Spaces:
Runtime error
Runtime error
Altamira Builder commited on
Commit ·
6ae6a09
1
Parent(s): ff80184
minimal FastAPI Docker test v2 - match SpacesExamples pattern
Browse files- Dockerfile +6 -4
- app.py +11 -6
- prestart.sh +0 -2
- router.py +0 -94
- worker.py +0 -94
Dockerfile
CHANGED
|
@@ -1,9 +1,11 @@
|
|
| 1 |
-
FROM python:3.
|
| 2 |
|
| 3 |
-
|
| 4 |
|
| 5 |
-
COPY
|
| 6 |
|
| 7 |
-
|
|
|
|
|
|
|
| 8 |
|
| 9 |
CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]
|
|
|
|
| 1 |
+
FROM python:3.9
|
| 2 |
|
| 3 |
+
WORKDIR /code
|
| 4 |
|
| 5 |
+
COPY ./requirements.txt /code/requirements.txt
|
| 6 |
|
| 7 |
+
RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
|
| 8 |
+
|
| 9 |
+
COPY . .
|
| 10 |
|
| 11 |
CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]
|
app.py
CHANGED
|
@@ -1,8 +1,13 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
|
| 4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
|
|
|
| 1 |
+
import http.server
|
| 2 |
+
import json
|
| 3 |
|
| 4 |
+
class Handler(http.server.BaseHTTPRequestHandler):
|
| 5 |
+
def do_GET(self):
|
| 6 |
+
self.send_response(200)
|
| 7 |
+
self.send_header("Content-Type", "application/json")
|
| 8 |
+
self.end_headers()
|
| 9 |
+
self.wfile.write(json.dumps({"status": "ok"}).encode())
|
| 10 |
|
| 11 |
+
if __name__ == "__main__":
|
| 12 |
+
server = http.server.HTTPServer(("0.0.0.0", 7860), Handler)
|
| 13 |
+
server.serve_forever()
|
prestart.sh
DELETED
|
@@ -1,2 +0,0 @@
|
|
| 1 |
-
#!/bin/bash
|
| 2 |
-
pip install huggingface_hub==0.25.2 --upgrade --quiet
|
|
|
|
|
|
|
|
|
router.py
DELETED
|
@@ -1,94 +0,0 @@
|
|
| 1 |
-
import asyncio
|
| 2 |
-
import time
|
| 3 |
-
from collections import deque
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
class PredictiveContextFilter:
|
| 7 |
-
def __init__(self, capacity: int = 4096, threshold: float = 0.8):
|
| 8 |
-
self.capacity = capacity
|
| 9 |
-
self.threshold = threshold
|
| 10 |
-
self.window = deque(maxlen=100)
|
| 11 |
-
|
| 12 |
-
def estimate_tokens(self, text: str) -> int:
|
| 13 |
-
return len(text) // 4
|
| 14 |
-
|
| 15 |
-
def should_compact(self, text: str) -> bool:
|
| 16 |
-
ratio = self.estimate_tokens(text) / self.capacity
|
| 17 |
-
return ratio >= self.threshold
|
| 18 |
-
|
| 19 |
-
def compact(self, text: str, target_ratio: float = 0.5) -> str:
|
| 20 |
-
if not self.should_compact(text):
|
| 21 |
-
return text
|
| 22 |
-
target_len = int(self.capacity * target_ratio * 4)
|
| 23 |
-
if len(text) <= target_len:
|
| 24 |
-
return text
|
| 25 |
-
half = target_len // 2
|
| 26 |
-
return text[:half] + "\n... [compacted] ...\n" + text[-half:]
|
| 27 |
-
|
| 28 |
-
async def monitor(self, text: str) -> str:
|
| 29 |
-
loop_count = 0
|
| 30 |
-
while self.should_compact(text) and loop_count < 3:
|
| 31 |
-
text = self.compact(text)
|
| 32 |
-
loop_count += 1
|
| 33 |
-
return text
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
class CircuitBreakerState:
|
| 37 |
-
CLOSED = "closed"
|
| 38 |
-
OPEN = "open"
|
| 39 |
