Files
turnstone/tests/test_admin_model_registry_refresh.py
T
Patrick Buckley 32fd8f29c7 feat(providers): api_surface toggle + mistral medium reasoning fix (#469)
* feat(providers): api_surface toggle + mistral medium reasoning fix

Mistral medium open-weights served by vLLM expects reasoning_effort via
the Responses API (`reasoning.effort`), not as a `chat_template_kwargs`
entry on Chat Completions.  The session was unconditionally injecting
`{"reasoning_effort": ...}` into `chat_template_kwargs` for every
openai-compatible request, which corrupted the prompt rendering for any
backend whose chat template didn't consume that key (Mistral medium,
Mistral cloud, Groq, OpenRouter).

Changes:
- Add `api_surface` ("chat" | "responses") to `ModelConfig.server_compat`
  and thread it through `create_provider` / `model_registry.get_provider`.
  `openai-compatible` defaults to Chat Completions; operators can flip
  individual aliases to Responses for endpoints that support it.
- New `vllm-mistral-medium` profile that pre-fills api_surface=responses
  on Detect for known Mistral medium model ids.
- Drop the unconditional `reasoning_effort` injection into
  `chat_template_kwargs`.  Operators running gpt-oss-style local
  templates that consume `reasoning_effort` from the chat template now
  opt in via `server_compat.extra_body.chat_template_kwargs`.
- New "API Surface" select in the Models admin tab; allowlist-validated
  server-side at create/update time; pre-filled by Detect via the
  profile suggestion.
- Evict the cached provider singleton in `ModelRegistry.reload()` when
  api_surface changes (previously only cfg.provider triggered eviction).
- Fix `_run_agent` fallback path to inherit the session's primary alias
  for capability and server_compat resolution; previously the fallback
  passed `alias=None`, which silently dropped per-model caps on the
  agent path.

Tests: 5117 passed (-m "not live"); ruff + mypy clean.

* fix(providers): don't auto-suggest Responses for Mistral medium

vLLM's Responses API surface for Mistral medium open-weights doesn't
wire up the Mistral tool-call parser as of vLLM 0.x — tool calls leak
into the response as ``[TOOL_CALLS]<name>{...}`` text instead of
structured tool_calls.  Chat Completions on the same engine handles
tools cleanly via ``--tool-call-parser mistral``, and reasoning can be
turned on via the vLLM CLI ``--reasoning-parser`` flag.

Drop the auto-suggest mapping so Detect falls back to the generic
``vllm`` profile.  Keep the ``vllm-mistral-medium`` profile definition
in place so an operator who specifically wants per-request effort and
accepts the tool-calling limitation can still pick "Responses API"
manually in the admin UI.

* fix(providers): address Copilot review on PR #469

- providers/__init__.py: drop the redundant *_responses_provider /
  *_chat_provider names; have create_provider use _openai_provider and
  _openai_compat_provider directly so they're not flagged as unused
  globals.
- console/server.py: tighten _validate_api_surface to a strict equality
  match against the canonical {"chat", "responses"} set.  The previous
  strip().lower() membership check accepted ' Responses '/'CHAT' but
  stored the raw string verbatim, which then failed to round-trip
  through the admin <select>.
- console/static/admin.js: gate the entire server_compat block (server
  type, api_surface, extra_body) on provider == "openai-compatible" at
  save time so toggling provider away can't leave a stale hidden surface
  selection in the persisted capabilities JSON.
- tests/test_session.py: splat the bad kwarg via **dict so CodeQL no
  longer flags the call as a wrong-name keyword (the point of the test
  is the runtime contract, not the static type).
- tests/test_admin_model_registry_refresh.py: add endpoint-level tests
  for the api_surface validation on both create and update — covers the
  bogus-value rejection, non-canonical-string rejection, and the happy
  path persisting through to the refreshed registry.
2026-05-03 13:37:50 -07:00

