ddtrace 1.1.0 → 1.12.1

data/ext/ddtrace_profiling_native_extension/collectors_thread_context.h

@@ -0,0 +1,14 @@
+#pragma once
+
+#include <ruby.h>
+
+void thread_context_collector_sample(
+  VALUE self_instance,
+  long current_monotonic_wall_time_ns,
+  VALUE profiler_overhead_stack_thread
+);
+void thread_context_collector_sample_allocation(VALUE self_instance, unsigned int sample_weight);
+VALUE thread_context_collector_sample_after_gc(VALUE self_instance);
+void thread_context_collector_on_gc_start(VALUE self_instance);
+void thread_context_collector_on_gc_finish(VALUE self_instance);
+VALUE enforce_thread_context_collector_instance(VALUE object);

data/ext/ddtrace_profiling_native_extension/extconf.rb

@@ -1,5 +1,3 @@
-# typed: ignore
-
 # rubocop:disable Style/StderrPuts
 # rubocop:disable Style/GlobalVars
 
@@ -10,13 +8,31 @@ SKIPPED_REASON_FILE = "#{__dir__}/skipped_reason.txt".freeze
 File.delete(SKIPPED_REASON_FILE) rescue nil
 
 def skip_building_extension!(reason)
-  $stderr.puts(Datadog::Profiling::NativeExtensionHelpers::Supported.failure_banner_for(**reason))
+  fail_install_if_missing_extension =
+    Datadog::Profiling::NativeExtensionHelpers.fail_install_if_missing_extension?
+
+  $stderr.puts(
+    Datadog::Profiling::NativeExtensionHelpers::Supported.failure_banner_for(
+      **reason,
+      fail_install: fail_install_if_missing_extension,
+    )
+  )
+
   File.write(
     SKIPPED_REASON_FILE,
     Datadog::Profiling::NativeExtensionHelpers::Supported.render_skipped_reason_file(**reason),
   )
 
-  File.write('Makefile', 'all install clean: # dummy makefile that does nothing')
+  if fail_install_if_missing_extension
+    require 'mkmf'
+    Logging.message(
+      '[ddtrace] Failure cause: ' \
+      "#{Datadog::Profiling::NativeExtensionHelpers::Supported.render_skipped_reason_file(**reason)}\n"
+    )
+  else
+    File.write('Makefile', 'all install clean: # dummy makefile that does nothing')
+  end
+
   exit
 end
 
@@ -24,7 +40,8 @@ unless Datadog::Profiling::NativeExtensionHelpers::Supported.supported?
   skip_building_extension!(Datadog::Profiling::NativeExtensionHelpers::Supported.unsupported_reason)
 end
 
-$stderr.puts(%(
+$stderr.puts(
+  %(
 +------------------------------------------------------------------------------+
 | ** Preparing to build the ddtrace profiling native extension... **           |
 |                                                                              |
@@ -41,12 +58,17 @@ $stderr.puts(%(
 | Thanks for using ddtrace! You rock!                                          |
 +------------------------------------------------------------------------------+
 
-))
+)
+)
 
 # NOTE: we MUST NOT require 'mkmf' before we check the #skip_building_extension? because the require triggers checks
 # that may fail on an environment not properly setup for building Ruby extensions.
 require 'mkmf'
 
+Logging.message("[ddtrace] Using compiler:\n")
+xsystem("#{CONFIG['CC']} -v")
+Logging.message("[ddtrace] End of compiler information\n")
+
 # mkmf on modern Rubies actually has an append_cflags that does something similar
 # (see https://github.com/ruby/ruby/pull/5760), but as usual we need a bit more boilerplate to deal with legacy Rubies
 def add_compiler_flag(flag)
@@ -57,13 +79,20 @@ def add_compiler_flag(flag)
   end
 end
 
