_from_running_in_current_thread`.
+//
+//   If the signal lands on a thread that is not holding onto the global VM lock, we can't proceed to the next step,
+// and we need to restart the sampling flow from step 1. (There's still quite a few improvements we can make here,
+// but this is the current state of the implementation).
+//
+// 3. Inside `handle_sampling_signal`, if it's getting executed by the Ruby thread that is holding the global VM lock,
+// we can call `rb_postponed_job_register_one` to ask the Ruby VM to call our `sample_from_postponed_job` function
+// "as soon as it can".
+//
+// 4. The Ruby VM calls our `sample_from_postponed_job` from a thread holding the global VM lock. A sample is recorded by
+// calling `cpu_and_wall_time_collector_sample`.
+//
+// ---
+
+// Contains state for a single CpuAndWallTimeWorker instance
+struct cpu_and_wall_time_worker_state {
+  // Important: This is not atomic nor is it guaranteed to replace memory barriers and the like. Aka this works for
+  // telling the sampling trigger loop to stop, but if we ever need to communicate more, we should move to actual
+  // atomic operations. stdatomic.h seems a nice thing to reach out for.
+  volatile bool should_run;
+
+  VALUE cpu_and_wall_time_collector_instance;
+  // When something goes wrong during sampling, we record the Ruby exception here, so that it can be "re-raised" on
+  // the CpuAndWallTimeWorker thread
+  VALUE failure_exception;
+};
+
+static VALUE _native_new(VALUE klass);
+static VALUE _native_initialize(DDTRACE_UNUSED VALUE _self, VALUE self_instance, VALUE cpu_and_wall_time_collector_instance);
+static void cpu_and_wall_time_worker_typed_data_mark(void *state_ptr);
+static VALUE _native_sampling_loop(VALUE self, VALUE instance);
+static VALUE _native_stop(DDTRACE_UNUSED VALUE _self, VALUE self_instance);
+static void install_sigprof_signal_handler(void (*signal_handler_function)(int, siginfo_t *, void *));
+static void remove_sigprof_signal_handler(void);
+static void block_sigprof_signal_handler_from_running_in_current_thread(void);
+static void handle_sampling_signal(DDTRACE_UNUSED int _signal, DDTRACE_UNUSED siginfo_t *_info, DDTRACE_UNUSED void *_ucontext);
+static void *run_sampling_trigger_loop(void *state_ptr);
+static void interrupt_sampling_trigger_loop(void *state_ptr);
+static void sample_from_postponed_job(DDTRACE_UNUSED void *_unused);
+static VALUE handle_sampling_failure(VALUE self_instance, VALUE exception);
+static VALUE _native_current_sigprof_signal_handler(DDTRACE_UNUSED VALUE self);
+static VALUE release_gvl_and_run_sampling_trigger_loop(VALUE instance);
+static VALUE _native_is_running(DDTRACE_UNUSED VALUE self, VALUE instance);
+static void testing_signal_handler(DDTRACE_UNUSED int _signal, DDTRACE_UNUSED siginfo_t *_info, DDTRACE_UNUSED void *_ucontext);
+static VALUE _native_install_testing_signal_handler(DDTRACE_UNUSED VALUE self);
+static VALUE _native_remove_testing_signal_handler(DDTRACE_UNUSED VALUE self);
+
+// Global state -- be very careful when accessing or modifying it
+
+// Note: Global state must only be mutated while holding the global VM lock (we piggy back on it to ensure correctness).
+// The active_sampler_instance needs to be global because we access it from the signal handler.
+static VALUE active_sampler_instance = Qnil;
+// ...We also store active_sampler_owner_thread to be able to tell who the active_sampler_instance belongs to (and also
+// to detect when it is outdated)
+static VALUE active_sampler_owner_thread = Qnil;
+
+void collectors_cpu_and_wall_time_worker_init(VALUE profiling_module) {
+  rb_global_variable(&active_sampler_instance);
+  rb_global_variable(&active_sampler_owner_thread);
+
+  VALUE collectors_module = rb_define_module_under(profiling_module, "Collectors");
+  VALUE collectors_cpu_and_wall_time_worker_class = rb_define_class_under(collectors_module, "CpuAndWallTimeWorker", rb_cObject);
+  // Hosts methods used for testing the native code using RSpec
+  VALUE testing_module = rb_define_module_under(collectors_cpu_and_wall_time_worker_class, "Testing");
