fix(otel): tolerate a bogus exclusion env when callback config wins

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/integrations/otel/model/config.py

@@ -370,16 +370,19 @@ def _db_system_for_excluded_service(service: str) -> str:
     return resolved
 
 
-def validate_otel_v2_excluded_services_env() -> None:
+def validate_otel_v2_excluded_services_env(settings: object) -> None:
     """Validate ``LITELLM_OTEL_EXCLUDED_SERVICES`` at boot even with no ``otel`` callback.
 
     Preset-only deployments build env-only configs through a path that swallows
     init errors, so a bogus value would otherwise degrade to the legacy callback
     silently. Splitting and normalizing here raises the same ``ValueError`` the
-    field raises.
+    field raises. An explicit ``callback_settings.otel.excluded_services`` wins
+    over the env var, so a bad env value is inert then and must not block boot.
     """
     if not is_otel_v2_enabled():
         return
+    if isinstance(settings, Mapping) and "excluded_services" in settings:
+        return
     raw: Final = os.environ.get("LITELLM_OTEL_EXCLUDED_SERVICES")
     if not raw:
         return

litellm/proxy/common_utils/callback_utils.py

@@ -181,7 +181,7 @@ def initialize_callbacks_on_proxy(
 
             from litellm.integrations.otel.model.config import validate_otel_v2_excluded_services_env
 
-            validate_otel_v2_excluded_services_env()
+            validate_otel_v2_excluded_services_env(callback_specific_params.get("otel") if "otel" in value else None)
 
             # check if callback is a custom logger compatible callback
             if isinstance(callback, str):

tests/integration/observability/test_otel_excluded_services.py

@@ -162,6 +162,7 @@ def _guardrail_block(config: dict) -> None:
     ]
 
 
+@pytest.mark.timeout(180)
 def test_excluded_services_drops_db_spans_at_tenant_only(
     gateway: Gateway,
     audit_sinks: SpanSinks,
@@ -288,6 +289,35 @@ def test_bogus_excluded_service_fails_proxy_start(
     assert "postgres, redis" in text, text[-3000:]
 
 
+def test_valid_config_excluded_services_tolerates_bogus_env(
+    gateway: Gateway,
+    audit_sinks: SpanSinks,
+    otel_audit_config: AuditConfigWriter,
+    langfuse_vars: dict[str, JsonValue],
+    tmp_path: Path,
+) -> None:
+    def with_langfuse_otel(config: dict) -> None:
+        config["litellm_settings"]["callbacks"] = ["otel", "langfuse_otel"]
+
+    config: Final = _config_with(
+        tmp_path, otel_audit_config, otel={"excluded_services": ["postgres"]}, extra=with_langfuse_otel
+    )
+    with owned_proxy(
+        gateway, tmp_path, {"LITELLM_OTEL_V2": "1", "LITELLM_OTEL_EXCLUDED_SERVICES": "auth"}, config=config, workers=2
+    ) as candidate:
+        start, _ = recorded_spans(audit_sinks.tenant)
+        traffic: Final = _drive(candidate, langfuse_vars)
+        tenant_trace: Final = _trace_id(audit_sinks.tenant, traffic)
+        _await_db_span(audit_sinks.tenant, tenant_trace, "redis", seconds=60)
+        tenant_spans: Final = _trace_spans(audit_sinks.tenant, tenant_trace, seconds=15)
+        _, all_tenant = recorded_spans(audit_sinks.tenant, start)
+        systems: Final = _db_systems(tenant_spans)
+        assert "redis" in systems, f"redis spans missing at tenant: {systems}"
+        assert "postgresql" not in _db_systems(all_tenant), (
+            f"postgresql spans reached tenant: {_db_systems(all_tenant)}"
+        )
+
+
 def test_bogus_excluded_services_env_fails_proxy_start_without_otel_callback(
     gateway: Gateway, otel_audit_config: AuditConfigWriter, tmp_path: Path
 ) -> None:

tests/unit/integrations/otel/test_otel_v2_config_baggage_parenting_guardrails.py

@@ -128,7 +128,7 @@ def test_excluded_services_env_is_validated_at_boot_when_enabled(monkeypatch):
     is_otel_v2_enabled.cache_clear()
     try:
         with pytest.raises(ValueError, match="'auth' is not a datastore service; allowed: postgres, redis"):
-            validate_otel_v2_excluded_services_env()
+            validate_otel_v2_excluded_services_env(None)
     finally:
         is_otel_v2_enabled.cache_clear()
 
@@ -140,7 +140,19 @@ def test_excluded_services_env_validation_accepts_datastore_names(monkeypatch):
     monkeypatch.setenv("LITELLM_OTEL_EXCLUDED_SERVICES", "redis, postgres")
     is_otel_v2_enabled.cache_clear()
     try:
-        validate_otel_v2_excluded_services_env()
+        validate_otel_v2_excluded_services_env(None)
+    finally:
+        is_otel_v2_enabled.cache_clear()
+
+
+def test_excluded_services_env_bad_value_is_inert_when_config_wins(monkeypatch):
+    from litellm.integrations.otel.model.config import is_otel_v2_enabled, validate_otel_v2_excluded_services_env
+
+    monkeypatch.setenv("LITELLM_OTEL_V2", "1")
+    monkeypatch.setenv("LITELLM_OTEL_EXCLUDED_SERVICES", "auth")
+    is_otel_v2_enabled.cache_clear()
+    try:
+        validate_otel_v2_excluded_services_env({"excluded_services": ["postgres"]})
     finally:
         is_otel_v2_enabled.cache_clear()