+# Because we can't control what compiler versions our customers use, shipping with -Werror by default is a no-go.
+# But we can enable it in CI, so that we quickly spot any new warnings that just got introduced.
+#
+# @ivoanjo TODO: Ruby 3.3.0-preview1 was causing issues in CI because `have_header('vm_core.h')` below triggers warnings;
+# I've chosen to disable `-Werror` for this Ruby version for now, and we can revisit this on a later 3.3 release.
+add_compiler_flag '-Werror' if ENV['DDTRACE_CI'] == 'true' && !RUBY_DESCRIPTION.include?('3.3.0preview1')
+
 # Older gcc releases may not default to C99 and we need to ask for this. This is also used:
 # * by upstream Ruby -- search for gnu99 in the codebase
 # * by msgpack, another ddtrace dependency
 #   (https://github.com/msgpack/msgpack-ruby/blob/18ce08f6d612fe973843c366ac9a0b74c4e50599/ext/msgpack/extconf.rb#L8)
 add_compiler_flag '-std=gnu99'
 
-# Gets really noisy when we include the MJIT header, let's omit it
+# Gets really noisy when we include the MJIT header, let's omit it (TODO: Use #pragma GCC diagnostic instead?)
 add_compiler_flag '-Wno-unused-function'
 
 # Allow defining variables at any point in a function
@@ -73,6 +102,9 @@ add_compiler_flag '-Wno-declaration-after-statement'
 # cause a segfault later. Let's ensure that never happens.
 add_compiler_flag '-Werror-implicit-function-declaration'
 
+# Warn on unused parameters to functions. Use `DDTRACE_UNUSED` to mark things as known-to-not-be-used.
+add_compiler_flag '-Wunused-parameter'
+
 # The native extension is not intended to expose any symbols/functions for other native libraries to use;
 # the sole exception being `Init_ddtrace_profiling_native_extension` which needs to be visible for Ruby to call it when
 # it `dlopen`s the library.
@@ -81,23 +113,54 @@ add_compiler_flag '-Werror-implicit-function-declaration'
 # For more details see https://gcc.gnu.org/wiki/Visibility
 add_compiler_flag '-fvisibility=hidden'
 
+# Avoid legacy C definitions
+add_compiler_flag '-Wold-style-definition'
+
+# Enable all other compiler warnings
+add_compiler_flag '-Wall'
+add_compiler_flag '-Wextra'
+
 if RUBY_PLATFORM.include?('linux')
   # Supposedly, the correct way to do this is
   # ```
   # have_library 'pthread'
   # have_func 'pthread_getcpuclockid'
   # ```
-  # but it broke the build on Windows and on older Ruby versions (2.1 and 2.2)
+  # but a) it broke the build on Windows, b) on older Ruby versions (2.2 and below) and c) It's slower to build
   # so instead we just assume that we have the function we need on Linux, and nowhere else
   $defs << '-DHAVE_PTHREAD_GETCPUCLOCKID'
 end
 
+# On older Rubies, we did not need to include the ractor header (this was built into the MJIT header)
+$defs << '-DNO_RACTOR_HEADER_INCLUDE' if RUBY_VERSION < '3.3'
+
+# On older Rubies, some of the Ractor internal APIs were directly accessible
+$defs << '-DUSE_RACTOR_INTERNAL_APIS_DIRECTLY' if RUBY_VERSION < '3.3'
+
+# On older Rubies, there was no struct rb_native_thread. See private_vm_api_acccess.c for details.
+$defs << '-DNO_RB_NATIVE_THREAD' if RUBY_VERSION < '3.2'
+
+# On older Rubies, there was no struct rb_thread_sched (it was struct rb_global_vm_lock_struct)
+$defs << '-DNO_RB_THREAD_SCHED' if RUBY_VERSION < '3.2'
+
+# On older Rubies, there was no tid member in the internal thread structure
+$defs << '-DNO_THREAD_TID' if RUBY_VERSION < '3.1'
+
 # On older Rubies, we need to use a backported version of this function. See private_vm_api_access.h for details.
 $defs << '-DUSE_BACKPORTED_RB_PROFILE_FRAME_METHOD_NAME' if RUBY_VERSION < '3'
 
+# On older Rubies, there are no Ractors
+$defs << '-DNO_RACTORS' if RUBY_VERSION < '3'
+
+# On older Rubies, rb_global_vm_lock_struct did not include the owner field
+$defs << '-DNO_GVL_OWNER' if RUBY_VERSION < '2.6'
+
 # On older Rubies, we need to use rb_thread_t instead of rb_execution_context_t
 $defs << '-DUSE_THREAD_INSTEAD_OF_EXECUTION_CONTEXT' if RUBY_VERSION < '2.5'
 