+
+  // Instances of the CpuAndWallTimeWorker class are "TypedData" objects.
+  // "TypedData" objects are special objects in the Ruby VM that can wrap C structs.
+  // In this case, it wraps the cpu_and_wall_time_worker_state.
+  //
+  // Because Ruby doesn't know how to initialize native-level structs, we MUST override the allocation function for objects
+  // of this class so that we can manage this part. Not overriding or disabling the allocation function is a common
+  // gotcha for "TypedData" objects that can very easily lead to VM crashes, see for instance
+  // https://bugs.ruby-lang.org/issues/18007 for a discussion around this.
+  rb_define_alloc_func(collectors_cpu_and_wall_time_worker_class, _native_new);
+
+  rb_define_singleton_method(collectors_cpu_and_wall_time_worker_class, "_native_initialize", _native_initialize, 2);
+  rb_define_singleton_method(collectors_cpu_and_wall_time_worker_class, "_native_sampling_loop", _native_sampling_loop, 1);
+  rb_define_singleton_method(collectors_cpu_and_wall_time_worker_class, "_native_stop", _native_stop, 1);
+  rb_define_singleton_method(testing_module, "_native_current_sigprof_signal_handler", _native_current_sigprof_signal_handler, 0);
+  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);
+}
+
+// This structure is used to define a Ruby object that stores a pointer to a struct cpu_and_wall_time_worker_state
+// See also https://github.com/ruby/ruby/blob/master/doc/extension.rdoc for how this works
+static const rb_data_type_t cpu_and_wall_time_worker_typed_data = {
+  .wrap_struct_name = "Datadog::Profiling::Collectors::CpuAndWallTimeWorker",
+  .function = {
+    .dmark = cpu_and_wall_time_worker_typed_data_mark,
+    .dfree = RUBY_DEFAULT_FREE,
+    .dsize = NULL, // We don't track profile memory usage (although it'd be cool if we did!)
+    //.dcompact = NULL, // FIXME: Add support for compaction
+  },
+  .flags = RUBY_TYPED_FREE_IMMEDIATELY
+};
+
+static VALUE _native_new(VALUE klass) {
+  struct cpu_and_wall_time_worker_state *state = ruby_xcalloc(1, sizeof(struct cpu_and_wall_time_worker_state));
+
+  state->should_run = false;
+  state->cpu_and_wall_time_collector_instance = Qnil;
+  state->failure_exception = Qnil;
+
+  return TypedData_Wrap_Struct(klass, &cpu_and_wall_time_worker_typed_data, state);
+}
+
+static VALUE _native_initialize(DDTRACE_UNUSED VALUE _self, VALUE self_instance, VALUE cpu_and_wall_time_collector_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);
+
+  state->cpu_and_wall_time_collector_instance = enforce_cpu_and_wall_time_collector_instance(cpu_and_wall_time_collector_instance);
+
+  return Qtrue;
+}
+
+// Since our state contains references to Ruby objects, we need to tell the Ruby GC about them
+static void cpu_and_wall_time_worker_typed_data_mark(void *state_ptr) {
+  struct cpu_and_wall_time_worker_state *state = (struct cpu_and_wall_time_worker_state *) state_ptr;
+
+  rb_gc_mark(state->cpu_and_wall_time_collector_instance);
+  rb_gc_mark(state->failure_exception);
+}
+
+// Called in a background thread created in CpuAndWallTimeWorker#start
+static VALUE _native_sampling_loop(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);
+
+  if (active_sampler_owner_thread != Qnil && is_thread_alive(active_sampler_owner_thread)) {
+    rb_raise(
+      rb_eRuntimeError,
+      "Could not start CpuAndWallTimeWorker: There's already another instance of CpuAndWallTimeWorker active in a different thread"
+    );
+  }
+
+  // This write to a global is thread-safe BECAUSE we're still holding on to the global VM lock at this point
+  active_sampler_instance = instance;
+  active_sampler_owner_thread = rb_thread_current();
+
+  state->should_run = true;
+
+  block_sigprof_signal_handler_from_running_in_current_thread(); // We want to interrupt the thread with the global VM lock, never this one
+
+  install_sigprof_signal_handler(handle_sampling_signal);
+
+  // Release GVL, get to the actual work!
+  int exception_state;