482 lines
19 KiB
Python

"""Console-side coord_registry auto-refresh on model-definition CRUD + reload.
The console builds ``app.state.coord_registry`` once at lifespan startup
and the coordinator session factory closes over that exact instance.
Without these refresh hooks, an admin who edits a model definition
through the UI sees the DB change immediately but coordinator sessions
keep calling the prior model name — the on-disk truth diverges from the
in-process registry until the console is restarted.
These tests cover both the helper (``_refresh_coord_registry``)
and the four wired endpoints (create / update / delete / explicit reload)
to lock in:
- in-place mutation: ``coord_registry`` object identity is preserved
across refreshes (factory closure must not be invalidated);
- failure isolation: a load or reload failure leaves the existing
registry intact rather than tearing down a working coordinator;
- no-op safety: the helper short-circuits when ``coord_registry`` is
``None`` so a coord-less console (no model rows at boot) doesn't
500 on routine model-definition CRUD.
"""
from __future__ import annotations
from typing import Any
from unittest.mock import MagicMock
import pytest
from starlette.applications import Starlette
from starlette.middleware import Middleware
from starlette.routing import Route
from starlette.testclient import TestClient
from tests._coord_test_helpers import _AuthMiddleware
from turnstone.console.server import (
_refresh_coord_registry,
admin_create_model_definition,
admin_delete_model_definition,
admin_model_reload,
admin_update_model_definition,
)
from turnstone.core.model_registry import ModelConfig, ModelRegistry
from turnstone.core.storage._sqlite import SQLiteBackend
# ---------------------------------------------------------------------------
# Fixtures
# ---------------------------------------------------------------------------
@pytest.fixture
def storage(tmp_path: Any) -> SQLiteBackend:
return SQLiteBackend(str(tmp_path / "models.db"))
def _seed_model_def(
storage: SQLiteBackend,
*,
definition_id: str,
alias: str,
model: str,
base_url: str = "http://localhost:8000/v1",
enabled: bool = True,
) -> None:
"""Insert a model definition row directly via the storage API."""
storage.create_model_definition(
definition_id=definition_id,
alias=alias,
model=model,
provider="openai-compatible",
base_url=base_url,
api_key="sk-test",
context_window=8192,
capabilities="{}",
enabled=enabled,
created_by="admin",
)
def _make_config(alias: str, model: str) -> ModelConfig:
return ModelConfig(
alias=alias,
base_url="http://localhost:8000/v1",
api_key="sk-test",
model=model,
context_window=8192,
provider="openai-compatible",
source="db",
)
def _make_registry(
*,
alias: str = "local",
model: str = "old-model",
extras: dict[str, str] | None = None,
) -> ModelRegistry:
"""Build a real ModelRegistry seeded with ``alias`` (the default) plus
any ``extras`` (alias → model). ``ModelRegistry.__init__`` rejects an
empty model dict so tests that exercise the helper need at least one
entry; pass ``extras`` for multi-alias scenarios (e.g. delete-by-alias).
"""
configs = {alias: _make_config(alias, model)}
for extra_alias, extra_model in (extras or {}).items():
configs[extra_alias] = _make_config(extra_alias, extra_model)
return ModelRegistry(configs, default=alias)
class _AppState:
"""Shim mirroring Starlette's ``app.state`` for direct helper tests."""
coord_registry: ModelRegistry | None = None
# ---------------------------------------------------------------------------
# Helper-level tests — ``_refresh_coord_registry`` semantics
# ---------------------------------------------------------------------------
def test_helper_rebuilds_registry_from_db(storage: SQLiteBackend) -> None:
"""Helper pulls the latest DB rows into the existing registry."""
_seed_model_def(storage, definition_id="m1", alias="local", model="new-model")
state = _AppState()
state.coord_registry = _make_registry(alias="local", model="old-model")
_refresh_coord_registry(state, storage)
assert state.coord_registry is not None
assert state.coord_registry.get_config("local").model == "new-model"
def test_helper_preserves_object_identity(storage: SQLiteBackend) -> None:
"""The factory closes over the registry object — refresh must mutate