+# On older Rubies, extensions can't use GET_VM()
+$defs << '-DNO_GET_VM' if RUBY_VERSION < '2.5'
+
 # On older Rubies...
 if RUBY_VERSION < '2.4'
   # ...we need to use RUBY_VM_NORMAL_ISEQ_P instead of VM_FRAME_RUBYFRAME_P
@@ -106,20 +169,36 @@ if RUBY_VERSION < '2.4'
   $defs << '-DUSE_LEGACY_RB_VM_FRAME_METHOD_ENTRY'
 end
 
-# For REALLY OLD Rubies...
-if RUBY_VERSION < '2.3'
-  # ...there was no rb_time_timespec_new function
-  $defs << '-DNO_RB_TIME_TIMESPEC_NEW'
-  # ...the VM changed enough that we need an alternative legacy rb_profile_frames
-  $defs << '-DUSE_LEGACY_RB_PROFILE_FRAMES'
+# If we got here, libdatadog is available and loaded
+ENV['PKG_CONFIG_PATH'] = "#{ENV['PKG_CONFIG_PATH']}:#{Libdatadog.pkgconfig_folder}"
+Logging.message("[ddtrace] PKG_CONFIG_PATH set to #{ENV['PKG_CONFIG_PATH'].inspect}\n")
+$stderr.puts("Using libdatadog #{Libdatadog::VERSION} from #{Libdatadog.pkgconfig_folder}")
+
+unless pkg_config('datadog_profiling_with_rpath')
+  Logging.message("[ddtrace] Ruby detected the pkg-config command is #{$PKGCONFIG.inspect}\n")
+
+  skip_building_extension!(
+    if Datadog::Profiling::NativeExtensionHelpers::Supported.pkg_config_missing?
+      Datadog::Profiling::NativeExtensionHelpers::Supported::PKG_CONFIG_IS_MISSING
+    else
+      # Less specific error message
+      Datadog::Profiling::NativeExtensionHelpers::Supported::FAILED_TO_CONFIGURE_LIBDATADOG
+    end
+  )
 end
 
-# If we got here, libddprof is available and loaded
-ENV['PKG_CONFIG_PATH'] = "#{ENV['PKG_CONFIG_PATH']}:#{Libddprof.pkgconfig_folder}"
-unless pkg_config('ddprof_ffi_with_rpath')
-  skip_building_extension!(Datadog::Profiling::NativeExtensionHelpers::Supported::FAILED_TO_CONFIGURE_LIBDDPROF)
+unless have_type('atomic_int', ['stdatomic.h'])
+  skip_building_extension!(Datadog::Profiling::NativeExtensionHelpers::Supported::COMPILER_ATOMIC_MISSING)
 end
 
+# See comments on the helper method being used for why we need to additionally set this.
+# The extremely excessive escaping around ORIGIN below seems to be correct and was determined after a lot of
+# experimentation. We need to get these special characters across a lot of tools untouched...
+$LDFLAGS += \
+  ' -Wl,-rpath,$$$\\\\{ORIGIN\\}/' \
+  "#{Datadog::Profiling::NativeExtensionHelpers.libdatadog_folder_relative_to_native_lib_folder}"
+Logging.message("[ddtrace] After pkg-config $LDFLAGS were set to: #{$LDFLAGS.inspect}\n")
+
 # Tag the native extension library with the Ruby version and Ruby platform.
 # This makes it easier for development (avoids "oops I forgot to rebuild when I switched my Ruby") and ensures that
 # the wrong library is never loaded.
@@ -149,17 +228,11 @@ if Datadog::Profiling::NativeExtensionHelpers::CAN_USE_MJIT_HEADER
 
   create_makefile EXTENSION_NAME
 else
-  # On older Rubies, we use the debase-ruby_core_source gem to get access to private VM headers.
+  # The MJIT header was introduced on 2.6 and removed on 3.3; for other Rubies we rely on
+  # the debase-ruby_core_source gem to get access to private VM headers.
   # This gem ships source code copies of these VM headers for the different Ruby VM versions;
   # see https://github.com/ruby-debug/debase-ruby_core_source for details
 
