aladac/tensors
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Turn server wrapper into transparent proxy with /reload

Browse Source 
Replace management endpoints (/start, /stop, /restart) with a transparent
reverse proxy and hot-reload architecture. The wrapper now sits in front of
sd-server, forwarding all requests and adding a /reload endpoint for model
swapping without restarting the wrapper itself.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Adam Ladachowski
2026-02-12 20:51:16 +00:00
parent b33fe120fa
commit 61e68fee15
6 changed files with 203 additions and 124 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/test_server.py
+89 -61
View File
@@ -1,13 +1,15 @@
"""Tests for tensors.server package (FastAPI sd-server manager)."""
"""Tests for tensors.server package (FastAPI sd-server proxy wrapper)."""
from __future__ import annotations
from unittest.mock import MagicMock, patch
from unittest.mock import AsyncMock, MagicMock, patch
import httpx
import pytest
from fastapi.testclient import TestClient
from tensors.server import create_app
from tensors.server.models import ServerConfig
from tensors.server.process import ProcessManager
@@ -22,7 +24,7 @@ def api() -> TestClient:
def _get_pm(api: TestClient) -> ProcessManager:
return api.app.state.pm # type: ignore[union-attr]
return api.app.state.pm # type: ignore[no-any-return, attr-defined]
class TestStatus:
@@ -37,7 +39,7 @@ class TestStatus:
mock_proc.poll.return_value = None
mock_proc.pid = 999
pm.proc = mock_proc
pm.config = {"model": "/m.safetensors", "port": 1234, "args": []}
pm.config = ServerConfig(model="/m.safetensors")
r = api.get("/status")
data = r.json()
assert data["running"] is True
@@ -54,69 +56,86 @@ class TestStatus:
assert data["exit_code"] == 1
class TestStart:
class TestReload:
@patch.object(ProcessManager, "wait_ready", new_callable=AsyncMock, return_value=True)
@patch("tensors.server.process.subprocess.Popen")
def test_start_success(self, mock_popen: MagicMock, api: TestClient) -> None:
def test_reload_swaps_model(self, mock_popen: MagicMock, mock_ready: AsyncMock, api: TestClient) -> None:
pm = _get_pm(api)
pm.config = ServerConfig(model="/old.gguf", port=5555, args=["--fa"])
mock_popen.return_value.pid = 42
mock_popen.return_value.poll.return_value = None
r = api.post("/start", json={"model": "/m.safetensors"})
assert r.status_code == 200
assert r.json()["started"] is True
assert r.json()["pid"] == 42
@patch("tensors.server.process.subprocess.Popen")
def test_start_already_running(self, mock_popen: MagicMock, api: TestClient) -> None:
pm = _get_pm(api)
mock_proc = MagicMock()
mock_proc.poll.return_value = None
pm.proc = mock_proc
r = api.post("/start", json={"model": "/m.safetensors"})
assert r.status_code == 409
class TestStop:
def test_stop_not_running(self, api: TestClient) -> None:
r = api.post("/stop")
assert r.status_code == 409
def test_stop_running(self, api: TestClient) -> None:
pm = _get_pm(api)
mock_proc = MagicMock()
mock_proc.poll.return_value = None
mock_proc.wait.return_value = 0
pm.proc = mock_proc
r = api.post("/stop")
assert r.status_code == 200
assert r.json()["stopped"] is True
mock_proc.send_signal.assert_called_once()
class TestRestart:
def test_restart_no_config_no_model(self, api: TestClient) -> None:
r = api.post("/restart", json={})
assert r.status_code == 400
@patch("tensors.server.process.subprocess.Popen")
def test_restart_with_new_model(self, mock_popen: MagicMock, api: TestClient) -> None:
mock_popen.return_value.pid = 100
mock_popen.return_value.poll.return_value = None
pm = _get_pm(api)
pm.config = {"model": "/old.safetensors", "port": 1234, "args": []}
r = api.post("/restart", json={"model": "/new.safetensors"})
r = api.post("/reload", json={"model": "/new.gguf"})
assert r.status_code == 200
data = r.json()
