diff --git a/gateway/routes/allowlist.py b/gateway/routes/allowlist.py index 92b73867e67..3733072a948 100644 --- a/gateway/routes/allowlist.py +++ b/gateway/routes/allowlist.py @@ -3,7 +3,8 @@ The gateway exposes the LLM data-plane surface: chat/completions, embeddings, audio, batches, files, fine-tuning, rerank, ocr, rag, video, search, image, responses, vector stores, passthrough providers, realtime websockets, MCP -tool-call endpoints, and operational endpoints (/health, /metrics). +tool-call endpoints, and operational endpoints (/health, /metrics, and the +/debug/memory/summary read of the serving worker's RSS). Any path not listed here is dropped from the gateway process so management/UI endpoints don't ride on the same pods. @@ -121,6 +122,7 @@ GATEWAY_EXACT_PATHS: frozenset[str] = frozenset( "/docs/oauth2-redirect", "/redoc", "/test", + "/debug/memory/summary", } ) diff --git a/tests/test_litellm/proxy/test_component_allowlists.py b/tests/test_litellm/proxy/test_component_allowlists.py index 0fdb43d60da..3073908fa54 100644 --- a/tests/test_litellm/proxy/test_component_allowlists.py +++ b/tests/test_litellm/proxy/test_component_allowlists.py @@ -196,6 +196,20 @@ def test_gateway_drops_ui_and_swagger_mounts(): f"Mount {path} must not be served by the gateway" +def test_gateway_keeps_memory_summary_and_trims_the_other_debug_routes(): + """The gateway serves /debug/memory/summary, since the RSS that matters is the + serving worker's and the memory regression e2e test reads it on every gateway + replica; the heavier and mutating /debug/memory routes stay on the backend.""" + debug_memory_routes = { + getattr(r, "path"): r for r in app.router.routes if str(getattr(r, "path", "")).startswith("/debug/memory/") + } + assert {"/debug/memory/summary", "/debug/memory/details", "/debug/memory/gc/configure"} <= set(debug_memory_routes) + assert _is_gateway_route(debug_memory_routes["/debug/memory/summary"]), \ + "/debug/memory/summary must survive the gateway route trim" + for path in ("/debug/memory/details", "/debug/memory/gc/configure"): + assert not _is_gateway_route(debug_memory_routes[path]), f"{path} must not be served by the gateway" + + def test_every_app_mount_is_assigned_to_a_component(): """Every Mount on the proxy app must be consciously assigned to a component.