+  rb_protect(release_gvl_and_run_sampling_trigger_loop, instance, &exception_state);
+
+  // The sample trigger loop finished (either cleanly or with an error); let's clean up
+
+  remove_sigprof_signal_handler();
+  active_sampler_instance = Qnil;
+  active_sampler_owner_thread = Qnil;
+
+  // Ensure that instance is not garbage collected while the native sampling loop is running; this is probably not needed, but just in case
+  RB_GC_GUARD(instance);
+
+  if (exception_state) rb_jump_tag(exception_state); // Re-raise any exception that happened
+
+  return Qnil;
+}
+
+static VALUE _native_stop(DDTRACE_UNUSED VALUE _self, 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);
+
+  state->should_run = false;
+
+  return Qtrue;
+}
+
+static void install_sigprof_signal_handler(void (*signal_handler_function)(int, siginfo_t *, void *)) {
+  struct sigaction existing_signal_handler_config = {.sa_sigaction = NULL};
+  struct sigaction signal_handler_config = {
+    .sa_flags = SA_RESTART | SA_SIGINFO,
+    .sa_sigaction = signal_handler_function
+  };
+  sigemptyset(&signal_handler_config.sa_mask);
+
+  if (sigaction(SIGPROF, &signal_handler_config, &existing_signal_handler_config) != 0) {
+    rb_sys_fail("Could not start CpuAndWallTimeWorker: Could not install signal handler");
+  }
+
+  // In some corner cases (e.g. after a fork), our signal handler may still be around, and that's ok
+  if (existing_signal_handler_config.sa_sigaction == handle_sampling_signal) return;
+
+  if (existing_signal_handler_config.sa_handler != NULL || existing_signal_handler_config.sa_sigaction != NULL) {
+    // A previous signal handler already existed. Currently we don't support this situation, so let's just back out
+    // of the installation.
+
+    if (sigaction(SIGPROF, &existing_signal_handler_config, NULL) != 0) {
+      rb_sys_fail(
+        "Could not start CpuAndWallTimeWorker: Could not re-install pre-existing SIGPROF signal handler. " \
+        "This may break the component had installed it."
+      );
+    }
+
+    rb_raise(rb_eRuntimeError, "Could not start CpuAndWallTimeWorker: There's a pre-existing SIGPROF signal handler");
+  }
+}
+
+static void remove_sigprof_signal_handler(void) {
+  struct sigaction signal_handler_config = {
+    .sa_handler = SIG_DFL, // Reset back to default
+    .sa_flags = SA_RESTART // TODO: Unclear if this is actually needed/does anything at all
+  };
+  sigemptyset(&signal_handler_config.sa_mask);
+
+  if (sigaction(SIGPROF, &signal_handler_config, NULL) != 0) rb_sys_fail("Failure while removing the signal handler");
+}
+
+static void block_sigprof_signal_handler_from_running_in_current_thread(void) {
+  sigset_t signals_to_block;
+  sigemptyset(&signals_to_block);
+  sigaddset(&signals_to_block, SIGPROF);
+  pthread_sigmask(SIG_BLOCK, &signals_to_block, NULL);
+}
+
+static void handle_sampling_signal(DDTRACE_UNUSED int _signal, DDTRACE_UNUSED siginfo_t *_info, DDTRACE_UNUSED void *_ucontext) {
+  if (!ruby_thread_has_gvl_p()) {
+    return; // Not safe to enqueue a sample from this thread
+  }
+
+  // We implicitly assume there can be no concurrent nor nested calls to handle_sampling_signal because
+  // a) we get triggered using SIGPROF, and the docs state second SIGPROF will not interrupt an existing one
+  // b) we validate we are in the thread that has the global VM lock; if a different thread gets a signal, it will return early
+  //    because it will not have the global VM lock
+  // TODO: Validate that this does not impact Ractors
+
+  // Note: rb_postponed_job_register_one ensures that if there's a previous sample_from_postponed_job queued for execution
+  // then we will not queue a second one. It does this by doing a linear scan on the existing jobs; in the future we
+  // may want to implement that check ourselves.
+
+  // TODO: Do something with result (potentially update tracking counters?)
+  /*int result =*/ rb_postponed_job_register_one(0, sample_from_postponed_job, NULL);
+}
+
+// The actual sampling trigger loop always runs **without** the global vm lock.