-  thread_native_for_ruby_2_1 = proc { true }
-  if RUBY_VERSION < '2.2'
-    # This header became public in Ruby 2.2, but we need to pull it from the private headers folder for 2.1
-    thread_native_for_ruby_2_1 = proc { have_header('thread_native.h') }
-    $defs << '-DRUBY_2_1_WORKAROUND'
-  end
-
   create_header
 
   require 'debase/ruby_core_source'
@@ -167,7 +240,11 @@ else
 
   Debase::RubyCoreSource
     .create_makefile_with_core(
-      proc { have_header('vm_core.h') && have_header('iseq.h') && thread_native_for_ruby_2_1.call },
+      proc do
+        have_header('vm_core.h') &&
+        have_header('iseq.h') &&
+        (RUBY_VERSION < '3.3' || have_header('ractor_core.h'))
+      end,
       EXTENSION_NAME,
     )
 end

data/ext/ddtrace_profiling_native_extension/helpers.h

@@ -0,0 +1,17 @@
+#pragma once
+
+// Used to mark symbols to be exported to the outside of the extension.
+// Consider very carefully before tagging a function with this.
+#define DDTRACE_EXPORT __attribute__ ((visibility ("default")))
+
+// Used to mark function arguments that are deliberately left unused
+#ifdef __GNUC__
+  #define DDTRACE_UNUSED  __attribute__((unused))
+#else
+  #define DDTRACE_UNUSED
+#endif
+
+// @ivoanjo: After trying to read through https://stackoverflow.com/questions/3437404/min-and-max-in-c I decided I
+// don't like C and I just implemented this as a function.
+inline static uint64_t uint64_max_of(uint64_t a, uint64_t b) { return a > b ? a : b; }
+inline static uint64_t uint64_min_of(uint64_t a, uint64_t b) { return a > b ? b : a; }

