Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,83 +1,106 @@
|
|
| 1 |
import os
|
| 2 |
import time
|
| 3 |
import random
|
|
|
|
| 4 |
import httpx
|
| 5 |
from typing import Dict
|
| 6 |
|
| 7 |
from fastapi import FastAPI, Request, HTTPException
|
| 8 |
-
from fastapi.responses import StreamingResponse,
|
| 9 |
|
| 10 |
app = FastAPI()
|
| 11 |
|
| 12 |
-
# ββ
|
|
|
|
|
|
|
| 13 |
MASTER_API_KEY = os.getenv("MASTER_API_KEY", "changeme")
|
| 14 |
BASE_URL = os.getenv("BASE_URL", "https://ollama.com")
|
| 15 |
|
| 16 |
-
# ββ
|
|
|
|
|
|
|
| 17 |
def load_keys(prefix="OLLAMA_KEY_"):
|
| 18 |
keys = []
|
| 19 |
for k, v in os.environ.items():
|
| 20 |
if k.startswith(prefix) and v:
|
| 21 |
keys.append(v.strip())
|
| 22 |
-
return
|
| 23 |
|
| 24 |
API_KEYS = load_keys()
|
|
|
|
| 25 |
|
| 26 |
-
# ββ
|
|
|
|
|
|
|
| 27 |
key_status: Dict[str, Dict] = {
|
| 28 |
-
k: {
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
for k in API_KEYS
|
| 30 |
}
|
| 31 |
|
| 32 |
-
# ββ
|
|
|
|
|
|
|
| 33 |
def auth(req: Request):
|
| 34 |
if req.headers.get("Authorization") != f"Bearer {MASTER_API_KEY}":
|
| 35 |
raise HTTPException(401, "Unauthorized")
|
| 36 |
|
| 37 |
-
# ββ
|
|
|
|
|
|
|
| 38 |
def pick_key():
|
| 39 |
now = time.time()
|
|
|
|
| 40 |
valid = [
|
| 41 |
k for k, v in key_status.items()
|
| 42 |
if v["status"] != "dead" and v["cooldown"] < now
|
| 43 |
]
|
|
|
|
| 44 |
return random.choice(valid) if valid else None
|
| 45 |
|
| 46 |
-
# ββ
|
|
|
|
|
|
|
| 47 |
def mark_limit(k):
|
| 48 |
key_status[k]["cooldown"] = time.time() + 60
|
|
|
|
| 49 |
|
| 50 |
def mark_dead(k):
|
| 51 |
key_status[k]["status"] = "dead"
|
| 52 |
|
| 53 |
def mark_ok(k):
|
| 54 |
key_status[k]["fail"] = 0
|
|
|
|
| 55 |
|
| 56 |
-
# ββ
|
|
|
|
|
|
|
| 57 |
@app.api_route("/v1/{path:path}", methods=["GET", "POST", "PUT", "DELETE"])
|
| 58 |
async def proxy(req: Request, path: str):
|
| 59 |
auth(req)
|
| 60 |
|
| 61 |
-
# build target url dinamis
|
| 62 |
target_url = f"{BASE_URL}/v1/{path}"
|
| 63 |
|
| 64 |
-
# ambil body (kalau ada)
|
| 65 |
body = await req.body()
|
| 66 |
|
| 67 |
-
#
|
| 68 |
headers = dict(req.headers)
|
| 69 |
headers.pop("host", None)
|
|
|
|
| 70 |
|
| 71 |
for _ in range(len(API_KEYS)):
|
| 72 |
key = pick_key()
|
|
|
|
| 73 |
if not key:
|
| 74 |
-
return
|
| 75 |
|
| 76 |
headers["Authorization"] = f"Bearer {key}"
|
| 77 |
|
| 78 |
try:
|
| 79 |
async with httpx.AsyncClient(timeout=None) as client:
|
| 80 |
-
|
| 81 |
r = await client.request(
|
| 82 |
method=req.method,
|
| 83 |
url=target_url,
|
|
@@ -86,44 +109,64 @@ async def proxy(req: Request, path: str):
|
|
| 86 |
params=req.query_params
|
| 87 |
)
|
| 88 |
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
if "text/event-stream" in r.headers.get("content-type", ""):
|
| 95 |
-
return StreamingResponse(
|
| 96 |
-
r.aiter_raw(),
|
| 97 |
-
media_type="text/event-stream"
|
| 98 |
-
)
|
| 99 |
-
|
| 100 |
-
return Response(
|
| 101 |
-
content=r.content,
|
| 102 |
-
status_code=r.status_code,
|
| 103 |
-
headers=dict(r.headers)
|
| 104 |
