Merge branch 'master' into ivana/update-sample-rate

Author: sentrivana

sentry_sdk/_types.py

@@ -1,10 +1,88 @@
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, TypeVar, Union
 
 
 # Re-exported for compat, since code out there in the wild might use this variable.
 MYPY = TYPE_CHECKING
 
 
+SENSITIVE_DATA_SUBSTITUTE = "[Filtered]"
+
+
+class AnnotatedValue:
+    """
+    Meta information for a data field in the event payload.
+    This is to tell Relay that we have tampered with the fields value.
+    See:
+    https://github.com/getsentry/relay/blob/be12cd49a0f06ea932ed9b9f93a655de5d6ad6d1/relay-general/src/types/meta.rs#L407-L423
+    """
+
+    __slots__ = ("value", "metadata")
+
+    def __init__(self, value, metadata):
+        # type: (Optional[Any], Dict[str, Any]) -> None
+        self.value = value
+        self.metadata = metadata
+
+    def __eq__(self, other):
+        # type: (Any) -> bool
+        if not isinstance(other, AnnotatedValue):
+            return False
+
+        return self.value == other.value and self.metadata == other.metadata
+
+    @classmethod
+    def removed_because_raw_data(cls):
+        # type: () -> AnnotatedValue
+        """The value was removed because it could not be parsed. This is done for request body values that are not json nor a form."""
+        return AnnotatedValue(
+            value="",
+            metadata={
+                "rem": [  # Remark
+                    [
+                        "!raw",  # Unparsable raw data
+                        "x",  # The fields original value was removed
+                    ]
+                ]
+            },
+        )
+
+    @classmethod
+    def removed_because_over_size_limit(cls):
+        # type: () -> AnnotatedValue
+        """The actual value was removed because the size of the field exceeded the configured maximum size (specified with the max_request_body_size sdk option)"""
+        return AnnotatedValue(
+            value="",
+            metadata={
+                "rem": [  # Remark
+                    [
+                        "!config",  # Because of configured maximum size
+                        "x",  # The fields original value was removed
+                    ]
+                ]
+            },
+        )
+
+    @classmethod
+    def substituted_because_contains_sensitive_data(cls):
+        # type: () -> AnnotatedValue
+        """The actual value was removed because it contained sensitive information."""
+        return AnnotatedValue(
+            value=SENSITIVE_DATA_SUBSTITUTE,
+            metadata={
+                "rem": [  # Remark
+                    [
+                        "!config",  # Because of SDK configuration (in this case the config is the hard coded removal of certain django cookies)
+                        "s",  # The fields original value was substituted
+                    ]
+                ]
+            },
+        )
+
+
+T = TypeVar("T")
+Annotated = Union[AnnotatedValue, T]
+
+
 if TYPE_CHECKING:
     from collections.abc import Container, MutableMapping, Sequence
 
@@ -19,7 +97,6 @@
     from typing import Optional
     from typing import Tuple
     from typing import Type
-    from typing import Union
     from typing_extensions import Literal, TypedDict
 
     class SDKInfo(TypedDict):
@@ -101,7 +178,7 @@ class SDKInfo(TypedDict):
             "request": dict[str, object],
             "sdk": Mapping[str, object],
             "server_name": str,
-            "spans": list[dict[str, object]],
+            "spans": Annotated[list[dict[str, object]]],
             "stacktrace": dict[
                 str, object
             ],  # We access this key in the code, but I am unsure whether we ever set it
@@ -118,6 +195,7 @@ class SDKInfo(TypedDict):
             "transaction_info": Mapping[str, Any],  # TODO: We can expand on this type
             "type": Literal["check_in", "transaction"],
             "user": dict[str, object],
+            "_dropped_spans": int,
             "_metrics_summary": dict[str, object],
         },
         total=False,

sentry_sdk/client.py

@@ -5,11 +5,12 @@
 from collections.abc import Mapping
 from datetime import datetime, timezone
 from importlib import import_module
-from typing import cast, overload
+from typing import TYPE_CHECKING, List, Dict, cast, overload
 import warnings
 
 from sentry_sdk._compat import PY37, check_uwsgi_thread_support
 from sentry_sdk.utils import (
+    AnnotatedValue,
     ContextVar,
     capture_internal_exceptions,
     current_stacktrace,
@@ -45,12 +46,9 @@
 from sentry_sdk.monitor import Monitor
 from sentry_sdk.spotlight import setup_spotlight
 