data/ext/ddtrace_profiling_native_extension/http_transport.c

@@ -0,0 +1,347 @@
+#include <ruby.h>
+#include <ruby/thread.h>
+#include <datadog/profiling.h>
+#include "helpers.h"
+#include "libdatadog_helpers.h"
+#include "ruby_helpers.h"
+
+// Used to report profiling data to Datadog.
+// This file implements the native bits of the Datadog::Profiling::HttpTransport class
+
+static VALUE ok_symbol = Qnil; // :ok in Ruby
+static VALUE error_symbol = Qnil; // :error in Ruby
+
+static ID agentless_id; // id of :agentless in Ruby
+static ID agent_id; // id of :agent in Ruby
+
+static ID log_failure_to_process_tag_id; // id of :log_failure_to_process_tag in Ruby
+
+static VALUE http_transport_class = Qnil;
+static VALUE library_version_string = Qnil;
+
+struct call_exporter_without_gvl_arguments {
+  ddog_prof_Exporter *exporter;
+  ddog_prof_Exporter_Request_BuildResult *build_result;
+  ddog_CancellationToken *cancel_token;
+  ddog_prof_Exporter_SendResult result;
+  bool send_ran;
+};
+
+inline static ddog_ByteSlice byte_slice_from_ruby_string(VALUE string);
+static VALUE _native_validate_exporter(VALUE self, VALUE exporter_configuration);
+static ddog_prof_Exporter_NewResult create_exporter(VALUE exporter_configuration, VALUE tags_as_array);
+static VALUE handle_exporter_failure(ddog_prof_Exporter_NewResult exporter_result);
+static ddog_Endpoint endpoint_from(VALUE exporter_configuration);
+static ddog_Vec_Tag convert_tags(VALUE tags_as_array);
+static void safely_log_failure_to_process_tag(ddog_Vec_Tag tags, VALUE err_details);
+static VALUE _native_do_export(
+  VALUE self,
+  VALUE exporter_configuration,
+  VALUE upload_timeout_milliseconds,
+  VALUE start_timespec_seconds,
+  VALUE start_timespec_nanoseconds,
+  VALUE finish_timespec_seconds,
+  VALUE finish_timespec_nanoseconds,
+  VALUE pprof_file_name,
+  VALUE pprof_data,
+  VALUE code_provenance_file_name,
+  VALUE code_provenance_data,
+  VALUE tags_as_array
+);
+static void *call_exporter_without_gvl(void *call_args);
+static void interrupt_exporter_call(void *cancel_token);
+static VALUE ddtrace_version(void);
+
+void http_transport_init(VALUE profiling_module) {
+  http_transport_class = rb_define_class_under(profiling_module, "HttpTransport", rb_cObject);
+
+  rb_define_singleton_method(http_transport_class, "_native_validate_exporter",  _native_validate_exporter, 1);
+  rb_define_singleton_method(http_transport_class, "_native_do_export",  _native_do_export, 11);
+
+  ok_symbol = ID2SYM(rb_intern_const("ok"));
+  error_symbol = ID2SYM(rb_intern_const("error"));
+  agentless_id = rb_intern_const("agentless");
+  agent_id = rb_intern_const("agent");
+  log_failure_to_process_tag_id = rb_intern_const("log_failure_to_process_tag");
+
+  library_version_string = ddtrace_version();
+  rb_global_variable(&library_version_string);
+}
+
+inline static ddog_ByteSlice byte_slice_from_ruby_string(VALUE string) {
+  ENFORCE_TYPE(string, T_STRING);
+  ddog_ByteSlice byte_slice = {.ptr = (uint8_t *) StringValuePtr(string), .len = RSTRING_LEN(string)};
+  return byte_slice;
+}
+
+static VALUE _native_validate_exporter(DDTRACE_UNUSED VALUE _self, VALUE exporter_configuration) {
+  ENFORCE_TYPE(exporter_configuration, T_ARRAY);
+  ddog_prof_Exporter_NewResult exporter_result = create_exporter(exporter_configuration, rb_ary_new());
+
+  VALUE failure_tuple = handle_exporter_failure(exporter_result);
+  if (!NIL_P(failure_tuple)) return failure_tuple;
+
+  // We don't actually need the exporter for now -- we just wanted to validate that we could create it with the
+  // settings we were given
+  ddog_prof_Exporter_drop(exporter_result.ok);
+
+  return rb_ary_new_from_args(2, ok_symbol, Qnil);
+}
+
+static ddog_prof_Exporter_NewResult create_exporter(VALUE exporter_configuration, VALUE tags_as_array) {
+  ENFORCE_TYPE(exporter_configuration, T_ARRAY);
+  ENFORCE_TYPE(tags_as_array, T_ARRAY);
+
+  // This needs to be called BEFORE convert_tags since it can raise an exception and thus cause the ddog_Vec_Tag
+  // to be leaked.
+  ddog_Endpoint endpoint = endpoint_from(exporter_configuration);
+
+  ddog_Vec_Tag tags = convert_tags(tags_as_array);
+
+  ddog_CharSlice library_name = DDOG_CHARSLICE_C("dd-trace-rb");
+  ddog_CharSlice library_version = char_slice_from_ruby_string(library_version_string);
+  ddog_CharSlice profiling_family = DDOG_CHARSLICE_C("ruby");
+
+  ddog_prof_Exporter_NewResult exporter_result =
+    ddog_prof_Exporter_new(library_name, library_version, profiling_family, &tags, endpoint);
+
+  ddog_Vec_Tag_drop(tags);
+
+  return exporter_result;
+}
+
+static VALUE handle_exporter_failure(ddog_prof_Exporter_NewResult exporter_result) {
+  return exporter_result.tag == DDOG_PROF_EXPORTER_NEW_RESULT_OK ?
+    Qnil :
+    rb_ary_new_from_args(2, error_symbol, get_error_details_and_drop(&exporter_result.err));
+}
+
+static ddog_Endpoint endpoint_from(VALUE exporter_configuration) {
+  ENFORCE_TYPE(exporter_configuration, T_ARRAY);
+
+  ID working_mode = SYM2ID(rb_ary_entry(exporter_configuration, 0)); // SYM2ID verifies its input so we can do this safely
+
+  if (working_mode != agentless_id && working_mode != agent_id) {
+    rb_raise(rb_eArgError, "Failed to initialize transport: Unexpected working mode, expected :agentless or :agent");
+  }
+
+  if (working_mode == agentless_id) {
+    VALUE site = rb_ary_entry(exporter_configuration, 1);
+    VALUE api_key = rb_ary_entry(exporter_configuration, 2);
+    ENFORCE_TYPE(site, T_STRING);
+    ENFORCE_TYPE(api_key, T_STRING);
+
+    return ddog_Endpoint_agentless(char_slice_from_ruby_string(site), char_slice_from_ruby_string(api_key));
+  } else { // agent_id
+    VALUE base_url = rb_ary_entry(exporter_configuration, 1);
+    ENFORCE_TYPE(base_url, T_STRING);
+
+    return ddog_Endpoint_agent(char_slice_from_ruby_string(base_url));
+  }
+}
+
+__attribute__((warn_unused_result))
+static ddog_Vec_Tag convert_tags(VALUE tags_as_array) {
+  ENFORCE_TYPE(tags_as_array, T_ARRAY);
+
+  long tags_count = RARRAY_LEN(tags_as_array);
+  ddog_Vec_Tag tags = ddog_Vec_Tag_new();
+
+  for (long i = 0; i < tags_count; i++) {
+    VALUE name_value_pair = rb_ary_entry(tags_as_array, i);
+
+    if (!RB_TYPE_P(name_value_pair, T_ARRAY)) {
+      ddog_Vec_Tag_drop(tags);
+      ENFORCE_TYPE(name_value_pair, T_ARRAY);
+    }
+
+    // Note: We can index the array without checking its size first because rb_ary_entry returns Qnil if out of bounds
+    VALUE tag_name = rb_ary_entry(name_value_pair, 0);
+    VALUE tag_value = rb_ary_entry(name_value_pair, 1);
+
+    if (!(RB_TYPE_P(tag_name, T_STRING) && RB_TYPE_P(tag_value, T_STRING))) {
+      ddog_Vec_Tag_drop(tags);
+      ENFORCE_TYPE(tag_name, T_STRING);
+      ENFORCE_TYPE(tag_value, T_STRING);
+    }
+
+    ddog_Vec_Tag_PushResult push_result =
+      ddog_Vec_Tag_push(&tags, char_slice_from_ruby_string(tag_name), char_slice_from_ruby_string(tag_value));
+
+    if (push_result.tag == DDOG_VEC_TAG_PUSH_RESULT_ERR) {
+      // libdatadog validates tags and may catch invalid tags that ddtrace didn't actually catch.
+      // We warn users about such tags, and then just ignore them.
+      safely_log_failure_to_process_tag(tags, get_error_details_and_drop(&push_result.err));
+    }
+  }
+
+  return tags;
+}
+
+static VALUE log_failure_to_process_tag(VALUE err_details) {
+  return rb_funcall(http_transport_class, log_failure_to_process_tag_id, 1, err_details);
+}
+
+// Since we are calling into Ruby code, it may raise an exception. This method ensure that dynamically-allocated tags
+// get cleaned before propagating the exception.
+static void safely_log_failure_to_process_tag(ddog_Vec_Tag tags, VALUE err_details) {
+  int exception_state;
+  rb_protect(log_failure_to_process_tag, err_details, &exception_state);
+
+  if (exception_state) {           // An exception was raised