assert data["restarted"] is True
assert "/new.safetensors" in str(data["cmd"])
assert data["ok"] is True
assert data["model"] == "/new.gguf"
assert data["pid"] == 42
# Verify new config preserved port and args from previous config
assert pm.config is not None
assert pm.config.port == 5555
assert pm.config.args == ["--fa"]
assert pm.config.model == "/new.gguf"
@patch.object(ProcessManager, "wait_ready", new_callable=AsyncMock, return_value=False)
@patch("tensors.server.process.subprocess.Popen")
def test_restart_keeps_previous_config(self, mock_popen: MagicMock, api: TestClient) -> None:
mock_popen.return_value.pid = 101
mock_popen.return_value.poll.return_value = None
def test_reload_fails_when_not_ready(self, mock_popen: MagicMock, mock_ready: AsyncMock, api: TestClient) -> None:
pm = _get_pm(api)
pm.config = {"model": "/m.safetensors", "port": 5555, "args": ["--fa"]}
r = api.post("/restart", json={})
pm.config = ServerConfig(model="/old.gguf")
mock_popen.return_value.pid = 43
mock_popen.return_value.poll.return_value = None
r = api.post("/reload", json={"model": "/bad.gguf"})
assert r.status_code == 503
assert "failed" in r.json()["error"]
def test_reload_requires_model(self, api: TestClient) -> None:
r = api.post("/reload", json={})
assert r.status_code == 422
class TestProxy:
def test_proxy_503_when_not_running(self, api: TestClient) -> None:
r = api.get("/v1/models")
assert r.status_code == 503
assert "not running" in r.json()["error"]
def test_proxy_forwards_request(self, api: TestClient) -> None:
pm = _get_pm(api)
mock_proc = MagicMock()
mock_proc.poll.return_value = None
mock_proc.pid = 100
pm.proc = mock_proc
pm.config = ServerConfig(model="/m.gguf", port=1234)
upstream_response = httpx.Response(
200,
json={"data": [{"id": "model-1"}]},
headers={"content-type": "application/json"},
)
mock_client = AsyncMock()
mock_client.request.return_value = upstream_response
api.app.state.client = mock_client # type: ignore[attr-defined]
r = api.get("/v1/models")
assert r.status_code == 200
assert "5555" in str(r.json()["cmd"])
assert r.json() == {"data": [{"id": "model-1"}]}
mock_client.request.assert_called_once()
def test_proxy_forwards_post_with_body(self, api: TestClient) -> None:
pm = _get_pm(api)
mock_proc = MagicMock()
mock_proc.poll.return_value = None
mock_proc.pid = 100
pm.proc = mock_proc
pm.config = ServerConfig(model="/m.gguf", port=1234)
upstream_response = httpx.Response(200, json={"ok": True})
mock_client = AsyncMock()
mock_client.request.return_value = upstream_response
api.app.state.client = mock_client # type: ignore[attr-defined]
r = api.post("/v1/chat/completions", json={"prompt": "hello"})
assert r.status_code == 200
mock_client.request.assert_called_once()
class TestProcessManager:
@@ -124,18 +143,27 @@ class TestProcessManager:
assert pm.status() == {"running": False}
def test_build_cmd(self, pm: ProcessManager) -> None:
config = {"model": "/m.gguf", "port": 1234, "args": ["--fa"]}
cmd = pm.build_cmd(config)
pm.config = ServerConfig(model="/m.gguf", port=1234, args=["--fa"])
cmd = pm.build_cmd()
assert "/m.gguf" in cmd
assert "--fa" in cmd
assert "1234" in cmd
def test_build_cmd_no_config(self, pm: ProcessManager) -> None:
with pytest.raises(RuntimeError, match="No config"):
pm.build_cmd()
@patch("tensors.server.process.subprocess.Popen")
def test_start_and_stop(self, mock_popen: MagicMock, pm: ProcessManager) -> None:
mock_popen.return_value.pid = 77
mock_popen.return_value.poll.return_value = None
mock_popen.return_value.wait.return_value = 0
pm.start({"model": "/m.gguf", "port": 1234, "args": []})
pm.start(ServerConfig(model="/m.gguf"))
assert pm.proc is not None
assert pm.stop() is True
assert pm.proc is None
def test_server_config_defaults(self) -> None:
cfg = ServerConfig(model="/m.gguf")
assert cfg.port == 1234
assert cfg.args == []