vllm-sr 0.3.0.dev20260916213254__tar.gz → 0.3.0.dev20260917022332__tar.gz

{vllm_sr-0.3.0.dev20260916213254 → vllm_sr-0.3.0.dev20260917022332}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vllm-sr
-Version: 0.3.0.dev20260916213254
+Version: 0.3.0.dev20260917022332
 Summary: vLLM Semantic Router - Intelligent routing for Mixture-of-Models
 Author: vLLM-SR Team
 License: Apache-2.0

{vllm_sr-0.3.0.dev20260916213254 → vllm_sr-0.3.0.dev20260917022332}/cli/templates/envoy.template.yaml

@@ -67,15 +67,9 @@ static_resources:
                   {% else %}
                   host_rewrite_literal: "{{ model.endpoints[0].host_authority }}"
                   {% endif %}
-                  {% if model.path_prefix %}
-                  # Prepend path prefix to /v1 requests (e.g., /compatible-mode + /v1/chat/completions = /compatible-mode/v1/chat/completions).
-                  # Match only the complete /v1 segment so route recomputation cannot rewrite /v1beta/... again.
-                  regex_rewrite:
-                    pattern:
-                      google_re2: {}
-                      regex: "^/v1([/?].*)?$"
-                    substitution: "{{ model.path_prefix }}\\1"
-                  {% endif %}
+                  # ext_proc has already resolved the complete provider path for
+                  # requests carrying x-selected-model, so this route must not
+                  # prefix it a second time.
               {% endfor %}
               # Default route (no x-selected-model header)
               - match:

{vllm_sr-0.3.0.dev20260916213254 → vllm_sr-0.3.0.dev20260917022332}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "vllm-sr"
-version = "0.3.0.dev20260916213254"
+version = "0.3.0.dev20260917022332"
 description = "vLLM Semantic Router - Intelligent routing for Mixture-of-Models"
 authors = [{name = "vLLM-SR Team"}]
 readme = "README.md"

{vllm_sr-0.3.0.dev20260916213254 → vllm_sr-0.3.0.dev20260917022332}/tests/test_catalog_provider_projection.py

@@ -108,7 +108,10 @@ routing:
     route = _model_route(rendered, "frontier")
     assert route["route"]["cluster"] == "model_frontier_cluster"
     assert route["route"]["host_rewrite_literal"] == "api.openai.com"
-    assert route["route"]["regex_rewrite"]["substitution"] == "/v1\\1"
+    assert "regex_rewrite" not in route["route"]
+    assert (
+        _default_route(rendered)["route"]["regex_rewrite"]["substitution"] == "/v1\\1"
+    )
     headers = {
         header["header"]["key"]: header["header"]["value"]
         for header in route.get("request_headers_to_add", [])
@@ -279,7 +282,10 @@ routing: {}
 
     route = _model_route(rendered, "frontier-responses")
     assert route["route"]["cluster"] == "model_frontier_2dresponses_cluster"
-    assert route["route"]["regex_rewrite"]["substitution"] == "/v1\\1"
+    assert "regex_rewrite" not in route["route"]
+    assert (
+        _default_route(rendered)["route"]["regex_rewrite"]["substitution"] == "/v1\\1"
+    )
     headers = {
         header["header"]["key"]: header["header"]["value"]
         for header in route.get("request_headers_to_add", [])
@@ -348,7 +354,11 @@ routing:
     )
     route = _model_route(rendered, "frontier")
     assert route["route"]["host_rewrite_literal"] == "gateway.example.test"
-    assert route["route"]["regex_rewrite"]["substitution"] == "/openai/v1\\1"
+    assert "regex_rewrite" not in route["route"]
+    assert (
+        _default_route(rendered)["route"]["regex_rewrite"]["substitution"]
+        == "/openai/v1\\1"
+    )
 
 
 def test_catalog_provider_projection_rejects_missing_model_mapping(tmp_path):
@@ -567,7 +577,11 @@ routing: {{}}
     }
     route = _model_route(rendered, "local-model")
     assert route["route"]["host_rewrite_literal"] == expected_authority
-    assert route["route"]["regex_rewrite"]["substitution"] == "/v1" + r"\1"
+    assert "regex_rewrite" not in route["route"]
+    assert (
+        _default_route(rendered)["route"]["regex_rewrite"]["substitution"]
+        == "/v1" + r"\1"
+    )
     assert ("transport_socket" in cluster) is expected_https
 
