datadog 2.3.0 → 2.5.0

data/ext/datadog_profiling_native_extension/collectors_thread_context.h

@@ -3,6 +3,8 @@
 #include <ruby.h>
 #include <stdbool.h>
 
+#include "gvl_profiling_helper.h"
+
 void thread_context_collector_sample(
   VALUE self_instance,
   long current_monotonic_wall_time_ns,
@@ -12,5 +14,17 @@ void thread_context_collector_sample_allocation(VALUE self_instance, unsigned in
 void thread_context_collector_sample_skipped_allocation_samples(VALUE self_instance, unsigned int skipped_samples);
 VALUE thread_context_collector_sample_after_gc(VALUE self_instance);
 void thread_context_collector_on_gc_start(VALUE self_instance);
-bool thread_context_collector_on_gc_finish(VALUE self_instance);
+__attribute__((warn_unused_result)) bool thread_context_collector_on_gc_finish(VALUE self_instance);
 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)) 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

data/ext/datadog_profiling_native_extension/datadog_ruby_common.c

@@ -28,33 +28,6 @@ VALUE datadog_gem_version(void) {
   return version_string;
 }
 
-__attribute__((warn_unused_result))
-ddog_prof_Endpoint endpoint_from(VALUE exporter_configuration) {
-  ENFORCE_TYPE(exporter_configuration, T_ARRAY);
-
-  VALUE exporter_working_mode = rb_ary_entry(exporter_configuration, 0);
-  ENFORCE_TYPE(exporter_working_mode, T_SYMBOL);
-  ID working_mode = SYM2ID(exporter_working_mode);
-
-  ID agentless_id = rb_intern("agentless");
-  ID agent_id = rb_intern("agent");
-
-  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);
-
-    return ddog_prof_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);
-
-    return ddog_prof_Endpoint_agent(char_slice_from_ruby_string(base_url));
-  }
-}
-
 static VALUE log_failure_to_process_tag(VALUE err_details) {
   VALUE datadog_module = rb_const_get(rb_cObject, rb_intern("Datadog"));
   VALUE logger = rb_funcall(datadog_module, rb_intern("logger"), 0);

data/ext/datadog_profiling_native_extension/datadog_ruby_common.h

@@ -27,7 +27,6 @@
 #define ENFORCE_BOOLEAN(value) \
   { if (RB_UNLIKELY(value != Qtrue && value != Qfalse)) raise_unexpected_type(value, ADD_QUOTES(value), "true or false", __FILE__, __LINE__, __func__); }
 
-// Called by ENFORCE_TYPE; should not be used directly
 NORETURN(void raise_unexpected_type(VALUE value, const char *value_name, const char *type_name, const char *file, int line, const char* function_name));
 
 // Helper to retrieve Datadog::VERSION::STRING
@@ -39,9 +38,6 @@ static inline ddog_CharSlice char_slice_from_ruby_string(VALUE string) {
   return char_slice;
 }
 
-__attribute__((warn_unused_result))
-ddog_prof_Endpoint endpoint_from(VALUE exporter_configuration);
-
 __attribute__((warn_unused_result))
 ddog_Vec_Tag convert_tags(VALUE tags_as_array);
 

data/ext/datadog_profiling_native_extension/extconf.rb

@@ -74,35 +74,25 @@ Logging.message("[datadog] Using compiler:\n")
 xsystem("#{CONFIG["CC"]} -v")
 Logging.message("[datadog] 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)
-  if try_cflags(flag)
-    $CFLAGS << " " << flag
-  else
-    $stderr.puts("WARNING: '#{flag}' not accepted by compiler, skipping it")
-  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.
-add_compiler_flag "-Werror" if ENV["DATADOG_GEM_CI"] == "true"
+append_cflags "-Werror" if ENV["DATADOG_GEM_CI"] == "true"
 
 # 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 datadog gem dependency
 #   (https://github.com/msgpack/msgpack-ruby/blob/18ce08f6d612fe973843c366ac9a0b74c4e50599/ext/msgpack/extconf.rb#L8)
-add_compiler_flag "-std=gnu99"
+append_cflags "-std=gnu99"
 
 # 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"
+append_cflags "-Wno-unused-function"
 
 # Allow defining variables at any point in a function
-add_compiler_flag "-Wno-declaration-after-statement"
+append_cflags "-Wno-declaration-after-statement"
 
 # If we forget to include a Ruby header, the function call may still appear to work, but then
 # cause a segfault later. Let's ensure that never happens.
-add_compiler_flag "-Werror-implicit-function-declaration"
+append_cflags "-Werror-implicit-function-declaration"
 
 # The native extension is not intended to expose any symbols/functions for other native libraries to use;
 # the sole exception being `Init_datadog_profiling_native_extension` which needs to be visible for Ruby to call it when
@@ -110,14 +100,14 @@ add_compiler_flag "-Werror-implicit-function-declaration"
 #
 # By setting this compiler flag, we tell it to assume that everything is private unless explicitly stated.
 # For more details see https://gcc.gnu.org/wiki/Visibility
-add_compiler_flag "-fvisibility=hidden"
+append_cflags "-fvisibility=hidden"
 
 # Avoid legacy C definitions
-add_compiler_flag "-Wold-style-definition"
+append_cflags "-Wold-style-definition"
 
 # Enable all other compiler warnings
-add_compiler_flag "-Wall"
-add_compiler_flag "-Wextra"
+append_cflags "-Wall"
+append_cflags "-Wextra"
 
 if ENV["DDTRACE_DEBUG"] == "true"
   $defs << "-DDD_DEBUG"
@@ -153,6 +143,12 @@ $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 GVL instrumentation API and APIs created to support it
+$defs << "-DNO_GVL_INSTRUMENTATION" if RUBY_VERSION < "3.2"
+
+# Supporting GVL instrumentation on 3.2 needs some workarounds
+$defs << "-DUSE_GVL_PROFILING_3_2_WORKAROUNDS" if RUBY_VERSION.start_with?("3.2")
+
 # 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"
 
@@ -255,21 +251,42 @@ if Datadog::Profiling::NativeExtensionHelpers::CAN_USE_MJIT_HEADER
 
   # Warn on unused parameters to functions. Use `DDTRACE_UNUSED` to mark things as known-to-not-be-used.
   # See the comment on the same flag below for why this is done last.
-  add_compiler_flag "-Wunused-parameter"
+  append_cflags "-Wunused-parameter"
 
   create_makefile EXTENSION_NAME
 else
   # 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.
+  # the datadog-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
+  # see https://github.com/DataDog/datadog-ruby_core_source for details
 
   create_header
 
-  require "debase/ruby_core_source"
+  require "datadog/ruby_core_source"
   dir_config("ruby") # allow user to pass in non-standard core include directory
 
