cw-datadog 2.23.0.2

data/ext/datadog_profiling_native_extension/heap_recorder.h

@@ -0,0 +1,177 @@
+#pragma once
+
+#include <datadog/profiling.h>
+#include <ruby.h>
+
+// A heap recorder keeps track of a collection of live heap objects.
+//
+// All allocations observed by this recorder for which a corresponding free was
+// not yet observed are deemed as alive and can be iterated on to produce a
+// live heap profile.
+//
+// NOTE: All public APIs of heap_recorder support receiving a NULL heap_recorder
+//       in which case the behaviour will be a noop.
+//
+// WARN: Unless otherwise stated the heap recorder APIs assume calls are done
+// under the GVL.
+typedef struct heap_recorder heap_recorder;
+
+// Extra data associated with each live object being tracked.
+typedef struct {
+  // The weight of this object from a sampling perspective.
+  //
+  // A notion of weight is preserved for each tracked object to allow for an approximate
+  // extrapolation to an unsampled view.
+  //
+  // Example: If we were sampling every 50 objects, then each sampled object
+  //          could be seen as being representative of 50 objects.
+  unsigned int weight;
+
+  // Size of this object in memory.
+  // NOTE: This only gets updated during heap_recorder_prepare_iteration and only
+  //       for those objects that meet the minimum iteration age requirements.
+  size_t size;
+
+  // The class of the object that we're tracking.
+  // NOTE: This is optional and will be set to NULL if not set.
+  ddog_prof_ManagedStringId class;
+
+  // The GC allocation gen in which we saw this object being allocated.
+  //
+  // This enables us to calculate the age of this object in terms of GC executions.
+  size_t alloc_gen;
+
+  // The age of this object in terms of GC generations.
+  // NOTE: This only gets updated during heap_recorder_prepare_iteration
+  size_t gen_age;
+
+  // Whether this object was previously seen as being frozen. If this is the case,
+  // we'll skip any further size updates since frozen objects are supposed to be
+  // immutable.
+  bool is_frozen;
+} live_object_data;
+
+// Data that is made available to iterators of heap recorder data for each live object
+// tracked therein.
+typedef struct {
+  ddog_prof_Slice_Location locations;
+  live_object_data object_data;
+} heap_recorder_iteration_data;
+
+// Initialize a new heap recorder.
+heap_recorder* heap_recorder_new(ddog_prof_ManagedStringStorage string_storage);
+
+// Free a previously initialized heap recorder.
+void heap_recorder_free(heap_recorder *heap_recorder);
+
+// Sets whether this heap recorder should keep track of sizes or not.
+//
+// If set to true, the heap recorder will attempt to determine the approximate sizes of
+// tracked objects and wield them during iteration.
+// If set to false, sizes returned during iteration should not be used/relied on (they
+// may be 0 or the last determined size before disabling the tracking of sizes).
+//
+// NOTE: Default is true, i.e., it will attempt to determine approximate sizes of tracked
+// objects.
+void heap_recorder_set_size_enabled(heap_recorder *heap_recorder, bool size_enabled);
+
+// Set sample rate used by this heap recorder.
+//
+// Controls how many recordings will be ignored before committing a heap allocation and
+// the weight of the committed heap allocation.
+//
+// A value of 1 will effectively track all objects that are passed through
+// start/end_heap_allocation_recording pairs. A value of 10 will only track every 10th
+// object passed through such calls and its effective weight for the purposes of heap
+// profiling will be multiplied by 10.
+//
+// NOTE: Default is 1, i.e., track all heap allocation recordings.
+//
+// WARN: Non-positive values will lead to an exception being thrown.
+void heap_recorder_set_sample_rate(heap_recorder *heap_recorder, int sample_rate);
+
+// Do any cleanup needed after forking.
+// WARN: Assumes this gets called before profiler is reinitialized on the fork
+void heap_recorder_after_fork(heap_recorder *heap_recorder);
+
+// Start a heap allocation recording on the heap recorder for a new object.