-from typing import TYPE_CHECKING
-
 if TYPE_CHECKING:
     from typing import Any
     from typing import Callable
-    from typing import Dict
     from typing import Optional
     from typing import Sequence
     from typing import Type
@@ -483,12 +481,14 @@ def _prepare_event(
     ):
         # type: (...) -> Optional[Event]
 
+        previous_total_spans = None  # type: Optional[int]
+
         if event.get("timestamp") is None:
             event["timestamp"] = datetime.now(timezone.utc)
 
         if scope is not None:
             is_transaction = event.get("type") == "transaction"
-            spans_before = len(event.get("spans", []))
+            spans_before = len(cast(List[Dict[str, object]], event.get("spans", [])))
             event_ = scope.apply_to_event(event, hint, self.options)
 
             # one of the event/error processors returned None
@@ -507,13 +507,18 @@ def _prepare_event(
                 return None
 
             event = event_
-
-            spans_delta = spans_before - len(event.get("spans", []))
+            spans_delta = spans_before - len(
+                cast(List[Dict[str, object]], event.get("spans", []))
+            )
             if is_transaction and spans_delta > 0 and self.transport is not None:
                 self.transport.record_lost_event(
                     "event_processor", data_category="span", quantity=spans_delta
                 )
 
+            dropped_spans = event.pop("_dropped_spans", 0) + spans_delta  # type: int
+            if dropped_spans > 0:
+                previous_total_spans = spans_before + dropped_spans
+
         if (
             self.options["attach_stacktrace"]
             and "exception" not in event
@@ -561,6 +566,11 @@ def _prepare_event(
             if event_scrubber:
                 event_scrubber.scrub_event(event)
 
+        if previous_total_spans is not None:
+            event["spans"] = AnnotatedValue(
+                event.get("spans", []), {"len": previous_total_spans}
+            )
+
         # Postprocess the event here so that annotated types do
         # generally not surface in before_send
         if event is not None:
@@ -598,7 +608,7 @@ def _prepare_event(
             and event.get("type") == "transaction"
         ):
             new_event = None
-            spans_before = len(event.get("spans", []))
+            spans_before = len(cast(List[Dict[str, object]], event.get("spans", [])))
             with capture_internal_exceptions():
                 new_event = before_send_transaction(event, hint or {})
             if new_event is None:

sentry_sdk/consts.py

@@ -528,6 +528,7 @@ def __init__(
         profiles_sample_rate=None,  # type: Optional[float]
         profiles_sampler=None,  # type: Optional[TracesSampler]
         profiler_mode=None,  # type: Optional[ProfilerMode]
+        profile_session_sample_rate=None,  # type: Optional[float]
         auto_enabling_integrations=True,  # type: bool
         disabled_integrations=None,  # type: Optional[Sequence[sentry_sdk.integrations.Integration]]
         auto_session_tracking=True,  # type: bool

sentry_sdk/profiler/continuous_profiler.py

@@ -1,5 +1,6 @@
 import atexit
 import os
+import random
 import sys
 import threading
 import time
@@ -83,11 +84,15 @@ def setup_continuous_profiler(options, sdk_info, capture_func):
     else:
         default_profiler_mode = ThreadContinuousScheduler.mode
 
-    experiments = options.get("_experiments", {})
+    if options.get("profiler_mode") is not None:
+        profiler_mode = options["profiler_mode"]
+    else:
+        # TODO: deprecate this and just use the existing `profiler_mode`
+        experiments = options.get("_experiments", {})
 
-    profiler_mode = (
-        experiments.get("continuous_profiling_mode") or default_profiler_mode
-    )
+        profiler_mode = (
+            experiments.get("continuous_profiling_mode") or default_profiler_mode
+        )
 
     frequency = DEFAULT_SAMPLING_FREQUENCY
 
@@ -118,35 +123,26 @@ def try_autostart_continuous_profiler():
     if _scheduler is None:
         return
 
