fix(tracing): keep W3C tracestate dd= member within 256 chars

ddtrace/internal/constants.py

@@ -152,6 +152,8 @@ def __getitem__(self, k: str) -> Any:
 DD_TRACE_TRACESTATE_MAX_BYTES = 512
 # Per W3C Trace Context, oversized list-members are preferred targets when truncating by size.
 DD_TRACE_TRACESTATE_ITEM_MAX_CHARS = 128
+# W3C limits each list-member value to 256 characters.
+W3C_DD_LIST_MEMBER_MAX_CHARS = 256
 
 SPAN_EVENTS_HAS_EXCEPTION = "_dd.span_events.has_exception"
 COLLECTOR_MAX_SIZE_PER_SPAN = 100

ddtrace/internal/opentelemetry/sampling.py

@@ -4,14 +4,14 @@
 
 from ddtrace.internal.constants import MAX_UINT_64BITS
 from ddtrace.internal.constants import SAMPLING_KNUTH_FACTOR
+from ddtrace.internal.constants import W3C_DD_LIST_MEMBER_MAX_CHARS
 from ddtrace.internal.constants import W3C_TRACESTATE_KEY
 from ddtrace.internal.utils.http import w3c_get_tracestate_list_member
 from ddtrace.internal.utils.http import w3c_update_tracestate_list_member
 
 
 _MAX_THRESHOLD = 1 << 56
 _MAX_ENCODABLE_THRESHOLD = _MAX_THRESHOLD - 1
-_MAX_OTEL_TRACESTATE_VALUE_CHARS = 256
 _VALID_RANDOM_VALUE = re.compile(r"^[0-9a-f]{14}$")
 _VALID_THRESHOLD = re.compile(r"^[0-9a-f]{1,14}$")
 
@@ -74,7 +74,7 @@ def _build_otel_member(random_value: Optional[str], threshold: Optional[str], un
     value_chars = 0
     for field in candidate_fields:
         field_chars = len(field) + (1 if fields else 0)
-        if value_chars + field_chars <= _MAX_OTEL_TRACESTATE_VALUE_CHARS:
+        if value_chars + field_chars <= W3C_DD_LIST_MEMBER_MAX_CHARS:
             fields.append(field)
             value_chars += field_chars
     return ";".join(fields)

ddtrace/internal/utils/http.py

@@ -24,6 +24,7 @@
 from ddtrace.internal.constants import DD_TRACE_TRACESTATE_MAX_ITEMS
 from ddtrace.internal.constants import DEFAULT_TIMEOUT
 from ddtrace.internal.constants import SAMPLING_DECISION_TRACE_TAG_KEY
+from ddtrace.internal.constants import W3C_DD_LIST_MEMBER_MAX_CHARS
 from ddtrace.internal.constants import W3C_TRACESTATE_ORIGIN_KEY
 from ddtrace.internal.constants import W3C_TRACESTATE_PARENT_ID_KEY
 from ddtrace.internal.constants import W3C_TRACESTATE_SAMPLING_PRIORITY_KEY
@@ -188,6 +189,9 @@ def _connector_context() -> Generator[HTTPConnection, None, None]:
     return _connector_context
 
 
+_W3C_DD_LIST_MEMBER_RESERVED_LEN = len("dd=") + len(f"{W3C_TRACESTATE_PARENT_ID_KEY}:{0:016x};")
+
+
 def w3c_get_dd_list_member(context):
     # Context -> str
     tags = []
@@ -211,7 +215,7 @@ def w3c_get_dd_list_member(context):
     if usr_id:
         tags.append("t.usr.id:{}".format(w3c_encode_tag((_W3C_TRACESTATE_INVALID_CHARS_REGEX_VALUE, "_", usr_id))))
 