+//
+// This heap allocation recording needs to be ended via ::end_heap_allocation_recording
+// before it will become fully committed and able to be iterated on.
+//
+// @param new_obj
+//   The newly allocated Ruby object/value.
+// @param weight
+//   The sampling weight of this object.
+//
+// WARN: It needs to be paired with a ::end_heap_allocation_recording call.
+void start_heap_allocation_recording(heap_recorder *heap_recorder, VALUE new_obj, unsigned int weight, ddog_CharSlice alloc_class);
+
+// End a previously started heap allocation recording on the heap recorder.
+//
+// It is at this point that an allocated object will become fully tracked and able to be iterated on.
+//
+// @param locations The stacktrace representing the location of the allocation.
+//
+// WARN: It is illegal to call this without previously having called ::start_heap_allocation_recording.
+// WARN: This method rescues exceptions with `rb_protect`, returning the exception state integer for the caller to handle.
+__attribute__((warn_unused_result))
+int end_heap_allocation_recording_with_rb_protect(heap_recorder *heap_recorder, ddog_prof_Slice_Location locations);
+
+// Update the heap recorder, **checking young objects only**. The idea here is to align with GC: most young objects never
+// survive enough GC generations, and thus periodically running this method reduces memory usage (we get rid of
+// these objects quicker) and hopefully reduces tail latency (because there's less objects at serialization time to check).
+void heap_recorder_update_young_objects(heap_recorder *heap_recorder);
+
+// Update the heap recorder to reflect the latest state of the VM and prepare internal structures
+// for efficient iteration.
+//
+// WARN: This must be called strictly before iteration. Failing to do so will result in exceptions.
+void heap_recorder_prepare_iteration(heap_recorder *heap_recorder);
+
+// Optimize the heap recorder by cleaning up any data that might have been prepared specifically
+// for the purpose of iterating over the heap recorder data.
+//
+// WARN: This must be called strictly after iteration to ensure proper cleanup and to keep the memory
+// profile of the heap recorder low.
+void heap_recorder_finish_iteration(heap_recorder *heap_recorder);
+
+// Iterate over each live object being tracked by the heap recorder.
+//
+// NOTE: Iteration can be called without holding the Ruby Global VM lock.
+// WARN: This must be called strictly after heap_recorder_prepare_iteration and before
+// heap_recorder_finish_iteration.
+//
+// @param for_each_callback
+//   A callback function that shall be called for each live object being tracked
+//   by the heap recorder. Alongside the iteration_data for each live object,
+//   a second argument will be forwarded with the contents of the optional
+//   for_each_callback_extra_arg. Iteration will continue until the callback
+//   returns false or we run out of objects.
+// @param for_each_callback_extra_arg
+//   Optional (NULL if empty) extra data that should be passed to the
+//   callback function alongside the data for each live tracked object.
+// @return true if iteration ran, false if something prevented it from running.
+bool heap_recorder_for_each_live_object(
+    heap_recorder *heap_recorder,
+    bool (*for_each_callback)(heap_recorder_iteration_data data, void* extra_arg),
+    void *for_each_callback_extra_arg);
+
+// Return a Ruby hash containing a snapshot of this recorder's interesting state at calling time.
+// WARN: This allocates in the Ruby VM and therefore should not be called without the
+//       VM lock or during GC.
+VALUE heap_recorder_state_snapshot(heap_recorder *heap_recorder);
+
+// v--- TEST-ONLY APIs ---v
+
+// Returns a Ruby string with a representation of internal data helpful to
+// troubleshoot issues such as unexpected test failures.
+VALUE heap_recorder_testonly_debug(heap_recorder *heap_recorder);
+
+// Check if a given object_id is being tracked or not
+VALUE heap_recorder_testonly_is_object_recorded(heap_recorder *heap_recorder, VALUE obj_id);
+
+// Used to ensure that a GC actually triggers an update of the objects
+void heap_recorder_testonly_reset_last_update(heap_recorder *heap_recorder);
+
+void heap_recorder_testonly_benchmark_intern(heap_recorder *heap_recorder, ddog_CharSlice string, int times, bool use_all);

data/ext/datadog_profiling_native_extension/helpers.h

@@ -0,0 +1,12 @@
+#pragma once
+
+#include <stdint.h>
+
+// @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.
+static inline uint64_t uint64_max_of(uint64_t a, uint64_t b) { return a > b ? a : b; }
+static inline uint64_t uint64_min_of(uint64_t a, uint64_t b) { return a > b ? b : a; }
+static inline long long_max_of(long a, long b) { return a > b ? a : b; }
+static inline long long_min_of(long a, long b) { return a > b ? b : a; }
+static inline double double_max_of(double a, double b) { return a > b ? a : b; }
+static inline double double_min_of(double a, double b) { return a > b ? b : a; }

data/ext/datadog_profiling_native_extension/http_transport.c

@@ -0,0 +1,280 @@
+#include <ruby.h>
+#include <ruby/thread.h>
+#include <datadog/profiling.h>
+#include "helpers.h"
+#include "libdatadog_helpers.h"
+#include "ruby_helpers.h"
+#include "encoded_profile.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 VALUE library_version_string = Qnil;
+
+typedef struct {
+  ddog_prof_ProfileExporter *exporter;
+  ddog_prof_Request_Result *build_result;
+  ddog_CancellationToken *cancel_token;
+  ddog_prof_Result_HttpStatus result;
+  bool send_ran;
+} call_exporter_without_gvl_arguments;
+
+static inline ddog_ByteSlice byte_slice_from_ruby_string(VALUE string);
+static VALUE _native_validate_exporter(VALUE self, VALUE exporter_configuration);
+static ddog_prof_ProfileExporter_Result create_exporter(VALUE exporter_configuration, VALUE tags_as_array);
+static VALUE handle_exporter_failure(ddog_prof_ProfileExporter_Result exporter_result);
+static VALUE _native_do_export(
+  VALUE self,
+  VALUE exporter_configuration,
+  VALUE upload_timeout_milliseconds,
+  VALUE flush
+);
+static void *call_exporter_without_gvl(void *call_args);
+static void interrupt_exporter_call(void *cancel_token);
+
+void http_transport_init(VALUE profiling_module) {
+  VALUE 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, 3);
+
+  ok_symbol = ID2SYM(rb_intern_const("ok"));
+  error_symbol = ID2SYM(rb_intern_const("error"));
+
+  library_version_string = datadog_gem_version();
+  rb_global_variable(&library_version_string);
+}
+
+static inline 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_ProfileExporter_Result 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_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);
+
+  if (working_mode == rb_intern("agentless")) {
+    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 if (working_mode == rb_intern("agent")) {
+    VALUE base_url = rb_ary_entry(exporter_configuration, 1);
+
+    return ddog_prof_Endpoint_agent(char_slice_from_ruby_string(base_url));
+  } else {
+    rb_raise(rb_eArgError, "Failed to initialize transport: Unexpected working mode, expected :agentless or :agent");
+  }
+}
+
+static ddog_prof_ProfileExporter_Result 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_prof_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_ProfileExporter_Result exporter_result =
+    ddog_prof_Exporter_new(library_name, library_version, profiling_family, &tags, endpoint);
+
+  ddog_Vec_Tag_drop(tags);
+
+  return exporter_result;
+}
+
+static void validate_token(ddog_CancellationToken token, const char *file, int line) {
+  if (token.inner == NULL) rb_raise(rb_eRuntimeError, "Unexpected: Validation token was empty at %s:%d", file, line);
+}
+
+static VALUE handle_exporter_failure(ddog_prof_ProfileExporter_Result exporter_result) {
+  return exporter_result.tag == DDOG_PROF_PROFILE_EXPORTER_RESULT_OK_HANDLE_PROFILE_EXPORTER ?
