perf(profiling): implement and use cheaper groupby (#4936)

## Description

The use of the `itertools.groupby` function requires the collection that
is passed to it to be sorted for the intended use. We make our own
implementation of the `groupby` operation to avoid incurring in the
extra cost of sorting, thus keeping the time complexity linear with the
collection size.

## Checklist
- [ ] Followed the [library release note
guidelines](https://ddtrace.readthedocs.io/en/stable/contributing.html#Release-Note-Guidelines)
when writing a release note.
- [ ] Add additional sections for `feat` and `fix` pull requests.
- [ ] [Library
documentation](https://github.com/DataDog/dd-trace-py/tree/1.x/docs)
and/or [Datadog's documentation
site](https://github.com/DataDog/documentation/) is updated. Link to doc
PR in description.

## Reviewer Checklist
- [ ] Title is accurate.
- [ ] Description motivates each change.
- [ ] No unnecessary changes were introduced in this PR.
- [ ] Avoid breaking
[API](https://ddtrace.readthedocs.io/en/stable/versioning.html#interfaces)
changes unless absolutely necessary.
- [ ] Tests provided or description of manual testing performed is
included in the code or PR.
- [ ] Release note has been added and follows the [library release note
guidelines](https://ddtrace.readthedocs.io/en/stable/contributing.html#Release-Note-Guidelines),
or else `changelog/no-changelog` label added.
- [ ] All relevant GitHub issues are correctly linked.
- [ ] Change contains telemetry where appropriate (logs, metrics, etc.).
- [ ] Telemetry is meaningful, actionable and does not have the
potential to leak sensitive data.

Author: P403n1x87

ddtrace/profiling/exporter/pprof.pyx

@@ -1,5 +1,4 @@
 import collections
-import itertools
 import operator
 import platform
 import sysconfig
@@ -143,10 +142,8 @@ class _StringTable(object):
         return len(self._strings)
 
 
-def _none_to_str(value: typing.Any) -> str:
-    if value is None:
-        return ""
-    return str(value)
+cdef str _none_to_str(object value):
+    return "" if value is None else str(value)
 
 
 def _get_thread_name(thread_id: typing.Optional[int], thread_name: typing.Optional[str]) -> str:
@@ -155,6 +152,15 @@ def _get_thread_name(thread_id: typing.Optional[int], thread_name: typing.Option
     return thread_name
 
 
+cdef groupby(object collection, object key):
+    cdef dict groups = {}
+
+    for item in collection:
+        groups.setdefault(key(item), []).append(item)
+
+    return groups.items()
+
+
 class pprof_LocationType(object):
     # pprof_pb2.Location
     id: int
@@ -460,20 +466,15 @@ class _PprofConverter(object):
                     "paths": sorted(lib_and_filename[1] for lib_and_filename in libs_and_filenames),
                 }
             )
-            for lib, libs_and_filenames in itertools.groupby(
-                sorted(
-                    set(
-                        filter(
-                            lambda p: p[0] is not None,
-                            (
-                                (_packages.filename_to_package(filename), filename)
-                                for filename, lineno, funcname in self._locations
-                            ),
-                        )
-                    ),
-                    key=_ITEMGETTER_ZERO,
-                ),
-                key=_ITEMGETTER_ZERO,
+            for lib, libs_and_filenames in groupby(
+                {
+                    _
+                    for _ in (
+                        (_packages.filename_to_package(filename), filename)
+                        for filename, lineno, funcname in self._locations
+                    )
+                    if _[0] is not None
+                }, _ITEMGETTER_ZERO
             )
         ] + STDLIB
 
@@ -601,10 +602,7 @@ class PprofExporter(exporter.Exporter):
     def _group_stack_events(
         self, events: typing.Iterable[event.StackBasedEvent]
     ) -> typing.Iterator[typing.Tuple[StackEventGroupKey, typing.Iterator[event.StackBasedEvent]]]:
-        return itertools.groupby(
-            sorted(events, key=self._stack_event_group_key),
-            key=self._stack_event_group_key,
-        )
+        return groupby(events, self._stack_event_group_key)
 
     def _lock_event_group_key(
         self,
@@ -627,10 +625,7 @@ class PprofExporter(exporter.Exporter):
     def _group_lock_events(
         self, events: typing.Iterable[_lock.LockEventBase]
     ) -> typing.Iterator[typing.Tuple[LockEventGroupKey, typing.Iterator[_lock.LockEventBase]]]:
-        return itertools.groupby(
-            sorted(events, key=self._lock_event_group_key),
-            key=self._lock_event_group_key,
-        )
+        return groupby(events, self._lock_event_group_key)
 
     def _stack_exception_group_key(self, event: stack_event.StackExceptionSampleEvent) -> StackExceptionEventGroupKey:
         exc_type = event.exc_type
@@ -654,10 +649,7 @@ class PprofExporter(exporter.Exporter):
     ) -> typing.Iterator[
         typing.Tuple[StackExceptionEventGroupKey, typing.Iterator[stack_event.StackExceptionSampleEvent]]
     ]:
-        return itertools.groupby(
-            sorted(events, key=self._stack_exception_group_key),
-            key=self._stack_exception_group_key,
-        )
+        return groupby(events, self._stack_exception_group_key)
 
     def _get_event_trace_resource(self, event: event.StackBasedEvent) -> str:
         trace_resource = ""

mypy.ini

@@ -1,7 +1,6 @@
 [mypy]
 files = ddtrace,
       ddtrace/profiling/_build.pyx,
-      ddtrace/profiling/exporter/pprof.pyx,
       docs
 # mypy thinks .pyx files are scripts and errors out if it finds multiple scripts
 scripts_are_modules = true

pyproject.toml

@@ -32,6 +32,7 @@ exclude = '''
   | ddtrace/profiling/_threading.pyx$
   | ddtrace/profiling/collector/stack.pyx$
   | ddtrace/profiling/exporter/pprof_.*pb2.py$
+  | ddtrace/profiling/exporter/pprof.pyx$
   | ddtrace/vendor/
   | \.eggs
   | \.git