llama-stack/llama_stack/distribution/server/server.py
Ashwin Bharambe 6bb57e72a7
Remove "routing_table" and "routing_key" concepts for the user (#201)
This PR makes several core changes to the developer experience surrounding Llama Stack.

Background: PR #92 introduced the notion of "routing" to the Llama Stack. It introduces three object types: (1) models, (2) shields and (3) memory banks. Each of these objects can be associated with a distinct provider. So you can get model A to be inferenced locally while model B, C can be inference remotely (e.g.)

However, this had a few drawbacks:

you could not address the provider instances -- i.e., if you configured "meta-reference" with a given model, you could not assign an identifier to this instance which you could re-use later.
the above meant that you could not register a "routing_key" (e.g. model) dynamically and say "please use this existing provider I have already configured" for a new model.
the terms "routing_table" and "routing_key" were exposed directly to the user. in my view, this is way too much overhead for a new user (which almost everyone is.) people come to the stack wanting to do ML and encounter a completely unexpected term.
What this PR does: This PR structures the run config with only a single prominent key:

- providers
Providers are instances of configured provider types. Here's an example which shows two instances of the remote::tgi provider which are serving two different models.

providers:
  inference:
  - provider_id: foo
    provider_type: remote::tgi
    config: { ... }
  - provider_id: bar
    provider_type: remote::tgi
    config: { ... }
Secondly, the PR adds dynamic registration of { models | shields | memory_banks } to the API surface. The distribution still acts like a "routing table" (as previously) except that it asks the backing providers for a listing of these objects. For example it asks a TGI or Ollama inference adapter what models it is serving. Only the models that are being actually served can be requested by the user for inference. Otherwise, the Stack server will throw an error.

When dynamically registering these objects, you can use the provider IDs shown above. Info about providers can be obtained using the Api.inspect set of endpoints (/providers, /routes, etc.)

The above examples shows the correspondence between inference providers and models registry items. Things work similarly for the safety <=> shields and memory <=> memory_banks pairs.

Registry: This PR also makes it so that Providers need to implement additional methods for registering and listing objects. For example, each Inference provider is now expected to implement the ModelsProtocolPrivate protocol (naming is not great!) which consists of two methods

register_model
list_models
The goal is to inform the provider that a certain model needs to be supported so the provider can make any relevant backend changes if needed (or throw an error if the model cannot be supported.)

There are many other cleanups included some of which are detailed in a follow-up comment.
2024-10-10 10:24:13 -07:00

