feat(tracing): Backfill missing sample_rand on PropagationContext

Whenever the `PropagationContext` continues an incoming trace (i.e. whenever the `trace_id` is set, rather than being randomly generated as for a new trace), check if the `sample_rand` is present and valid in the incoming DSC. If the `sample_rand` is missing, generate it deterministically based on the `trace_id` and backfill it into the DSC on the `PropagationContext`.

When generating the backfilled `sample_rand`, we ensure the generated value is consistent with the incoming trace's sampling decision and sample rate, if both of these are present. Otherwise, we generate a new value in the range [0, 1).

Future PRs will address propagating the `sample_rand` to transactions generated with `continue_trace` (allowing the `sample_rand` to be propagated on outgoing traces), and will also allow `sample_rand` to be used for making sampling decisions.

Ref #3998

Author: szokeasaurusrex

sentry_sdk/tracing_utils.py

@@ -1,11 +1,13 @@
 import contextlib
+from decimal import ROUND_DOWN, Decimal
 import inspect
 import os
 import re
 import sys
 from collections.abc import Mapping
 from datetime import timedelta
 from functools import wraps
+from random import Random
 from urllib.parse import quote, unquote
 import uuid
 
@@ -19,6 +21,7 @@
     match_regex_list,
     qualname_from_function,
     to_string,
+    try_convert,
     is_sentry_url,
     _is_external_source,
     _is_in_project_root,
@@ -45,6 +48,7 @@
     "[ \t]*$"  # whitespace
 )
 
+
 # This is a normal base64 regex, modified to reflect that fact that we strip the
 # trailing = or == off
 base64_stripped = (
@@ -418,13 +422,17 @@ def from_incoming_data(cls, incoming_data):
                     propagation_context = PropagationContext()
                 propagation_context.update(sentrytrace_data)
 
+        if propagation_context is not None:
+            propagation_context._fill_sample_rand()
+
         return propagation_context
 
     @property
     def trace_id(self):
         # type: () -> str
         """The trace id of the Sentry trace."""
         if not self._trace_id:
+            # New trace, don't fill in sample_rand
             self._trace_id = uuid.uuid4().hex
 
         return self._trace_id
@@ -469,6 +477,60 @@ def __repr__(self):
             self.dynamic_sampling_context,
         )
 
+    def _fill_sample_rand(self):
+        # type: () -> None
+        """
+        Ensure that there is a valid sample_rand value in the dynamic_sampling_context.
+
+        If there is a valid sample_rand value in the dynamic_sampling_context, we keep it.
+        Otherwise, we generate a sample_rand value according to the following:
+
+          - If we have a parent_sampled value and a sample_rate in the DSC, we compute
+            a sample_rand value randomly in the range:
+                - [0, sample_rate) if parent_sampled is True,
+                - or, in the range [sample_rate, 1) if parent_sampled is False.
+
+          - If either parent_sampled or sample_rate is missing, we generate a random
+            value in the range [0, 1).
+
+        The sample_rand is deterministically generated from the trace_id, if present.
+
+        This function does nothing if there is no dynamic_sampling_context.
+        """
+        if self.dynamic_sampling_context is None:
+            return
+
+        sample_rand = try_convert(
+            Decimal, self.dynamic_sampling_context.get("sample_rand")
+        )
+        if sample_rand is not None and 0 <= sample_rand < 1:
+            # sample_rand is present and valid, so don't overwrite it
+            return
+
+        # Get the sample rate and compute the transformation that will map the random value
+        # to the desired range: [0, 1), [0, sample_rate), or [sample_rate, 1).
+        sample_rate = try_convert(
+            float, self.dynamic_sampling_context.get("sample_rate")
+        )
+        lower, upper = _sample_rand_range(self.parent_sampled, sample_rate)
+
+        try:
+            sample_rand = _generate_sample_rand(self.trace_id, interval=(lower, upper))
+        except ValueError:
+            # ValueError is raised if the interval is invalid, i.e. lower >= upper.
+            # lower >= upper might happen if the incoming trace's sampled flag
+            # and sample_rate are inconsistent, e.g. sample_rate=0.0 but sampled=True.
+            # We cannot generate a sensible sample_rand value in this case.
+            logger.debug(
+                f"Could not backfill sample_rand, since parent_sampled={self.parent_sampled} "
+                f"and sample_rate={sample_rate}."
+            )
+            return
+
+        self.dynamic_sampling_context["sample_rand"] = (
+            f"{sample_rand:.6f}"  # noqa: E231
+        )
+
 
 class Baggage:
     """
@@ -748,6 +810,49 @@ def get_current_span(scope=None):
     return current_span
 
 
+def _generate_sample_rand(
+    trace_id,  # type: Optional[str]
+    *,
+    interval=(0.0, 1.0),  # type: tuple[float, float]
+):
+    # type: (...) -> Decimal
+    """Generate a sample_rand value from a trace ID.
+
+    The generated value will be pseudorandomly chosen from the provided
+    interval. Specifically, given (lower, upper) = interval, the generated
+    value will be in the range [lower, upper). The value has 6-digit precision,
+    so when printing with .6f, the value will never be rounded up.
+
+    The pseudorandom number generator is seeded with the trace ID.
+    """
+    lower, upper = interval
+    if not lower < upper:  # using `if lower >= upper` would handle NaNs incorrectly
+        raise ValueError("Invalid interval: lower must be less than upper")
+
+    rng = Random(trace_id)
+    sample_rand = upper
+    while sample_rand >= upper:
+        sample_rand = rng.uniform(lower, upper)
+
+    # Round down to exactly six decimal-digit precision.
+    return Decimal(sample_rand).quantize(Decimal("0.000001"), rounding=ROUND_DOWN)
+
+
+def _sample_rand_range(parent_sampled, sample_rate):
+    # type: (Optional[bool], Optional[float]) -> tuple[float, float]
+    """
+    Compute the lower (inclusive) and upper (exclusive) bounds of the range of values
+    that a generated sample_rand value must fall into, given the parent_sampled and
+    sample_rate values.
+    """
+    if parent_sampled is None or sample_rate is None:
+        return 0.0, 1.0
+    elif parent_sampled is True:
+        return 0.0, sample_rate
+    else:  # parent_sampled is False
+        return sample_rate, 1.0
+
+
 # Circular imports
 from sentry_sdk.tracing import (
     BAGGAGE_HEADER_NAME,

sentry_sdk/utils.py

@@ -1888,3 +1888,20 @@ def should_be_treated_as_error(ty, value):
         return False
 
     return True
+
+
+if TYPE_CHECKING:
+    T = TypeVar("T")
+
+
+def try_convert(convert_func, value):
+    # type: (Callable[[Any], T], Any) -> Optional[T]
+    """
+    Attempt to convert from an unknown type to a specific type, using the
+    given function. Return None if the conversion fails, i.e. if the function
+    raises an exception.
+    """
+    try:
+        return convert_func(value)
+    except Exception:
+        return None

tests/test_api.py

@@ -111,7 +111,7 @@ def test_continue_trace(sentry_init):
     transaction = continue_trace(
         {
             "sentry-trace": "{}-{}-{}".format(trace_id, parent_span_id, parent_sampled),
-            "baggage": "sentry-trace_id=566e3688a61d4bc888951642d6f14a19",
+            "baggage": "sentry-trace_id=566e3688a61d4bc888951642d6f14a19,sentry-sample_rand=0.123456",
         },
         name="some name",
     )
@@ -123,7 +123,8 @@ def test_continue_trace(sentry_init):
         assert propagation_context.parent_span_id == parent_span_id
         assert propagation_context.parent_sampled == parent_sampled
         assert propagation_context.dynamic_sampling_context == {
-            "trace_id": "566e3688a61d4bc888951642d6f14a19"
+            "trace_id": "566e3688a61d4bc888951642d6f14a19",
+            "sample_rand": "0.123456",
         }
 
 

tests/test_propagationcontext.py

@@ -1,6 +1,19 @@
+from unittest import mock
+from unittest.mock import Mock
+
+import pytest
+
 from sentry_sdk.tracing_utils import PropagationContext
 
 
+SAMPLED_FLAG = {
+    None: "",
+    False: "-0",
+    True: "-1",
+}
+"""Maps the `sampled` value to the flag appended to the sentry-trace header."""
+
+
 def test_empty_context():
     ctx = PropagationContext()
 