-
HALF_OPEN = "half_open"
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
class InferenceRouterCircuitBreaker:
|
| 43 |
-
def __init__(self, failure_threshold: int = 5, recovery_timeout: float = 30.0):
|
| 44 |
-
self.failure_threshold = failure_threshold
|
| 45 |
-
self.recovery_timeout = recovery_timeout
|
| 46 |
-
self.state = CircuitBreakerState.CLOSED
|
| 47 |
-
self.failure_count = 0
|
| 48 |
-
self.last_failure_time = 0.0
|
| 49 |
-
|
| 50 |
-
def record_success(self):
|
| 51 |
-
self.failure_count = 0
|
| 52 |
-
if self.state == CircuitBreakerState.HALF_OPEN:
|
| 53 |
-
self.state = CircuitBreakerState.CLOSED
|
| 54 |
-
|
| 55 |
-
def record_failure(self):
|
| 56 |
-
self.failure_count += 1
|
| 57 |
-
self.last_failure_time = time.time()
|
| 58 |
-
if self.failure_count >= self.failure_threshold:
|
| 59 |
-
self.state = CircuitBreakerState.OPEN
|
| 60 |
-
|
| 61 |
-
async def call(self, coro_factory, fallback=None):
|
| 62 |
-
if self.state == CircuitBreakerState.OPEN:
|
| 63 |
-
if time.time() - self.last_failure_time >= self.recovery_timeout:
|
| 64 |
-
self.state = CircuitBreakerState.HALF_OPEN
|
| 65 |
-
elif fallback is not None:
|
| 66 |
-
return fallback
|
| 67 |
-
else:
|
| 68 |
-
raise RuntimeError("Circuit breaker is OPEN")
|
| 69 |
-
try:
|
| 70 |
-
result = await coro_factory()
|
| 71 |
-
self.record_success()
|
| 72 |
-
return result
|
| 73 |
-
except Exception as e:
|
| 74 |
-
self.record_failure()
|
| 75 |
-
if fallback is not None:
|
| 76 |
-
return fallback
|
| 77 |
-
raise e
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
class ParallelEngine:
|
| 81 |
-
def __init__(self, max_concurrency: int = 4):
|
| 82 |
-
self.semaphore = asyncio.Semaphore(max_concurrency)
|
| 83 |
-
self.filter = PredictiveContextFilter()
|
| 84 |
-
self.breaker = InferenceRouterCircuitBreaker()
|
| 85 |
-
|
| 86 |
-
async def run_with_filter(self, text: str, coro_factory) -> str:
|
| 87 |
-
compacted = await self.filter.monitor(text)
|
| 88 |
-
return await self.breaker.call(lambda: coro_factory(compacted))
|
| 89 |
-
|
| 90 |
-
async def run_parallel(self, tasks: list):
|
| 91 |
-
async def bounded(task):
|
| 92 |
-
async with self.semaphore:
|
| 93 |
-
return await task
|
| 94 |
-
return await asyncio.gather(*(bounded(t) for t in tasks))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
worker.py
DELETED
|
@@ -1,94 +0,0 @@
|
|
| 1 |
-
import asyncio
|
| 2 |
-
import json
|
| 3 |
-
import shutil
|
| 4 |
-
import subprocess
|
| 5 |
-
from pathlib import Path
|
| 6 |
-
|
| 7 |
-
from fastapi import FastAPI, HTTPException, Request
|
| 8 |
-
from huggingface_hub import HfApi
|
| 9 |
-
|
| 10 |
-
app = FastAPI(title="Altamira Worker")
|
| 11 |
-
|
| 12 |
-
CACHE_DIR = Path("/tmp/altamira-cache")
|
| 13 |
-
WORKSPACE_DIR = Path("/tmp/altamira-workspace")
|
| 14 |
-
STAGE_DIRS = [CACHE_DIR / "stage_0", CACHE_DIR / "stage_1", CACHE_DIR / "stage_2"]
|
| 15 |
-
|
| 16 |
-
api = HfApi()
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