+static void *run_sampling_trigger_loop(void *state_ptr) {
+  struct cpu_and_wall_time_worker_state *state = (struct cpu_and_wall_time_worker_state *) state_ptr;
+
+  struct timespec time_between_signals = {.tv_nsec = 10 * 1000 * 1000 /* 10ms */};
+
+  while (state->should_run) {
+    // TODO: This is still a placeholder for a more complex mechanism. In particular:
+    // * We want to signal a particular thread or threads, not the process in general
+    // * We want to track if a signal landed on the thread holding the global VM lock and do something about it
+    // * We want to do more than having a fixed sampling rate
+
+    kill(getpid(), SIGPROF);
+    nanosleep(&time_between_signals, NULL);
+  }
+
+  return NULL; // Unused
+}
+
+// This is called by the Ruby VM when it wants to shut down the background thread
+static void interrupt_sampling_trigger_loop(void *state_ptr) {
+  struct cpu_and_wall_time_worker_state *state = (struct cpu_and_wall_time_worker_state *) state_ptr;
+
+  state->should_run = false;
+}
+
+static void sample_from_postponed_job(DDTRACE_UNUSED void *_unused) {
+  VALUE instance = active_sampler_instance; // Read from global variable
+
+  // This can potentially happen if the CpuAndWallTimeWorker was stopped while the postponed job was waiting to be executed; nothing to do
+  if (instance == Qnil) return;
+
+  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);
+
+  // Trigger sampling using the Collectors::CpuAndWallTime; rescue against any exceptions that happen during sampling
+  VALUE (*function_to_call_safely)(VALUE) = cpu_and_wall_time_collector_sample;
+  VALUE function_to_call_safely_arg = state->cpu_and_wall_time_collector_instance;
+  VALUE (*exception_handler_function)(VALUE, VALUE) = handle_sampling_failure;
+  VALUE exception_handler_function_arg = instance;
+  rb_rescue2(
+    function_to_call_safely,
+    function_to_call_safely_arg,
+    exception_handler_function,
+    exception_handler_function_arg,
+    rb_eException, // rb_eException is the base class of all Ruby exceptions
+    0 // Required by API to be the last argument
+  );
+}
+
+static VALUE handle_sampling_failure(VALUE self_instance, VALUE exception) {
+  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);
+
+  state->should_run = false;
+  state->failure_exception = exception;
+
+  return Qnil;
+}
+
+static VALUE _native_current_sigprof_signal_handler(DDTRACE_UNUSED VALUE self) {
+  struct sigaction existing_signal_handler_config = {.sa_sigaction = NULL};
+  if (sigaction(SIGPROF, NULL, &existing_signal_handler_config) != 0) {
+    rb_sys_fail("Failed to probe existing handler");
+  }
+
+  if (existing_signal_handler_config.sa_sigaction == handle_sampling_signal) {
+    return ID2SYM(rb_intern("profiling"));
+  } else if (existing_signal_handler_config.sa_sigaction != NULL) {
+    return ID2SYM(rb_intern("other"));
+  } else {
+    return Qnil;
+  }
+}
+
+static VALUE release_gvl_and_run_sampling_trigger_loop(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);
+
+  rb_thread_call_without_gvl(run_sampling_trigger_loop, state, interrupt_sampling_trigger_loop, state);
+
+  // If we stopped sampling due to an exception, re-raise it (now in the worker thread)
+  if (state->failure_exception != Qnil) rb_exc_raise(state->failure_exception);
+
+  return Qnil;
+}
+
+static VALUE _native_is_running(DDTRACE_UNUSED VALUE self, VALUE instance) {
+  return \
+    (active_sampler_owner_thread != Qnil && is_thread_alive(active_sampler_owner_thread) && active_sampler_instance == instance) ?
+    Qtrue : Qfalse;
+}
+
+static void testing_signal_handler(DDTRACE_UNUSED int _signal, DDTRACE_UNUSED siginfo_t *_info, DDTRACE_UNUSED void *_ucontext) {
+  /* Does nothing on purpose */
+}
+
+static VALUE _native_install_testing_signal_handler(DDTRACE_UNUSED VALUE self) {
+  install_sigprof_signal_handler(testing_signal_handler);
+  return Qtrue;
+}
+
+static VALUE _native_remove_testing_signal_handler(DDTRACE_UNUSED VALUE self) {
+  remove_sigprof_signal_handler();
+  return Qtrue;
+}