-    # Ensure that the scheduler only autostarts once per process.
-    # This is necessary because many web servers use forks to spawn
-    # additional processes. And the profiler is only spawned on the
-    # master process, then it often only profiles the main process
-    # and not the ones where the requests are being handled.
-    #
-    # Additionally, we only want this autostart behaviour once per
-    # process. If the user explicitly calls `stop_profiler`, it should
-    # be respected and not start the profiler again.
-    if not _scheduler.should_autostart():
+    if not _scheduler.is_auto_start_enabled():
         return
 
-    _scheduler.ensure_running()
+    _scheduler.manual_start()
 
 
 def start_profiler():
     # type: () -> None
     if _scheduler is None:
         return
 
-    _scheduler.ensure_running()
+    _scheduler.manual_start()
 
 
 def stop_profiler():
     # type: () -> None
     if _scheduler is None:
         return
 
-    _scheduler.teardown()
+    _scheduler.manual_stop()
 
 
 def teardown_continuous_profiler():
@@ -164,6 +160,16 @@ def get_profiler_id():
     return _scheduler.profiler_id
 
 
+def determine_profile_session_sampling_decision(sample_rate):
+    # type: (Union[float, None]) -> bool
+
+    # `None` is treated as `0.0`
+    if not sample_rate:
+        return False
+
+    return random.random() < float(sample_rate)
+
+
 class ContinuousScheduler:
     mode = "unknown"  # type: ContinuousProfilerMode
 
@@ -175,16 +181,43 @@ def __init__(self, frequency, options, sdk_info, capture_func):
         self.capture_func = capture_func
         self.sampler = self.make_sampler()
         self.buffer = None  # type: Optional[ProfileBuffer]
+        self.pid = None  # type: Optional[int]
 
         self.running = False
 
-    def should_autostart(self):
+        profile_session_sample_rate = self.options.get("profile_session_sample_rate")
+        self.sampled = determine_profile_session_sampling_decision(
+            profile_session_sample_rate
+        )
+
+    def is_auto_start_enabled(self):
         # type: () -> bool
+
+        # Ensure that the scheduler only autostarts once per process.
+        # This is necessary because many web servers use forks to spawn
+        # additional processes. And the profiler is only spawned on the
+        # master process, then it often only profiles the main process
+        # and not the ones where the requests are being handled.
+        if self.pid == os.getpid():
+            return False
+
         experiments = self.options.get("_experiments")
         if not experiments:
             return False
+
         return experiments.get("continuous_profiling_auto_start")
 
+    def manual_start(self):
+        # type: () -> None
+        if not self.sampled:
+            return
+
+        self.ensure_running()
+
+    def manual_stop(self):
+        # type: () -> None
+        self.teardown()
+
     def ensure_running(self):
         # type: () -> None
         raise NotImplementedError
@@ -277,15 +310,11 @@ def __init__(self, frequency, options, sdk_info, capture_func):
         super().__init__(frequency, options, sdk_info, capture_func)
 
         self.thread = None  # type: Optional[threading.Thread]
-        self.pid = None  # type: Optional[int]
         self.lock = threading.Lock()
 
-    def should_autostart(self):
-        # type: () -> bool
-        return super().should_autostart() and self.pid != os.getpid()
-
     def ensure_running(self):
         # type: () -> None
+
         pid = os.getpid()
 
         # is running on the right process
@@ -356,13 +385,8 @@ def __init__(self, frequency, options, sdk_info, capture_func):
         super().__init__(frequency, options, sdk_info, capture_func)
 
         self.thread = None  # type: Optional[_ThreadPool]
-        self.pid = None  # type: Optional[int]
         self.lock = threading.Lock()
 
-    def should_autostart(self):
-        # type: () -> bool
-        return super().should_autostart() and self.pid != os.getpid()
-
     def ensure_running(self):
         # type: () -> None
         pid = os.getpid()
@@ -393,7 +417,6 @@ def ensure_running(self):
                 # longer allows us to spawn a thread and we have to bail.
                 self.running = False
                 self.thread = None
-                return
 
     def teardown(self):
         # type: () -> None
@@ -407,7 +430,7 @@ def teardown(self):
         self.buffer = None
 
 
-PROFILE_BUFFER_SECONDS = 10
+PROFILE_BUFFER_SECONDS = 60
 
 
 class ProfileBuffer: