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

@@ -6,6 +6,7 @@
 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 +20,7 @@
     match_regex_list,
     qualname_from_function,
     to_string,
+    try_float,
     is_sentry_url,
     _is_external_source,
     _is_in_project_root,
@@ -418,13 +420,17 @@ 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 +475,55 @@ 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_float(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_float(self.dynamic_sampling_context.get("sample_rate"))
+        lower, upper = _sample_rand_range(self.parent_sampled, sample_rate)
+
+        if 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.
+            return
+
+        random = Random(self.trace_id)
+        sample_rand = upper
+        while sample_rand == upper:
+            # The built-in uniform() method can, in some cases, return the
+            # upper bound. We request a new value until we get a different
+            # value.
+            sample_rand = random.uniform(lower, upper)
+
+        self.dynamic_sampling_context["sample_rand"] = str(sample_rand)
+
 
 class Baggage:
     """
@@ -748,6 +803,21 @@ def get_current_span(scope=None):
     return current_span
 
 
+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 +1888,12 @@ def should_be_treated_as_error(ty, value):
         return False
 
     return True
+
+
+def try_float(value):
+    # type: (Any) -> Optional[float]
+    """Small utility to convert a value to a float, if possible."""
+    try:
+        return float(value)
+    except (ValueError, TypeError):
+        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=0.1234567890",
         },
         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": "0.1234567890",
         }
 
 

tests/test_propagationcontext.py

@@ -1,6 +1,16 @@
+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 +61,15 @@ 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 +93,46 @@ 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 0.8766381713144122 based on the trace_id
+    assert ctx.dynamic_sampling_context["sample_rand"] == "0.5"
+
+
+@pytest.mark.parametrize(
+    ("parent_sampled", "sample_rate", "expected_sample_rand"),
+    (
+        (None, None, "0.8766381713144122"),
+        (None, "0.5", "0.8766381713144122"),
+        (False, None, "0.8766381713144122"),
+        (True, None, "0.8766381713144122"),
+        (False, "0.0", "0.8766381713144122"),
+        (False, "0.01", "0.8778717896012681"),
+        (True, "0.01", "0.008766381713144122"),
+        (False, "0.1", "0.888974354182971"),
+        (True, "0.1", "0.08766381713144122"),
+        (False, "0.5", "0.9383190856572061"),
+        (True, "0.5", "0.4383190856572061"),
+        (True, "1.0", "0.8766381713144122"),
+    ),
+)
+def test_sample_rand_filled(parent_sampled, sample_rate, expected_sample_rand):
+    """When continuing a trace, we want to fill in the sample_rand value if it's missing."""
+    dsc = {}
+    if sample_rate is not None:
+        dsc["sample_rate"] = sample_rate
+
+    ctx = PropagationContext().from_incoming_data(
+        {
+            "sentry-trace": f"00000000000000000000000000000000-0000000000000000{SAMPLED_FLAG[parent_sampled]}",
+            "baggage": f"sentry-sample_rate={sample_rate}",
+        }
+    )
+
+    assert ctx.dynamic_sampling_context["sample_rand"] == expected_sample_rand