-    current_tags_len = sum(len(i) for i in tags)
+    current_tags_len = _W3C_DD_LIST_MEMBER_RESERVED_LEN + len(";".join(tags))
     for k, v in _get_metas_to_propagate(context):
         if k not in [SAMPLING_DECISION_TRACE_TAG_KEY, _USER_ID_KEY]:
             # for key replace ",", "=", and characters outside the ASCII range 0x20 to 0x7E
@@ -221,11 +225,11 @@ def w3c_get_dd_list_member(context):
                 w3c_encode_tag((_W3C_TRACESTATE_INVALID_CHARS_REGEX_KEY, "_", k)),
                 w3c_encode_tag((_W3C_TRACESTATE_INVALID_CHARS_REGEX_VALUE, "_", v)),
             )
-            # we need to keep the total length under 256 char
-            potential_current_tags_len = current_tags_len + len(next_tag)
-            if not potential_current_tags_len > 256:
+            # account for ; before next tag entry
+            next_tag_len = len(next_tag) + (1 if tags else 0)
+            if current_tags_len + next_tag_len <= W3C_DD_LIST_MEMBER_MAX_CHARS:
                 tags.append(next_tag)
-                current_tags_len += len(next_tag)
+                current_tags_len += next_tag_len
             else:
                 log.debug("tracestate would exceed 256 char limit with tag: %s. Tag will not be added.", next_tag)
 

releasenotes/notes/fix-w3c-tracestate-dd-member-length-3f1c9a7e5d2b4c60.yaml

@@ -0,0 +1,5 @@
+---
+fixes:
+  - |
+    tracing: Fixes an issue where the ``dd=`` list-member of the W3C ``tracestate`` header could exceed
+    256 characters, causing proxies that enforce the W3C limit to drop the whole ``tracestate`` header.

tests/tracer/test_propagation.py

@@ -98,6 +98,38 @@ def test_inject_deep_child_propagates_trace_level_tags(tracer):  # noqa: F811
                 assert "_dd.p.other=value" in tags
 
 
+def test_inject_tracestate_dd_member_within_256_chars(tracer):  # noqa: F811
+    """The injected dd= list-member, including the "dd=" prefix, ";" separators and the
+    "p:" field added at injection, must not exceed 256 characters.
+    """
+    meta = {
+        "_dd.p.llmobs_parent_id": "10707708093473052057",
+        "_dd.p.llmobs_trace_id": "141883241939442010658227520463581072062",
+        "_dd.p.llmobs_ml_app": "datasciencetiger",
+        "_dd.p.llmobs_sid": "11a0eec0-ad0a-4f4d-8ae8-7f358ed1b153",
+        "_dd.p.llmobs_sr": "1",
+        "_dd.p.llmobs_sd": "1",
+        "_dd.p.llmobs_pagent_span_id": "14260131734498464268",
+        "_dd.p.llmobs_pagent_name": "datasciencetiger-orchestrator",
+    }
+    ctx = Context(trace_id=1234, sampling_priority=1, dd_origin="rum", meta=meta)
+    tracer.context_provider.activate(ctx)
+    with tracer.trace("global_root_span") as span:
+        headers = {}
+        HTTPPropagator.inject(span.context, headers)
+
+        dd_members = [m for m in headers[_HTTP_HEADER_TRACESTATE].split(",") if m.startswith("dd=")]
+        assert len(dd_members) == 1
+        dd_member = dd_members[0]
+        assert len(dd_member) <= 256, dd_member
+        assert dd_member.startswith(f"dd=p:{span.span_id:016x};s:1;o:rum;")
+        # Only whole tags are dropped, never truncated
+        for tag in dd_member[len("dd=") :].split(";"):
+            if tag.startswith("t.llmobs_"):
+                key, _, value = tag.partition(":")
+                assert meta[key.replace("t.", "_dd.p.", 1)] == value
+
+
 def test_inject_with_baggage_http_propagation(tracer):  # noqa: F811
     with override_global_config(dict(_propagation_http_baggage_enabled=True)):
         ctx = Context(trace_id=1234, sampling_priority=2, dd_origin="synthetics")