@@ -51,13 +64,15 @@ def test_lazy_uuids():
 
 def test_property_setters():
     ctx = PropagationContext()
+
     ctx.trace_id = "X234567890abcdef1234567890abcdef"
     ctx.span_id = "X234567890abcdef"
 
     assert ctx._trace_id == "X234567890abcdef1234567890abcdef"
     assert ctx.trace_id == "X234567890abcdef1234567890abcdef"
     assert ctx._span_id == "X234567890abcdef"
     assert ctx.span_id == "X234567890abcdef"
+    assert ctx.dynamic_sampling_context is None
 
 
 def test_update():
@@ -81,3 +96,87 @@ def test_update():
     assert ctx.dynamic_sampling_context is None
 
     assert not hasattr(ctx, "foo")
+
+
+def test_existing_sample_rand_kept():
+    ctx = PropagationContext(
+        trace_id="00000000000000000000000000000000",
+        dynamic_sampling_context={"sample_rand": "0.5"},
+    )
+
+    # If sample_rand was regenerated, the value would be 0.919221 based on the trace_id
+    assert ctx.dynamic_sampling_context["sample_rand"] == "0.5"
+
+
+@pytest.mark.parametrize(
+    ("parent_sampled", "sample_rate", "expected_interval"),
+    (
+        # Note that parent_sampled and sample_rate do not scale the
+        # sample_rand value, only determine the range of the value.
+        # Expected values are determined by parent_sampled, sample_rate,
+        # and the trace_id.
+        (None, None, (0.0, 1.0)),
+        (None, "0.5", (0.0, 1.0)),
+        (False, None, (0.0, 1.0)),
+        (True, None, (0.0, 1.0)),
+        (False, "0.0", (0.0, 1.0)),
+        (False, "0.01", (0.01, 1.0)),
+        (True, "0.01", (0.0, 0.01)),
+        (False, "0.1", (0.1, 1.0)),
+        (True, "0.1", (0.0, 0.1)),
+        (False, "0.5", (0.5, 1.0)),
+        (True, "0.5", (0.0, 0.5)),
+        (True, "1.0", (0.0, 1.0)),
+    ),
+)
+def test_sample_rand_filled(parent_sampled, sample_rate, expected_interval):
+    """When continuing a trace, we want to fill in the sample_rand value if it's missing."""
+    if sample_rate is not None:
+        sample_rate_str = f",sentry-sample_rate={sample_rate}"  # noqa: E231
+    else:
+        sample_rate_str = ""
+
+    # for convenience, we'll just return the lower bound of the interval
+    mock_uniform = mock.Mock(return_value=expected_interval[0])
+
+    def mock_random_class(seed):
+        assert seed == "00000000000000000000000000000000", "seed should be the trace_id"
+        rv = Mock()
+        rv.uniform = mock_uniform
+        return rv
+
+    with mock.patch("sentry_sdk.tracing_utils.Random", mock_random_class):
+        ctx = PropagationContext().from_incoming_data(
+            {
+                "sentry-trace": f"00000000000000000000000000000000-0000000000000000{SAMPLED_FLAG[parent_sampled]}",
+                # Placeholder is needed, since we only add sample_rand if sentry items are present in baggage
+                "baggage": f"sentry-placeholder=asdf{sample_rate_str}",
+            }
+        )
+
+    assert (
+        ctx.dynamic_sampling_context["sample_rand"]
+        == f"{expected_interval[0]:.6f}"  # noqa: E231
+    )
+    assert mock_uniform.call_count == 1
+    assert mock_uniform.call_args[0] == expected_interval
+
+
+def test_sample_rand_rounds_down():
+    # Mock value that should round down to 0.999_999
+    mock_uniform = mock.Mock(return_value=0.999_999_9)
+
+    def mock_random_class(_):
+        rv = Mock()
+        rv.uniform = mock_uniform
+        return rv
+
+    with mock.patch("sentry_sdk.tracing_utils.Random", mock_random_class):
+        ctx = PropagationContext().from_incoming_data(
+            {
+                "sentry-trace": "00000000000000000000000000000000-0000000000000000",
+                "baggage": "sentry-placeholder=asdf",
+            }
+        )
+
+    assert ctx.dynamic_sampling_context["sample_rand"] == "0.999999"