-  Debase::RubyCoreSource
+  # This is a workaround for a weird issue...
+  #
+  # The mkmf tool defines a `with_cppflags` helper that datadog-ruby_core_source uses. This helper temporarily
+  # replaces `$CPPFLAGS` (aka the C pre-processor [not c++!] flags) with a different set when doing something.
+  #
+  # The datadog-ruby_core_source gem uses `with_cppflags` during makefile generation to inject extra headers into the
+  # path. But because `with_cppflags` replaces `$CPPFLAGS`, well, the default `$CPPFLAGS` are not included in the
+  # makefile.
+  #
+  # This is a problem because the default `$CPPFLAGS` carries configuration that was set when Ruby was being built.
+  # Thus, if we ignore it, we don't compile the profiler with the exact same configuration as Ruby.
+  # In practice, this can generate crashes and weird bugs if the Ruby configuration is tweaked in a manner that
+  # changes some of the internal structures that the profiler relies on. Concretely, setting for instance
+  # `VM_CHECK_MODE=1` when building Ruby will trigger this issue (because somethings in structures the profiler reads
+  # are ifdef'd out using this setting).
+  #
+  # To workaround this issue, we override `with_cppflags` for datadog-ruby_core_source to still include `$CPPFLAGS`.
+  Datadog::RubyCoreSource.define_singleton_method(:with_cppflags) do |newflags, &block|
+    super("#{newflags} #{$CPPFLAGS}", &block)
+  end
+
+  Datadog::RubyCoreSource
     .create_makefile_with_core(
       proc do
         headers_available =
@@ -282,7 +299,7 @@ else
           # This is added as late as possible because in some Rubies we support (e.g. 3.3), adding this flag before
           # checking if internal VM headers are available causes those checks to fail because of this warning (and not
           # because the headers are not available.)
-          add_compiler_flag "-Wunused-parameter"
+          append_cflags "-Wunused-parameter"
         end
 
         headers_available

data/ext/datadog_profiling_native_extension/gvl_profiling_helper.c

@@ -0,0 +1,50 @@
+#include <ruby.h>
+#include <ruby/thread.h>
+#include "gvl_profiling_helper.h"
+
+#if !defined(NO_GVL_INSTRUMENTATION) && !defined(USE_GVL_PROFILING_3_2_WORKAROUNDS) // Ruby 3.3+
+  rb_internal_thread_specific_key_t gvl_waiting_tls_key;
+
+  void gvl_profiling_init(void) {
+    gvl_waiting_tls_key = rb_internal_thread_specific_key_create();
+  }
+
+#endif
+
+#ifdef USE_GVL_PROFILING_3_2_WORKAROUNDS // Ruby 3.2
+  __thread gvl_profiling_thread gvl_waiting_tls;
+  static bool gvl_profiling_state_thread_tracking_workaround_installed = false;
+
+  static void on_thread_start(
+    DDTRACE_UNUSED rb_event_flag_t _unused1,
+    DDTRACE_UNUSED const rb_internal_thread_event_data_t *_unused2,
+    DDTRACE_UNUSED void *_unused3
+  ) {
+    gvl_waiting_tls = (gvl_profiling_thread) {.thread = NULL};
+  }
+
+  // Hack: We're using the gvl_waiting_tls native thread-local to store per-thread information. Unfortunately, Ruby puts a big hole
+  // in our plan because it reuses native threads -- specifically, in Ruby 3.2, native threads are still 1:1 to Ruby
+  // threads (M:N wasn't a thing yet) BUT once a Ruby thread dies, the VM will keep the native thread around for a
+  // bit, and if another Ruby thread starts right after, Ruby will reuse the native thread, rather than create a new one.
+  //
+  // This will mean that the new Ruby thread will still have the same native thread-local data that we set on the
+  // old thread. For the purposes of our tracking, where we're keeping a pointer to the current thread object in
+  // thread-local storage **this is disastrous** since it means we'll be pointing at the wrong thread (and its
+  // memory may have been freed or reused since!)
+  //
+  // To work around this issue, once GVL profiling is enabled, we install an event hook on thread start
+  // events that clears the thread-local data. This guarantees that there will be no stale data -- any existing
+  // data will be cleared at thread start.
+  //
+  // Note that once installed, this event hook becomes permanent -- stopping the profiler does not stop this event
+  // hook, unlike all others. This is because we can't afford to miss any thread start events while the
+  // profiler is stopped (e.g. during reconfiguration) as that would mean stale data once the profiler starts again.
+  void gvl_profiling_state_thread_tracking_workaround(void) {
+    if (gvl_profiling_state_thread_tracking_workaround_installed) return;
+
+    rb_internal_thread_add_event_hook(on_thread_start, RUBY_INTERNAL_THREAD_EVENT_STARTED, NULL);
+
+    gvl_profiling_state_thread_tracking_workaround_installed = true;
+  }
+#endif