def _ensure_dirs():
|
| 20 |
-
for d in [CACHE_DIR, WORKSPACE_DIR] + STAGE_DIRS:
|
| 21 |
-
d.mkdir(parents=True, exist_ok=True)
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
def _rotate_stages():
|
| 25 |
-
for i in range(len(STAGE_DIRS) - 1, 0, -1):
|
| 26 |
-
src = STAGE_DIRS[i - 1]
|
| 27 |
-
dst = STAGE_DIRS[i]
|
| 28 |
-
if src.exists():
|
| 29 |
-
if dst.exists():
|
| 30 |
-
shutil.rmtree(dst)
|
| 31 |
-
shutil.copytree(src, dst, dirs_exist_ok=True)
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
def _cleanup_stale(keep_last: int = 2):
|
| 35 |
-
for d in STAGE_DIRS[:-keep_last] if keep_last < len(STAGE_DIRS) else []:
|
| 36 |
-
if d.exists():
|
| 37 |
-
shutil.rmtree(d)
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
@app.post("/sync")
|
| 41 |
-
async def sync_repo(repo_id: str, local_path: str = "/tmp/altamira-checkout"):
|
| 42 |
-
_ensure_dirs()
|
| 43 |
-
try:
|
| 44 |
-
subprocess.run(
|
| 45 |
-
["git", "clone", f"https://huggingface.co/spaces/{repo_id}", local_path],
|
| 46 |
-
capture_output=True, text=True, check=False
|
| 47 |
-
)
|
| 48 |
-
if not Path(local_path).exists():
|
| 49 |
-
subprocess.run(
|
| 50 |
-
["git", "init", local_path],
|
| 51 |
-
capture_output=True, check=True,
|
| 52 |
-
)
|
| 53 |
-
_rotate_stages()
|
| 54 |
-
shutil.copytree(local_path, STAGE_DIRS[0], dirs_exist_ok=True)
|
| 55 |
-
_cleanup_stale()
|
| 56 |
-
return {"status": "synced", "repo": repo_id, "stage": str(STAGE_DIRS[0])}
|
| 57 |
-
except Exception as e:
|
| 58 |
-
raise HTTPException(status_code=500, detail=str(e))
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
@app.post("/rotate")
|
| 62 |
-
async def rotate_cache():
|
| 63 |
-
_ensure_dirs()
|
| 64 |
-
_rotate_stages()
|
| 65 |
-
_cleanup_stale()
|
| 66 |
-
return {"status": "rotated", "stages": [str(d) for d in STAGE_DIRS]}
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
@app.post("/webhook/results")
|
| 70 |
-
async def receive_results(request: Request):
|
| 71 |
-
try:
|
| 72 |
-
payload = await request.json()
|
| 73 |
-
token = request.query_params.get("token")
|
| 74 |
-
if not token:
|
| 75 |
-
raise HTTPException(status_code=401, detail="Verification key required")
|
| 76 |
-
|
| 77 |
-
# Log diagnostic payload to local file for Orchestrator to pull
|
| 78 |
-
log_file = Path("/tmp/altamira-results.json")
|
| 79 |
-
with open(log_file, "a") as f:
|
| 80 |
-
f.write(json.dumps(payload) + "\n")
|
| 81 |
-
|
| 82 |
-
return {"status": "received", "payload_size": len(str(payload))}
|
| 83 |
-
except Exception as e:
|
| 84 |
-
raise HTTPException(status_code=400, detail=f"Invalid payload: {str(e)}")
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
@app.get("/health")
|
| 88 |
-
async def health():
|
| 89 |
-
return {"status": "healthy", "worker": "altamira"}
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
if __name__ == "__main__":
|
| 93 |
-
import uvicorn
|
| 94 |
-
uvicorn.run(app, host="0.0.0.0", port=7860)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|