in place rather than swap the attribute."""
_seed_model_def(storage, definition_id="m1", alias="local", model="new-model")
state = _AppState()
state.coord_registry = _make_registry()
before = id(state.coord_registry)
_refresh_coord_registry(state, storage)
assert id(state.coord_registry) == before
def test_helper_noop_when_coord_registry_none(storage: SQLiteBackend) -> None:
"""Console boot with no model rows leaves coord_registry = None.
The helper must not 500 in that state — CRUD that lands the FIRST
row would otherwise fail before the operator can recover."""
_seed_model_def(storage, definition_id="m1", alias="local", model="m")
state = _AppState()
state.coord_registry = None
_refresh_coord_registry(state, storage) # must not raise
assert state.coord_registry is None
def test_helper_preserves_registry_when_load_fails(
storage: SQLiteBackend, monkeypatch: pytest.MonkeyPatch
) -> None:
"""An unexpected error from ``load_model_registry`` (e.g. config.toml
parse failure, programming bug) must not tear down a working
registry — log + leave the existing instance intact."""
state = _AppState()
state.coord_registry = _make_registry(alias="local", model="old-model")
def _boom(**_kw: Any) -> ModelRegistry:
raise RuntimeError("simulated loader failure")
monkeypatch.setattr("turnstone.core.model_registry.load_model_registry", _boom)
_refresh_coord_registry(state, storage)
assert state.coord_registry is not None
assert state.coord_registry.get_config("local").model == "old-model"
def test_helper_preserves_registry_when_strict_load_fails(
storage: SQLiteBackend, monkeypatch: pytest.MonkeyPatch
) -> None:
"""``load_model_registry`` normally swallows storage read errors and
would return a config.toml-only registry on a transient DB outage —
applying that via ``reload()`` would silently drop every DB-sourced
alias. The helper passes ``strict=True`` so the loader re-raises
instead, the helper's outer except catches it, and the existing
registry survives intact."""
_seed_model_def(storage, definition_id="m1", alias="local", model="db-model")
state = _AppState()
state.coord_registry = _make_registry(alias="local", model="db-model")
def _broken(**_kw: Any) -> Any:
raise RuntimeError("simulated transient DB outage")
monkeypatch.setattr(storage, "list_model_definitions", _broken)
_refresh_coord_registry(state, storage)
assert state.coord_registry is not None
# Existing registry untouched — strict=True surfaced the storage
# error to the helper before the loader's silent fallback could
# produce a truncated registry for reload().
assert state.coord_registry.get_config("local").model == "db-model"
def test_helper_preserves_registry_when_no_enabled_rows(storage: SQLiteBackend) -> None:
"""All rows disabled/deleted: ModelRegistry.__init__ rejects an empty
model dict (raises ValueError). Helper must catch and preserve the
existing registry so coord stays usable while admin restores rows."""
_seed_model_def(storage, definition_id="m1", alias="local", model="m", enabled=False)
state = _AppState()
state.coord_registry = _make_registry(alias="local", model="cached-model")
_refresh_coord_registry(state, storage)
assert state.coord_registry is not None
assert state.coord_registry.get_config("local").model == "cached-model"
def test_helper_preserves_registry_on_reload_validation_error(
storage: SQLiteBackend, monkeypatch: pytest.MonkeyPatch
) -> None:
"""A reload that raises mid-mutation (e.g. validation guard) must
leave the existing registry instance functional."""
_seed_model_def(storage, definition_id="m1", alias="local", model="new-model")
state = _AppState()
state.coord_registry = _make_registry(alias="local", model="old-model")
def _broken_reload(*_a: Any, **_kw: Any) -> None:
raise ValueError("simulated reload validation failure")
monkeypatch.setattr(state.coord_registry, "reload", _broken_reload)
_refresh_coord_registry(state, storage)
# Existing registry still reachable; the broken reload was a no-op
# at the public-facing level.
assert state.coord_registry is not None
assert state.coord_registry.get_config("local").model == "old-model"
# ---------------------------------------------------------------------------
# Endpoint-level integration tests — verify wiring
# ---------------------------------------------------------------------------
def _make_client(storage: SQLiteBackend, registry: ModelRegistry | None) -> TestClient:
"""Build a TestClient wired to the four model-definition endpoints.
Uses the shared header-driven ``_AuthMiddleware`` from
``tests/_coord_test_helpers``; default headers below grant
``admin.models`` permission so the endpoint gate passes.
"""
app = Starlette(
routes=[
Route(
"/v1/api/admin/model-definitions",
admin_create_model_definition,
methods=["POST"],
),
Route(
"/v1/api/admin/model-definitions/reload",
admin_model_reload,
methods=["POST"],
),
Route(
"/v1/api/admin/model-definitions/{definition_id}",
admin_update_model_definition,
methods=["PUT"],
),
Route(
"/v1/api/admin/model-definitions/{definition_id}",
admin_delete_model_definition,
methods=["DELETE"],
),
],
middleware=[Middleware(_AuthMiddleware)],
)
app.state.auth_storage = storage
app.state.coord_registry = registry
# Reload endpoint also touches these — stub them so the test focuses
# on the registry-refresh behaviour without dragging in a full
# collector / proxy_client wiring.
app.state.collector = MagicMock()
app.state.collector.get_all_nodes.return_value = []
app.state.proxy_client = MagicMock()
app.state.config_store = MagicMock()
client = TestClient(app)
client.headers.update({"X-Test-User": "admin", "X-Test-Perms": "admin.models"})
return client
def test_create_endpoint_refreshes_registry(storage: SQLiteBackend) -> None:
"""POST /api/admin/model-definitions bumps the in-process registry
so newly-spawned coord sessions see the new alias immediately."""
# Pre-existing alias (registry needs at least one row)
_seed_model_def(storage, definition_id="m1", alias="local", model="m")
registry = _make_registry(alias="local", model="m")
client = _make_client(storage, registry)
resp = client.post(
"/v1/api/admin/model-definitions",
json={
"alias": "fast",
"model": "fast-model",
"provider": "openai-compatible",
"base_url": "http://localhost:9000/v1",
"api_key": "sk-x",
"context_window": 4096,
},
)
assert resp.status_code == 200, resp.text
assert registry.has_alias("fast")
assert registry.get_config("fast").model == "fast-model"
def test_update_endpoint_refreshes_registry(storage: SQLiteBackend) -> None:
"""PUT swaps the underlying model name behind a stable alias — the
user's reported regression."""
_seed_model_def(storage, definition_id="m1", alias="local", model="old-model")
registry = _make_registry(alias="local", model="old-model")
client = _make_client(storage, registry)
resp = client.put(
"/v1/api/admin/model-definitions/m1",
json={"model": "new-model"},
)
assert resp.status_code == 200, resp.text
assert registry.get_config("local").model == "new-model"
def test_update_endpoint_skips_refresh_on_empty_body(
storage: SQLiteBackend, monkeypatch: pytest.MonkeyPatch
) -> None:
"""An empty PUT body must skip the registry refresh — the
``if updates:`` gate exists because ``load_model_registry`` is
non-trivial and a no-op refresh on every PUT would burn cycles
rebuilding state that hasn't changed. Spy on the helper to lock
the gate down: a regression that drops the conditional would
register a call here and trip the assertion.
"""
from turnstone.console import server as server_module
_seed_model_def(storage, definition_id="m1", alias="local", model="locked-in")
registry = _make_registry(alias="local", model="locked-in")
client = _make_client(storage, registry)
calls: list[tuple[Any, Any]] = []
def _spy(app_state: Any, storage: Any) -> None:
calls.append((app_state, storage))
monkeypatch.setattr(server_module, "_refresh_coord_registry", _spy)
resp = client.put("/v1/api/admin/model-definitions/m1", json={})
assert resp.status_code == 200, resp.text
assert calls == [] # gate held: empty body did not trigger a refresh
def test_create_rejects_invalid_api_surface(storage: SQLiteBackend) -> None:
"""POST with a bogus server_compat.api_surface returns 400 rather than
persisting a value that would make get_provider() raise on every later
ChatSession init for the alias."""
_seed_model_def(storage, definition_id="m1", alias="local", model="m")
registry = _make_registry(alias="local", model="m")
client = _make_client(storage, registry)
resp = client.post(
"/v1/api/admin/model-definitions",
json={
"alias": "bad",