+    Qnil :
+    rb_ary_new_from_args(2, error_symbol, get_error_details_and_drop(&exporter_result.err));
+}
+
+// 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_ProfileExporter *exporter,
+  ddog_prof_EncodedProfile *profile,
+  ddog_prof_Exporter_Slice_File files_to_compress_and_export,
+  ddog_CharSlice internal_metadata,
+  ddog_CharSlice info
+) {
+  ddog_prof_Request_Result build_result = ddog_prof_Exporter_Request_build(
+    exporter,
+    profile,
+    files_to_compress_and_export,
+    /* files_to_export_unmodified: */ ddog_prof_Exporter_Slice_File_empty(),
+    /* optional_additional_tags: */ NULL,
+    &internal_metadata,
+    &info
+  );
+
+  if (build_result.tag == DDOG_PROF_REQUEST_RESULT_ERR_HANDLE_REQUEST) {
+    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_request = ddog_CancellationToken_new();
+  ddog_CancellationToken cancel_token_interrupt = ddog_CancellationToken_clone(&cancel_token_request);
+
+  validate_token(cancel_token_request, __FILE__, __LINE__);
+  validate_token(cancel_token_interrupt, __FILE__, __LINE__);
+
+  // 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
+  call_exporter_without_gvl_arguments args =
+    {.exporter = exporter, .build_result = &build_result, .cancel_token = &cancel_token_request, .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_interrupt);
+
+    // 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_request);
+  ddog_CancellationToken_drop(&cancel_token_interrupt);
+  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_Result_HttpStatus result = args.result;
+
+  return result.tag == DDOG_PROF_RESULT_HTTP_STATUS_OK_HTTP_STATUS ?
+    rb_ary_new_from_args(2, ok_symbol, UINT2NUM(result.ok.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 flush
+) {
+  VALUE encoded_profile = rb_funcall(flush, rb_intern("encoded_profile"), 0);
+  VALUE code_provenance_file_name = rb_funcall(flush, rb_intern("code_provenance_file_name"), 0);
+  VALUE code_provenance_data = rb_funcall(flush, rb_intern("code_provenance_data"), 0);
+  VALUE tags_as_array = rb_funcall(flush, rb_intern("tags_as_array"), 0);
+  VALUE internal_metadata_json = rb_funcall(flush, rb_intern("internal_metadata_json"), 0);
+  VALUE info_json = rb_funcall(flush, rb_intern("info_json"), 0);
+
+  ENFORCE_TYPE(upload_timeout_milliseconds, T_FIXNUM);
+  enforce_encoded_profile_instance(encoded_profile);
+  ENFORCE_TYPE(code_provenance_file_name, T_STRING);
+  ENFORCE_TYPE(tags_as_array, T_ARRAY);
+  ENFORCE_TYPE(internal_metadata_json, T_STRING);
+  ENFORCE_TYPE(info_json, 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);
+
+  int to_compress_length = have_code_provenance ? 1 : 0;
+  ddog_prof_Exporter_File to_compress[to_compress_length];
+  ddog_prof_Exporter_Slice_File files_to_compress_and_export = {.ptr = to_compress, .len = to_compress_length};
+
+  if (have_code_provenance) {
+    to_compress[0] = (ddog_prof_Exporter_File) {
+      .name = char_slice_from_ruby_string(code_provenance_file_name),
+      .file = byte_slice_from_ruby_string(code_provenance_data),
+    };
+  }
+
+  ddog_CharSlice internal_metadata = char_slice_from_ruby_string(internal_metadata_json);
+  ddog_CharSlice info = char_slice_from_ruby_string(info_json);
+
+  ddog_prof_ProfileExporter_Result 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;
+
+  ddog_VoidResult timeout_result = ddog_prof_Exporter_set_timeout(&exporter_result.ok, timeout_milliseconds);
+  if (timeout_result.tag == DDOG_VOID_RESULT_ERR) {
+    // NOTE: Seems a bit harsh to fail the upload if we can't set a timeout. OTOH, this is only expected to fail
+    // if the exporter is not well built. Because such a situation should already be caught above I think it's
+    // preferable to leave this here as a virtually unreachable exception rather than ignoring it.