data/ext/ddtrace_profiling_native_extension/extconf.rb

@@ -144,6 +144,8 @@ if RUBY_VERSION < '2.3'
   $defs << '-DUSE_LEGACY_RB_PROFILE_FRAMES'
   # ... you couldn't name threads
   $defs << '-DNO_THREAD_NAMES'
+  # ...the ruby_thread_has_gvl_p function was not exposed to users outside of the VM
+  $defs << '-DNO_THREAD_HAS_GVL'
 end
 
 # If we got here, libdatadog is available and loaded

data/ext/ddtrace_profiling_native_extension/private_vm_api_access.c

@@ -681,3 +681,12 @@ int ddtrace_rb_profile_frames(VALUE thread, int start, int limit, VALUE *buff, i
 }
 
 #endif // USE_LEGACY_RB_PROFILE_FRAMES
+
+#ifdef NO_THREAD_HAS_GVL
+int ruby_thread_has_gvl_p(void) {
+  // TODO: The CpuAndWallTimeWorker needs this function, but Ruby 2.2 doesn't expose it... For now this placeholder
+  // will enable the profiling native extension to continue to compile on Ruby 2.2, but the CpuAndWallTimeWorker will
+  // not work properly on 2.2. Will be addressed later.
+  return 0;
+}
+#endif // NO_THREAD_HAS_GVL

data/ext/ddtrace_profiling_native_extension/profiling.c

@@ -5,6 +5,7 @@
 
 // Each class/module here is implemented in their separate file
 void collectors_cpu_and_wall_time_init(VALUE profiling_module);
+void collectors_cpu_and_wall_time_worker_init(VALUE profiling_module);
 void collectors_stack_init(VALUE profiling_module);
 void http_transport_init(VALUE profiling_module);
 void stack_recorder_init(VALUE profiling_module);
@@ -22,6 +23,7 @@ void DDTRACE_EXPORT Init_ddtrace_profiling_native_extension(void) {
   rb_define_singleton_method(native_extension_module, "clock_id_for", clock_id_for, 1); // from clock_id.h
 
   collectors_cpu_and_wall_time_init(profiling_module);
+  collectors_cpu_and_wall_time_worker_init(profiling_module);
   collectors_stack_init(profiling_module);
   http_transport_init(profiling_module);
   stack_recorder_init(profiling_module);

data/ext/ddtrace_profiling_native_extension/stack_recorder.c

@@ -324,8 +324,9 @@ static void *call_serialize_without_gvl(void *call_args) {
   return NULL; // Unused
 }
 