data/ext/datadog_profiling_native_extension/gvl_profiling_helper.h

@@ -0,0 +1,75 @@
+#pragma once
+
+// This helper is used by the Datadog::Profiling::Collectors::ThreadContext to store data used when profiling the GVL.
+// It's tested through that class' interfaces.
+// ---
+
+#include "extconf.h"
+
+#if !defined(NO_GVL_INSTRUMENTATION) && !defined(USE_GVL_PROFILING_3_2_WORKAROUNDS) // Ruby 3.3+
+  #include <ruby.h>
+  #include <ruby/thread.h>
+  #include "datadog_ruby_common.h"
+
+  typedef struct { VALUE thread; } gvl_profiling_thread;
+  extern rb_internal_thread_specific_key_t gvl_waiting_tls_key;
+
+  void gvl_profiling_init(void);
+
+  static inline gvl_profiling_thread thread_from_thread_object(VALUE thread) {
+    return (gvl_profiling_thread) {.thread = thread};
+  }
+
+  static inline gvl_profiling_thread thread_from_event(const rb_internal_thread_event_data_t *event_data) {
+    return thread_from_thread_object(event_data->thread);
+  }
+
+  static inline intptr_t gvl_profiling_state_get(gvl_profiling_thread thread) {
+    return (intptr_t) rb_internal_thread_specific_get(thread.thread, gvl_waiting_tls_key);
+  }
+
+  static inline void gvl_profiling_state_set(gvl_profiling_thread thread, intptr_t value) {
+    rb_internal_thread_specific_set(thread.thread, gvl_waiting_tls_key, (void *) value);
+  }
+#endif
+
+#ifdef USE_GVL_PROFILING_3_2_WORKAROUNDS // Ruby 3.2
+  typedef struct { void *thread; } gvl_profiling_thread;
+  extern __thread gvl_profiling_thread gvl_waiting_tls;
+
+  static inline void gvl_profiling_init(void) { }
+
+  // This header gets included in private_vm_access.c which can't include datadog_ruby_common.h so we replicate this
+  // helper here
+  #ifdef __GNUC__
+    #define DDTRACE_UNUSED  __attribute__((unused))
+  #else
+    #define DDTRACE_UNUSED
+  #endif
+
+  // NOTE: This is a hack that relies on the knowledge that on Ruby 3.2 the
+  // RUBY_INTERNAL_THREAD_EVENT_READY and RUBY_INTERNAL_THREAD_EVENT_RESUMED events always get called on the thread they
+  // are about. Thus, we can use our thread local storage hack to get this data, even though the event doesn't include it.
+  static inline gvl_profiling_thread thread_from_event(DDTRACE_UNUSED const void *event_data) {
+    return gvl_waiting_tls;
+  }
+
+  void gvl_profiling_state_thread_tracking_workaround(void);
+  gvl_profiling_thread gvl_profiling_state_maybe_initialize(void);
+
+  // Implementing these on Ruby 3.2 requires access to private VM things, so the following methods are
+  // implemented in `private_vm_api_access.c`
+  gvl_profiling_thread thread_from_thread_object(VALUE thread);
+  intptr_t gvl_profiling_state_get(gvl_profiling_thread thread);
+  void gvl_profiling_state_set(gvl_profiling_thread thread, intptr_t value);
+#endif
+
+#ifndef NO_GVL_INSTRUMENTATION // For all Rubies supporting GVL profiling (3.2+)
+  static inline intptr_t gvl_profiling_state_thread_object_get(VALUE thread) {
+    return gvl_profiling_state_get(thread_from_thread_object(thread));
+  }
+
+  static inline void gvl_profiling_state_thread_object_set(VALUE thread, intptr_t value) {
+    gvl_profiling_state_set(thread_from_thread_object(thread), value);
+  }
+#endif