-
)
|
| 105 |
|
| 106 |
-
|
| 107 |
-
elif r.status_code == 429:
|
| 108 |
-
mark_limit(key)
|
| 109 |
|
| 110 |
-
#
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 115 |
|
| 116 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
key_status[key]["fail"] += 1
|
|
|
|
| 118 |
if key_status[key]["fail"] >= 3:
|
| 119 |
mark_dead(key)
|
| 120 |
|
| 121 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 122 |
|
| 123 |
-
# ββ
|
|
|
|
|
|
|
| 124 |
@app.get("/")
|
| 125 |
-
def
|
| 126 |
return {
|
| 127 |
-
"status": "
|
| 128 |
-
"keys": len(API_KEYS)
|
|
|
|
| 129 |
}
|
|
|
|
| 1 |
import os
|
| 2 |
import time
|
| 3 |
import random
|
| 4 |
+
import json
|
| 5 |
import httpx
|
| 6 |
from typing import Dict
|
| 7 |
|
| 8 |
from fastapi import FastAPI, Request, HTTPException
|
| 9 |
+
from fastapi.responses import Response, StreamingResponse, JSONResponse
|
| 10 |
|
| 11 |
app = FastAPI()
|
| 12 |
|
| 13 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 14 |
+
# CONFIG
|
| 15 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 16 |
MASTER_API_KEY = os.getenv("MASTER_API_KEY", "changeme")
|
| 17 |
BASE_URL = os.getenv("BASE_URL", "https://ollama.com")
|
| 18 |
|
| 19 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 20 |
+
# LOAD KEYS (OLLAMA_KEY_1, OLLAMA_KEY_2, ...)
|
| 21 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 22 |
def load_keys(prefix="OLLAMA_KEY_"):
|
| 23 |
keys = []
|
| 24 |
for k, v in os.environ.items():
|
| 25 |
if k.startswith(prefix) and v:
|
| 26 |
keys.append(v.strip())
|
| 27 |
+
return keys
|
| 28 |
|
| 29 |
API_KEYS = load_keys()
|
| 30 |
+
print(f"[INIT] Loaded {len(API_KEYS)} keys")
|
| 31 |
|
| 32 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 33 |
+
# KEY STATE
|
| 34 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 35 |
key_status: Dict[str, Dict] = {
|
| 36 |
+
k: {
|
| 37 |
+
"fail": 0,
|
| 38 |
+
"cooldown": 0,
|
| 39 |
+
"status": "active"
|
| 40 |
+
}
|
| 41 |
for k in API_KEYS
|
| 42 |
}
|
| 43 |
|
| 44 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 45 |
+
# AUTH
|
| 46 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 47 |
def auth(req: Request):
|
| 48 |
if req.headers.get("Authorization") != f"Bearer {MASTER_API_KEY}":
|
| 49 |
raise HTTPException(401, "Unauthorized")
|
| 50 |
|
| 51 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 52 |
+
# PICK KEY
|
| 53 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 54 |
def pick_key():
|
| 55 |
now = time.time()
|
| 56 |
+
|
| 57 |
valid = [
|
| 58 |
k for k, v in key_status.items()
|
| 59 |
if v["status"] != "dead" and v["cooldown"] < now
|
| 60 |
]
|
| 61 |
+
|
| 62 |
return random.choice(valid) if valid else None
|
| 63 |
|
| 64 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 65 |
+
# STATUS HANDLER
|
| 66 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 67 |
def mark_limit(k):
|
| 68 |
key_status[k]["cooldown"] = time.time() + 60
|
| 69 |
+
key_status[k]["status"] = "limited"
|
| 70 |
|
| 71 |
def mark_dead(k):
|
| 72 |
key_status[k]["status"] = "dead"
|
| 73 |
|
| 74 |