-void enforce_recorder_instance(VALUE object) {
+VALUE enforce_recorder_instance(VALUE object) {
   Check_TypedStruct(object, &stack_recorder_typed_data);
+  return object;
 }
 
 static struct active_slot_pair sampler_lock_active_profile(struct stack_recorder_state *state) {

data/ext/ddtrace_profiling_native_extension/stack_recorder.h

@@ -15,12 +15,13 @@
 // ```
 #define VALUE_STRING(string) {.ptr = "" string, .len = sizeof(string) - 1}
 
-#define      CPU_TIME_VALUE {.type_ = VALUE_STRING("cpu-time"),      .unit = VALUE_STRING("nanoseconds")}
-#define   CPU_SAMPLES_VALUE {.type_ = VALUE_STRING("cpu-samples"),   .unit = VALUE_STRING("count")}
-#define     WALL_TIME_VALUE {.type_ = VALUE_STRING("wall-time"),     .unit = VALUE_STRING("nanoseconds")}
-#define ALLOC_SAMPLES_VALUE {.type_ = VALUE_STRING("alloc-samples"), .unit = VALUE_STRING("count")}
-#define   ALLOC_SPACE_VALUE {.type_ = VALUE_STRING("alloc-space"),   .unit = VALUE_STRING("bytes")}
-#define    HEAP_SPACE_VALUE {.type_ = VALUE_STRING("heap-space"),    .unit = VALUE_STRING("bytes")}
+#define CPU_TIME_VALUE          {.type_ = VALUE_STRING("cpu-time"),          .unit = VALUE_STRING("nanoseconds")}
+#define CPU_SAMPLES_VALUE       {.type_ = VALUE_STRING("cpu-samples"),       .unit = VALUE_STRING("count")}
+#define WALL_TIME_VALUE         {.type_ = VALUE_STRING("wall-time"),         .unit = VALUE_STRING("nanoseconds")}
+#define ALLOC_SIZE_VALUE        {.type_ = VALUE_STRING("alloc-size"),        .unit = VALUE_STRING("bytes")}
+#define ALLOC_SAMPLES_VALUE     {.type_ = VALUE_STRING("alloc-samples"),     .unit = VALUE_STRING("count")}
+#define HEAP_LIVE_SIZE_VALUE    {.type_ = VALUE_STRING("heap-live-size"),    .unit = VALUE_STRING("bytes")}
+#define HEAP_LIVE_SAMPLES_VALUE {.type_ = VALUE_STRING("heap-live-samples"), .unit = VALUE_STRING("count")}
 
 static const ddprof_ffi_ValueType enabled_value_types[] = {
   #define CPU_TIME_VALUE_POS 0
@@ -34,4 +35,4 @@ static const ddprof_ffi_ValueType enabled_value_types[] = {
 #define ENABLED_VALUE_TYPES_COUNT (sizeof(enabled_value_types) / sizeof(ddprof_ffi_ValueType))
 
 void record_sample(VALUE recorder_instance, ddprof_ffi_Sample sample);
-void enforce_recorder_instance(VALUE object);
+VALUE enforce_recorder_instance(VALUE object);

data/lib/datadog/ci/contrib/cucumber/integration.rb

@@ -35,7 +35,7 @@ module Datadog
             false
           end
 
-          def default_configuration
+          def new_configuration
             Configuration::Settings.new
           end
 

data/lib/datadog/ci/contrib/rspec/integration.rb

@@ -36,7 +36,7 @@ module Datadog
             false
           end
 
-          def default_configuration
+          def new_configuration
             Configuration::Settings.new
           end