+    ddog_Vec_Tag_drop(tags); // clean up
+    rb_jump_tag(exception_state);  // "Re-raise" exception
+  }
+}
+
+// Note: This function handles a bunch of libdatadog dynamically-allocated objects, so it MUST not use any Ruby APIs
+// which can raise exceptions, otherwise the objects will be leaked.
+static VALUE perform_export(
+  ddog_prof_Exporter *exporter,
+  ddog_Timespec start,
+  ddog_Timespec finish,
+  ddog_prof_Exporter_Slice_File slice_files,
+  ddog_Vec_Tag *additional_tags,
+  uint64_t timeout_milliseconds
+) {
+  ddog_prof_ProfiledEndpointsStats *endpoints_stats = NULL; // Not in use yet
+  ddog_prof_Exporter_Request_BuildResult build_result =
+    ddog_prof_Exporter_Request_build(exporter, start, finish, slice_files, additional_tags, endpoints_stats, timeout_milliseconds);
+
+  if (build_result.tag == DDOG_PROF_EXPORTER_REQUEST_BUILD_RESULT_ERR) {
+    ddog_prof_Exporter_drop(exporter);
+    return rb_ary_new_from_args(2, error_symbol, get_error_details_and_drop(&build_result.err));
+  }
+
+  ddog_CancellationToken *cancel_token = ddog_CancellationToken_new();
+
+  // We'll release the Global VM Lock while we're calling send, so that the Ruby VM can continue to work while this
+  // is pending
+  struct call_exporter_without_gvl_arguments args =
+    {.exporter = exporter, .build_result = &build_result, .cancel_token = cancel_token, .send_ran = false};
+
+  // We use rb_thread_call_without_gvl2 instead of rb_thread_call_without_gvl as the gvl2 variant never raises any
+  // exceptions.
+  //
+  // (With rb_thread_call_without_gvl, if someone calls Thread#kill or something like it on the current thread,
+  // the exception will be raised without us being able to clean up dynamically-allocated stuff, which would leak.)
+  //
+  // Instead, we take care of our own exception checking, and delay the exception raising (`rb_jump_tag` call) until
+  // after we cleaned up any dynamically-allocated resources.
+  //
+  // We run rb_thread_call_without_gvl2 in a loop since an "interrupt" may cause it to return before even running
+  // our code. In such a case, we retry the call -- unless the interrupt was caused by an exception being pending,
+  // and in that case we also give up and break out of the loop.
+  int pending_exception = 0;
+
+  while (!args.send_ran && !pending_exception) {
+    rb_thread_call_without_gvl2(call_exporter_without_gvl, &args, interrupt_exporter_call, cancel_token);
+
+    // To make sure we don't leak memory, we never check for pending exceptions if send ran
+    if (!args.send_ran) pending_exception = check_if_pending_exception();
+  }
+
+  // Cleanup exporter and token, no longer needed
+  ddog_CancellationToken_drop(cancel_token);
+  ddog_prof_Exporter_drop(exporter);
+
+  if (pending_exception) {
+    // If we got here send did not run, so we need to explicitly dispose of the request
+    ddog_prof_Exporter_Request_drop(&build_result.ok);
+
+    // Let Ruby propagate the exception. This will not return.
+    rb_jump_tag(pending_exception);
+  }
+
+  // The request itself does not need to be freed as libdatadog takes ownership of it as part of sending.
+
+  ddog_prof_Exporter_SendResult result = args.result;
+
+  return result.tag == DDOG_PROF_EXPORTER_SEND_RESULT_HTTP_RESPONSE ?
+    rb_ary_new_from_args(2, ok_symbol, UINT2NUM(result.http_response.code)) :
+    rb_ary_new_from_args(2, error_symbol, get_error_details_and_drop(&result.err));
+}
+
+static VALUE _native_do_export(
+  DDTRACE_UNUSED VALUE _self,
+  VALUE exporter_configuration,
+  VALUE upload_timeout_milliseconds,
+  VALUE start_timespec_seconds,
+  VALUE start_timespec_nanoseconds,
+  VALUE finish_timespec_seconds,
+  VALUE finish_timespec_nanoseconds,
+  VALUE pprof_file_name,
+  VALUE pprof_data,
+  VALUE code_provenance_file_name,
+  VALUE code_provenance_data,
+  VALUE tags_as_array
+) {
+  ENFORCE_TYPE(upload_timeout_milliseconds, T_FIXNUM);