+    ddog_prof_Exporter_drop(&exporter_result.ok);
+    return rb_ary_new_from_args(2, error_symbol, get_error_details_and_drop(&timeout_result.err));
+  }
+
+  return perform_export(
+    &exporter_result.ok,
+    to_ddog_prof_EncodedProfile(encoded_profile),
+    files_to_compress_and_export,
+    internal_metadata,
+    info
+  );
+}
+
+static void *call_exporter_without_gvl(void *call_args) {
+  call_exporter_without_gvl_arguments *args = (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) {
+  // TODO: False here can mean two things: it was already cancelled OR it failed to cancel.
+  // Would be nice to change libdatadog to be able to distinguish between them...
+  ddog_CancellationToken_cancel((ddog_CancellationToken *) cancel_token);
+}

data/ext/datadog_profiling_native_extension/libdatadog_helpers.c

@@ -0,0 +1,84 @@
+#include "libdatadog_helpers.h"
+
+#include <ruby.h>
+
+const char *ruby_value_type_to_string(enum ruby_value_type type) {
+  return ruby_value_type_to_char_slice(type).ptr;
+}
+
+ddog_CharSlice ruby_value_type_to_char_slice(enum ruby_value_type type) {
+  switch (type) {
+    case(RUBY_T_NONE    ): return DDOG_CHARSLICE_C("T_NONE");
+    case(RUBY_T_OBJECT  ): return DDOG_CHARSLICE_C("T_OBJECT");
+    case(RUBY_T_CLASS   ): return DDOG_CHARSLICE_C("T_CLASS");
+    case(RUBY_T_MODULE  ): return DDOG_CHARSLICE_C("T_MODULE");
+    case(RUBY_T_FLOAT   ): return DDOG_CHARSLICE_C("T_FLOAT");
+    case(RUBY_T_STRING  ): return DDOG_CHARSLICE_C("T_STRING");
+    case(RUBY_T_REGEXP  ): return DDOG_CHARSLICE_C("T_REGEXP");
+    case(RUBY_T_ARRAY   ): return DDOG_CHARSLICE_C("T_ARRAY");
+    case(RUBY_T_HASH    ): return DDOG_CHARSLICE_C("T_HASH");
+    case(RUBY_T_STRUCT  ): return DDOG_CHARSLICE_C("T_STRUCT");
+    case(RUBY_T_BIGNUM  ): return DDOG_CHARSLICE_C("T_BIGNUM");
+    case(RUBY_T_FILE    ): return DDOG_CHARSLICE_C("T_FILE");
+    case(RUBY_T_DATA    ): return DDOG_CHARSLICE_C("T_DATA");
+    case(RUBY_T_MATCH   ): return DDOG_CHARSLICE_C("T_MATCH");
+    case(RUBY_T_COMPLEX ): return DDOG_CHARSLICE_C("T_COMPLEX");
+    case(RUBY_T_RATIONAL): return DDOG_CHARSLICE_C("T_RATIONAL");
+    case(RUBY_T_NIL     ): return DDOG_CHARSLICE_C("T_NIL");
+    case(RUBY_T_TRUE    ): return DDOG_CHARSLICE_C("T_TRUE");
+    case(RUBY_T_FALSE   ): return DDOG_CHARSLICE_C("T_FALSE");
+    case(RUBY_T_SYMBOL  ): return DDOG_CHARSLICE_C("T_SYMBOL");
+    case(RUBY_T_FIXNUM  ): return DDOG_CHARSLICE_C("T_FIXNUM");
+    case(RUBY_T_UNDEF   ): return DDOG_CHARSLICE_C("T_UNDEF");
+    case(RUBY_T_IMEMO   ): return DDOG_CHARSLICE_C("T_IMEMO");
+    case(RUBY_T_NODE    ): return DDOG_CHARSLICE_C("T_NODE");
+    case(RUBY_T_ICLASS  ): return DDOG_CHARSLICE_C("T_ICLASS");
+    case(RUBY_T_ZOMBIE  ): return DDOG_CHARSLICE_C("T_ZOMBIE");
+    #ifndef NO_T_MOVED
+    case(RUBY_T_MOVED   ): return DDOG_CHARSLICE_C("T_MOVED");
+    #endif
+                  default: return DDOG_CHARSLICE_C("BUG: Unknown value for ruby_value_type");
+  }
+}
+
+size_t read_ddogerr_string_and_drop(ddog_Error *error, char *string, size_t capacity) {