def mark_ok(k):
|
| 75 |
key_status[k]["fail"] = 0
|
| 76 |
+
key_status[k]["status"] = "active"
|
| 77 |
|
| 78 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 79 |
+
# UNIVERSAL OPENAI PROXY (/v1/*)
|
| 80 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 81 |
@app.api_route("/v1/{path:path}", methods=["GET", "POST", "PUT", "DELETE"])
|
| 82 |
async def proxy(req: Request, path: str):
|
| 83 |
auth(req)
|
| 84 |
|
|
|
|
| 85 |
target_url = f"{BASE_URL}/v1/{path}"
|
| 86 |
|
|
|
|
| 87 |
body = await req.body()
|
| 88 |
|
| 89 |
+
# clean headers
|
| 90 |
headers = dict(req.headers)
|
| 91 |
headers.pop("host", None)
|
| 92 |
+
headers.pop("content-length", None)
|
| 93 |
|
| 94 |
for _ in range(len(API_KEYS)):
|
| 95 |
key = pick_key()
|
| 96 |
+
|
| 97 |
if not key:
|
| 98 |
+
return JSONResponse({"error": "all keys exhausted"}, 503)
|
| 99 |
|
| 100 |
headers["Authorization"] = f"Bearer {key}"
|
| 101 |
|
| 102 |
try:
|
| 103 |
async with httpx.AsyncClient(timeout=None) as client:
|
|
|
|
| 104 |
r = await client.request(
|
| 105 |
method=req.method,
|
| 106 |
url=target_url,
|
|
|
|
| 109 |
params=req.query_params
|
| 110 |
)
|
| 111 |
|
| 112 |
+
# βββββββββββββββββββββββββββββ
|
| 113 |
+
# SUCCESS
|
| 114 |
+
# βββββββββββββββββββββββββββββ
|
| 115 |
+
if r.status_code < 400:
|
| 116 |
+
mark_ok(key)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
|
| 118 |
+
content_type = r.headers.get("content-type", "")
|
|
|
|
|
|
|
| 119 |
|
| 120 |
+
# streaming (SSE)
|
| 121 |
+
if "text/event-stream" in content_type:
|
| 122 |
+
return StreamingResponse(
|
| 123 |
+
r.aiter_raw(),
|
| 124 |
+
media_type="text/event-stream"
|
| 125 |
+
)
|
| 126 |
+
|
| 127 |
+
return Response(
|
| 128 |
+
content=r.content,
|
| 129 |
+
status_code=r.status_code,
|
| 130 |
+
headers=dict(r.headers)
|
| 131 |
+
)
|
| 132 |
|
| 133 |
+
# βββββββββββββββββββββββββββββ
|
| 134 |
+
# RATE LIMIT
|
| 135 |
+
# βββββββββββββββββββββββββββββ
|
| 136 |
+
if r.status_code == 429:
|
| 137 |
+
mark_limit(key)
|
| 138 |
+
|
| 139 |
+
# βββββββββββββββββββββββββββββ
|
| 140 |
+
# SERVER ERROR
|
| 141 |
+
# βββββββββββββββββββββββββββββ
|
| 142 |
+
if r.status_code >= 500:
|
| 143 |
+
key_status[key]["fail"] += 1
|
| 144 |
+
if key_status[key]["fail"] >= 3:
|
| 145 |
+
mark_dead(key)
|
| 146 |
+
|
| 147 |
+
except Exception as e:
|
| 148 |
key_status[key]["fail"] += 1
|
| 149 |
+
|
| 150 |
if key_status[key]["fail"] >= 3:
|
| 151 |
mark_dead(key)
|
| 152 |
|
| 153 |
+
print(f"[ERROR] key={key[:6]} err={str(e)}")
|
| 154 |
+
|
| 155 |
+
return JSONResponse(
|
| 156 |
+
{
|
| 157 |
+
"error": "all keys failed",
|
| 158 |
+
"hint": "check API keys or BASE_URL"
|
| 159 |
+
},
|
| 160 |
+
500
|
| 161 |
+
)
|
| 162 |
|
| 163 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 164 |
+
# HEALTH CHECK
|
| 165 |
+
# βββββββββββββββββββββββββββββββββββββββββββββ
|
| 166 |
@app.get("/")
|
| 167 |
+
def root():
|
| 168 |
return {
|
| 169 |
+
"status": "ok",
|
| 170 |
+
"keys": len(API_KEYS),
|
| 171 |
+
"mode": "openai-compatible-proxy"
|
| 172 |
}
|