+  ENFORCE_TYPE(start_timespec_seconds, T_FIXNUM);
+  ENFORCE_TYPE(start_timespec_nanoseconds, T_FIXNUM);
+  ENFORCE_TYPE(finish_timespec_seconds, T_FIXNUM);
+  ENFORCE_TYPE(finish_timespec_nanoseconds, T_FIXNUM);
+  ENFORCE_TYPE(pprof_file_name, T_STRING);
+  ENFORCE_TYPE(pprof_data, T_STRING);
+  ENFORCE_TYPE(code_provenance_file_name, T_STRING);
+
+  // Code provenance can be disabled and in that case will be set to nil
+  bool have_code_provenance = !NIL_P(code_provenance_data);
+  if (have_code_provenance) ENFORCE_TYPE(code_provenance_data, T_STRING);
+
+  uint64_t timeout_milliseconds = NUM2ULONG(upload_timeout_milliseconds);
+
+  ddog_Timespec start =
+    {.seconds = NUM2LONG(start_timespec_seconds), .nanoseconds = NUM2UINT(start_timespec_nanoseconds)};
+  ddog_Timespec finish =
+    {.seconds = NUM2LONG(finish_timespec_seconds), .nanoseconds = NUM2UINT(finish_timespec_nanoseconds)};
+
+  int files_to_report = 1 + (have_code_provenance ? 1 : 0);
+  ddog_prof_Exporter_File files[files_to_report];
+  ddog_prof_Exporter_Slice_File slice_files = {.ptr = files, .len = files_to_report};
+
+  files[0] = (ddog_prof_Exporter_File) {
+    .name = char_slice_from_ruby_string(pprof_file_name),
+    .file = byte_slice_from_ruby_string(pprof_data)
+  };
+  if (have_code_provenance) {
+    files[1] = (ddog_prof_Exporter_File) {
+      .name = char_slice_from_ruby_string(code_provenance_file_name),
+      .file = byte_slice_from_ruby_string(code_provenance_data)
+    };
+  }
+
+  ddog_Vec_Tag *null_additional_tags = NULL;
+
+  ddog_prof_Exporter_NewResult exporter_result = create_exporter(exporter_configuration, tags_as_array);
+  // Note: Do not add anything that can raise exceptions after this line, as otherwise the exporter memory will leak
+
+  VALUE failure_tuple = handle_exporter_failure(exporter_result);
+  if (!NIL_P(failure_tuple)) return failure_tuple;
+
+  return perform_export(exporter_result.ok, start, finish, slice_files, null_additional_tags, timeout_milliseconds);
+}
+
+static void *call_exporter_without_gvl(void *call_args) {
+  struct call_exporter_without_gvl_arguments *args = (struct call_exporter_without_gvl_arguments*) call_args;
+
+  args->result = ddog_prof_Exporter_send(args->exporter, &args->build_result->ok, args->cancel_token);
+  args->send_ran = true;
+
+  return NULL; // Unused
+}
+
+// Called by Ruby when it wants to interrupt call_exporter_without_gvl above, e.g. when the app wants to exit cleanly
+static void interrupt_exporter_call(void *cancel_token) {
+  ddog_CancellationToken_cancel((ddog_CancellationToken *) cancel_token);
+}
+
+static VALUE ddtrace_version(void) {
+  VALUE ddtrace_module = rb_const_get(rb_cObject, rb_intern("DDTrace"));
+  ENFORCE_TYPE(ddtrace_module, T_MODULE);
+  VALUE version_module = rb_const_get(ddtrace_module, rb_intern("VERSION"));
+  ENFORCE_TYPE(version_module, T_MODULE);
+  VALUE version_string = rb_const_get(version_module, rb_intern("STRING"));
+  ENFORCE_TYPE(version_string, T_STRING);
+  return version_string;
+}

data/ext/ddtrace_profiling_native_extension/libdatadog_helpers.h

@@ -0,0 +1,25 @@
+#pragma once
+
+#include <datadog/profiling.h>
+#include "ruby_helpers.h"
+
+inline static ddog_CharSlice char_slice_from_ruby_string(VALUE string) {
+  ENFORCE_TYPE(string, T_STRING);
+  ddog_CharSlice char_slice = {.ptr = StringValuePtr(string), .len = RSTRING_LEN(string)};
+  return char_slice;
+}
+
+inline static VALUE ruby_string_from_vec_u8(ddog_Vec_U8 string) {
+  return rb_str_new((char *) string.ptr, string.len);
+}
+
+inline static VALUE ruby_string_from_error(const ddog_Error *error) {
+  ddog_CharSlice char_slice = ddog_Error_message(error);
+  return rb_str_new(char_slice.ptr, char_slice.len);
+}
+
+inline static VALUE get_error_details_and_drop(ddog_Error *error) {
+  VALUE result = ruby_string_from_error(error);
+  ddog_Error_drop(error);
+  return result;
+}