 

{vllm_sr-0.3.0.dev20260916213254 → vllm_sr-0.3.0.dev20260917022332}/tests/test_config_generator.py

@@ -248,12 +248,15 @@ routing:
     route_action = route["route"]
     assert route_action["auto_host_rewrite"] is True
     assert "host_rewrite_literal" not in route_action
-    assert route_action["regex_rewrite"]["pattern"]["regex"] == r"^/v1([/?].*)?$"
-    assert route_action["regex_rewrite"]["substitution"] == "/v1\\1"
+    assert "regex_rewrite" not in route_action
 
     default_route_action = _default_route(rendered)["route"]
     assert default_route_action["auto_host_rewrite"] is True
     assert "host_rewrite_literal" not in default_route_action
+    assert (
+        default_route_action["regex_rewrite"]["pattern"]["regex"] == r"^/v1([/?].*)?$"
+    )
+    assert default_route_action["regex_rewrite"]["substitution"] == "/v1\\1"
 
 
 def test_base_url_path_rewrite_is_idempotent(tmp_path, monkeypatch):
@@ -295,38 +298,34 @@ routing:
         router_api_host="localhost",
     )
 
-    for route in (_model_route(rendered, "gemini-model"), _default_route(rendered)):
-        rewrite = route["route"]["regex_rewrite"]
-        pattern = rewrite["pattern"]["regex"]
-        substitution = rewrite["substitution"]
-
-        assert pattern == r"^/v1([/?].*)?$"
-        for request_path, upstream_path in (
-            ("/v1", "/v1beta/openai"),
-            (
-                "/v1/chat/completions",
-                "/v1beta/openai/chat/completions",
-            ),
-            (
-                "/v1?api-version=test",
-                "/v1beta/openai?api-version=test",
-            ),
-        ):
-            rewritten_path = re.sub(pattern, substitution, request_path)
-            assert rewritten_path == upstream_path
-            assert re.sub(pattern, substitution, rewritten_path) == upstream_path
-
-
-@pytest.mark.skip(
-    reason=(
-        "TODO(issue-2885): fix root-cause rewrite idempotency for backend base "
-        "paths that still begin with the /v1 segment after rewriting."
-    )
-)
-def test_base_url_path_rewrite_idempotency_todo_for_v1_segment_prefix(
+    # ext_proc owns the complete path after model selection. The fallback route
+    # still rewrites the original ingress path when semantic processing is skipped.
+    assert "regex_rewrite" not in _model_route(rendered, "gemini-model")["route"]
+
+    rewrite = _default_route(rendered)["route"]["regex_rewrite"]
+    pattern = rewrite["pattern"]["regex"]
+    substitution = rewrite["substitution"]
+    assert pattern == r"^/v1([/?].*)?$"
+    for request_path, upstream_path in (
+        ("/v1", "/v1beta/openai"),
+        (
+            "/v1/chat/completions",
+            "/v1beta/openai/chat/completions",
+        ),
+        (
+            "/v1?api-version=test",
+            "/v1beta/openai?api-version=test",
+        ),
+    ):
+        rewritten_path = re.sub(pattern, substitution, request_path)
+        assert rewritten_path == upstream_path
+        assert re.sub(pattern, substitution, rewritten_path) == upstream_path
+
+
+def test_selected_model_route_does_not_rewrite_overlapping_provider_path(
     tmp_path, monkeypatch
 ):
-    """Document the known gap: /v1/chat -> /v1/provider/chat rewrites twice today."""
+    """Do not confuse an overlapping provider prefix with an ingress path."""
     rendered = _render_envoy_config(
         tmp_path,
         monkeypatch,
@@ -344,7 +343,7 @@ providers:
       provider_model_id: "provider-model"
       backend_refs:
         - name: "provider"
-          base_url: "https://api.example.com/v1/provider"
+          base_url: "https://api.example.com/v1/chat"
           provider: "openai"
           weight: 100
 routing:
@@ -365,15 +364,19 @@ routing:
         router_api_host="localhost",
     )
 