"model": "x",
"provider": "openai-compatible",
"base_url": "http://localhost:9000/v1",
"api_key": "sk-x",
"capabilities": {"server_compat": {"api_surface": "BOGUS"}},
},
)
assert resp.status_code == 400, resp.text
assert "api_surface" in resp.json()["error"]
# And the alias is not persisted
assert not registry.has_alias("bad")
def test_create_rejects_non_canonical_api_surface(storage: SQLiteBackend) -> None:
"""Strict validation: ' Responses ' / 'CHAT' don't round-trip through the
admin <select>, so they're rejected even though they'd survive a
case-insensitive membership check."""
_seed_model_def(storage, definition_id="m1", alias="local", model="m")
registry = _make_registry(alias="local", model="m")
client = _make_client(storage, registry)
for bad in (" responses ", "RESPONSES", "Chat"):
resp = client.post(
"/v1/api/admin/model-definitions",
json={
"alias": "noncanon",
"model": "x",
"provider": "openai-compatible",
"base_url": "http://localhost:9000/v1",
"api_key": "sk-x",
"capabilities": {"server_compat": {"api_surface": bad}},
},
)
assert resp.status_code == 400, f"{bad!r}: {resp.text}"
def test_create_accepts_valid_api_surface(storage: SQLiteBackend) -> None:
"""Canonical 'chat' / 'responses' / unset are all accepted and persisted."""
_seed_model_def(storage, definition_id="m1", alias="local", model="m")
registry = _make_registry(alias="local", model="m")
client = _make_client(storage, registry)
resp = client.post(
"/v1/api/admin/model-definitions",
json={
"alias": "responses-alias",
"model": "x",
"provider": "openai-compatible",
"base_url": "http://localhost:9000/v1",
"api_key": "sk-x",
"capabilities": {"server_compat": {"api_surface": "responses"}},
},
)
assert resp.status_code == 200, resp.text
assert registry.has_alias("responses-alias")
def test_update_rejects_invalid_api_surface(storage: SQLiteBackend) -> None:
"""PUT path also gates the validation, so an admin can't smuggle a bad
value into an existing alias."""
_seed_model_def(storage, definition_id="m1", alias="local", model="m")
registry = _make_registry(alias="local", model="m")
client = _make_client(storage, registry)
resp = client.put(
"/v1/api/admin/model-definitions/m1",
json={"capabilities": {"server_compat": {"api_surface": "junk"}}},
)
assert resp.status_code == 400, resp.text
assert "api_surface" in resp.json()["error"]
def test_delete_endpoint_refreshes_registry(storage: SQLiteBackend) -> None:
"""DELETE drops the alias from the in-process registry too — a
coord session that tried to resolve the deleted alias would
otherwise hit a stale cached client."""
_seed_model_def(storage, definition_id="m1", alias="local", model="m")
_seed_model_def(storage, definition_id="m2", alias="extra", model="x")
registry = _make_registry(alias="local", model="m", extras={"extra": "x"})
client = _make_client(storage, registry)
resp = client.delete("/v1/api/admin/model-definitions/m2")
assert resp.status_code == 200, resp.text
assert not registry.has_alias("extra")
assert registry.has_alias("local") # default alias unaffected
def test_reload_endpoint_refreshes_registry(
storage: SQLiteBackend, monkeypatch: pytest.MonkeyPatch
) -> None:
"""The explicit reload button must refresh the console's own
registry — until this PR it only fanned out to nodes."""
_seed_model_def(storage, definition_id="m1", alias="local", model="initial")
registry = _make_registry(alias="local", model="initial")
client = _make_client(storage, registry)
# Bypass the CRUD endpoints to mimic an out-of-band DB change (e.g.
# an operator psql session) and verify the explicit reload path
# still pulls the change in.
storage.update_model_definition("m1", model="reloaded-model")
# Stub the async cluster fan-out helpers — they require a fully-wired
# collector / proxy_client which is orthogonal to the helper under test.
async def _noop_publish(_request: Any) -> None:
return None
async def _noop_notify(_request: Any) -> dict[str, Any]:
return {}
monkeypatch.setattr("turnstone.console.server._publish_config_change", _noop_publish)
monkeypatch.setattr("turnstone.console.server._notify_nodes_model_reload", _noop_notify)
resp = client.post("/v1/api/admin/model-definitions/reload")
assert resp.status_code == 200, resp.text
assert registry.get_config("local").model == "reloaded-model"