+  if (capacity == 0 || string == NULL) {
+    // short-circuit, we can't write anything
+    ddog_Error_drop(error);
+    return 0;
+  }
+
+  ddog_CharSlice error_msg_slice = ddog_Error_message(error);
+  size_t error_msg_size = error_msg_slice.len;
+  // Account for extra null char for proper cstring
+  if (error_msg_size >= capacity) {
+    // Error message too big, lets truncate it to capacity - 1 to allow for extra null at end
+    error_msg_size = capacity - 1;
+  }
+  strncpy(string, error_msg_slice.ptr, error_msg_size);
+  string[error_msg_size] = '\0';
+  ddog_Error_drop(error);
+  return error_msg_size;
+}
+
+ddog_prof_ManagedStringId intern_or_raise(ddog_prof_ManagedStringStorage string_storage, ddog_CharSlice string) {
+  if (string.len == 0) return (ddog_prof_ManagedStringId) { 0 }; // Id 0 is always an empty string, no need to ask
+
+  ddog_prof_ManagedStringStorageInternResult intern_result = ddog_prof_ManagedStringStorage_intern(string_storage, string);
+  if (intern_result.tag == DDOG_PROF_MANAGED_STRING_STORAGE_INTERN_RESULT_ERR) {
+    rb_raise(rb_eRuntimeError, "Failed to intern string: %"PRIsVALUE, get_error_details_and_drop(&intern_result.err));
+  }
+  return intern_result.ok;
+}
+
+void intern_all_or_raise(
+  ddog_prof_ManagedStringStorage string_storage,
+  ddog_prof_Slice_CharSlice strings,
+  ddog_prof_ManagedStringId *output_ids,
+  uintptr_t output_ids_size
+) {
+  ddog_prof_MaybeError result = ddog_prof_ManagedStringStorage_intern_all(string_storage, strings, output_ids, output_ids_size);
+  if (result.tag == DDOG_PROF_OPTION_ERROR_SOME_ERROR) {
+    rb_raise(rb_eRuntimeError, "Failed to intern_all: %"PRIsVALUE, get_error_details_and_drop(&result.some));
+  }
+}

data/ext/datadog_profiling_native_extension/libdatadog_helpers.h

@@ -0,0 +1,28 @@
+#pragma once
+
+#include <datadog/profiling.h>
+#include "ruby_helpers.h"
+
+static inline VALUE ruby_string_from_vec_u8(ddog_Vec_U8 string) {
+  return rb_str_new((char *) string.ptr, string.len);
+}
+
+// Utility function to be able to extract an error cstring from a ddog_Error.
+// Returns the amount of characters written to string (which are necessarily
+// bounded by capacity - 1 since the string will be null-terminated).
+size_t read_ddogerr_string_and_drop(ddog_Error *error, char *string, size_t capacity);
+
+// Used for pretty printing this Ruby enum. Returns "T_UNKNOWN_OR_MISSING_RUBY_VALUE_TYPE_ENTRY" for unknown elements.
+// In practice, there's a few types that the profiler will probably never encounter, but I've added all entries of
+// ruby_value_type that Ruby uses so that we can also use this for debugging.
+const char *ruby_value_type_to_string(enum ruby_value_type type);
+ddog_CharSlice ruby_value_type_to_char_slice(enum ruby_value_type type);
+
+ddog_prof_ManagedStringId intern_or_raise(ddog_prof_ManagedStringStorage string_storage, ddog_CharSlice string);
+
+void intern_all_or_raise(
+  ddog_prof_ManagedStringStorage string_storage,
+  ddog_prof_Slice_CharSlice strings,
+  ddog_prof_ManagedStringId *output_ids,
+  uintptr_t output_ids_size
+);