-    for route in (_model_route(rendered, "provider-model"), _default_route(rendered)):
-        rewrite = route["route"]["regex_rewrite"]
-        pattern = rewrite["pattern"]["regex"]
-        substitution = rewrite["substitution"]
-        upstream_path = "/v1/provider/chat/completions"
+    assert "regex_rewrite" not in _model_route(rendered, "provider-model")["route"]
 
-        rewritten_path = re.sub(pattern, substitution, "/v1/chat/completions")
-        assert rewritten_path == upstream_path
-        assert re.sub(pattern, substitution, rewritten_path) == upstream_path
+    rewrite = _default_route(rendered)["route"]["regex_rewrite"]
+    assert rewrite["pattern"]["regex"] == r"^/v1([/?].*)?$"
+    assert rewrite["substitution"] == "/v1/chat\\1"
+    assert (
+        re.sub(
+            rewrite["pattern"]["regex"],
+            rewrite["substitution"],
+            "/v1/chat/completions",
+        )
+        == "/v1/chat/chat/completions"
+    )
 
 
 def test_provider_reliability_renders_retry_outlier_and_least_request(
@@ -447,7 +450,7 @@ def test_backend_ref_domain_with_path_produces_correct_envoy_cluster_and_route(
 ):
     """Backend ref https://api.example.com/compatible-mode/v1 should produce
     address=api.example.com, port=443, host_authority=api.example.com (standard
-    port omitted), LOGICAL_DNS cluster, and regex_rewrite for path prefix."""
+    port omitted), LOGICAL_DNS cluster, and a fallback path-prefix rewrite."""
     rendered = _render_envoy_config(
         tmp_path,
         monkeypatch,
@@ -499,8 +502,15 @@ routing:
     route_action = route["route"]
     # standard port 443 → host_authority should omit port
     assert route_action["host_rewrite_literal"] == "api.example.com"
-    assert route_action["regex_rewrite"]["pattern"]["regex"] == r"^/v1([/?].*)?$"
-    assert route_action["regex_rewrite"]["substitution"] == "/compatible-mode/v1\\1"
+    assert "regex_rewrite" not in route_action
+    default_route_action = _default_route(rendered)["route"]
+    assert (
+        default_route_action["regex_rewrite"]["pattern"]["regex"] == r"^/v1([/?].*)?$"
+    )
+    assert (
+        default_route_action["regex_rewrite"]["substitution"]
+        == "/compatible-mode/v1\\1"
+    )
 
 
 def test_backend_ref_https_base_url_uses_tls_and_explicit_extra_headers(
@@ -568,7 +578,11 @@ routing:
     route = _model_route(rendered, "test-model")
     route_action = route["route"]
     assert route_action["host_rewrite_literal"] == "openrouter.ai"
-    assert route_action["regex_rewrite"]["substitution"] == "/api/v1\\1"
+    assert "regex_rewrite" not in route_action
+    assert (
+        _default_route(rendered)["route"]["regex_rewrite"]["substitution"]
+        == "/api/v1\\1"
+    )
 
     headers = {
         item["header"]["key"]: item["header"]["value"]

{vllm_sr-0.3.0.dev20260916213254 → vllm_sr-0.3.0.dev20260917022332}/vllm_sr.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vllm-sr
-Version: 0.3.0.dev20260916213254
+Version: 0.3.0.dev20260917022332
 Summary: vLLM Semantic Router - Intelligent routing for Mixture-of-Models
 Author: vLLM-SR Team
 License: Apache-2.0