343 lines
10 KiB
Python

# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
import asyncio
import functools
import inspect
import json
import signal
import traceback
from contextlib import asynccontextmanager
from ssl import SSLError
from typing import Any, Dict, Optional
import fire
import httpx
import yaml
from fastapi import Body, FastAPI, HTTPException, Request, Response
from fastapi.exceptions import RequestValidationError
from fastapi.responses import JSONResponse, StreamingResponse
from pydantic import BaseModel, ValidationError
from termcolor import cprint
from typing_extensions import Annotated
from llama_stack.distribution.distribution import builtin_automatically_routed_apis
from llama_stack.providers.utils.telemetry.tracing import (
end_trace,
setup_logger,
SpanStatus,
start_trace,
)
from llama_stack.distribution.datatypes import * # noqa: F403
from llama_stack.distribution.request_headers import set_request_provider_data
from llama_stack.distribution.resolver import resolve_impls_with_routing
from .endpoints import get_all_api_endpoints
def create_sse_event(data: Any) -> str:
if isinstance(data, BaseModel):
data = data.json()
else:
data = json.dumps(data)
return f"data: {data}\n\n"
async def global_exception_handler(request: Request, exc: Exception):
traceback.print_exception(exc)
http_exc = translate_exception(exc)
return JSONResponse(
status_code=http_exc.status_code, content={"error": {"detail": http_exc.detail}}
)
def translate_exception(exc: Exception) -> Union[HTTPException, RequestValidationError]:
if isinstance(exc, ValidationError):
exc = RequestValidationError(exc.raw_errors)
if isinstance(exc, RequestValidationError):
return HTTPException(
status_code=400,
detail={
"errors": [
{
"loc": list(error["loc"]),
"msg": error["msg"],
"type": error["type"],
}
for error in exc.errors()
]
},
)
elif isinstance(exc, ValueError):
return HTTPException(status_code=400, detail=f"Invalid value: {str(exc)}")
elif isinstance(exc, PermissionError):
return HTTPException(status_code=403, detail=f"Permission denied: {str(exc)}")
elif isinstance(exc, TimeoutError):
return HTTPException(status_code=504, detail=f"Operation timed out: {str(exc)}")
elif isinstance(exc, NotImplementedError):
return HTTPException(status_code=501, detail=f"Not implemented: {str(exc)}")
else:
return HTTPException(
status_code=500,
detail="Internal server error: An unexpected error occurred.",
)
async def passthrough(
request: Request,
downstream_url: str,
downstream_headers: Optional[Dict[str, str]] = None,
):
await start_trace(request.path, {"downstream_url": downstream_url})
headers = dict(request.headers)
headers.pop("host", None)
headers.update(downstream_headers or {})
content = await request.body()
client = httpx.AsyncClient()
erred = False
try:
req = client.build_request(
method=request.method,
url=downstream_url,
headers=headers,
content=content,
params=request.query_params,
)
response = await client.send(req, stream=True)
async def stream_response():
async for chunk in response.aiter_raw(chunk_size=64):
yield chunk
await response.aclose()
await client.aclose()
return StreamingResponse(
stream_response(),
status_code=response.status_code,
headers=dict(response.headers),
media_type=response.headers.get("content-type"),
)
except httpx.ReadTimeout:
erred = True
return Response(content="Downstream server timed out", status_code=504)
except httpx.NetworkError as e:
erred = True
return Response(content=f"Network error: {str(e)}", status_code=502)
except httpx.TooManyRedirects:
erred = True
return Response(content="Too many redirects", status_code=502)
except SSLError as e:
erred = True
return Response(content=f"SSL error: {str(e)}", status_code=502)
except httpx.HTTPStatusError as e:
erred = True
return Response(content=str(e), status_code=e.response.status_code)
except Exception as e:
erred = True
return Response(content=f"Unexpected error: {str(e)}", status_code=500)
finally:
await end_trace(SpanStatus.OK if not erred else SpanStatus.ERROR)
def handle_sigint(app, *args, **kwargs):
print("SIGINT or CTRL-C detected. Exiting gracefully...")
async def run_shutdown():
for impl in app.__llama_stack_impls__.values():
print(f"Shutting down {impl}")
await impl.shutdown()
asyncio.run(run_shutdown())
loop = asyncio.get_event_loop()
for task in asyncio.all_tasks(loop):
task.cancel()
loop.stop()
@asynccontextmanager
async def lifespan(app: FastAPI):
print("Starting up")
yield
print("Shutting down")
for impl in app.__llama_stack_impls__.values():
await impl.shutdown()
def create_dynamic_passthrough(
downstream_url: str, downstream_headers: Optional[Dict[str, str]] = None
):
async def endpoint(request: Request):
return await passthrough(request, downstream_url, downstream_headers)
return endpoint
def is_streaming_request(func_name: str, request: Request, **kwargs):
# TODO: pass the api method and punt it to the Protocol definition directly
return kwargs.get("stream", False)
async def maybe_await(value):
if inspect.iscoroutine(value):
return await value
return value
async def sse_generator(event_gen):
try:
async for item in event_gen:
yield create_sse_event(item)
await asyncio.sleep(0.01)
except asyncio.CancelledError:
print("Generator cancelled")
await event_gen.aclose()
except Exception as e:
traceback.print_exception(e)
yield create_sse_event(
{
"error": {
"message": str(translate_exception(e)),
},
}
)
finally:
await end_trace()
def create_dynamic_typed_route(func: Any, method: str):
async def endpoint(request: Request, **kwargs):
await start_trace(func.__name__)
set_request_provider_data(request.headers)
is_streaming = is_streaming_request(func.__name__, request, **kwargs)
try:
if is_streaming:
return StreamingResponse(
sse_generator(func(**kwargs)), media_type="text/event-stream"
)
else:
value = func(**kwargs)
return await maybe_await(value)
except Exception as e:
traceback.print_exception(e)
raise translate_exception(e) from e
finally:
await end_trace()
sig = inspect.signature(func)
new_params = [
inspect.Parameter(
"request", inspect.Parameter.POSITIONAL_OR_KEYWORD, annotation=Request
)
]
new_params.extend(sig.parameters.values())
if method == "post":
# make sure every parameter is annotated with Body() so FASTAPI doesn't
# do anything too intelligent and ask for some parameters in the query
# and some in the body
new_params = [new_params[0]] + [
param.replace(annotation=Annotated[param.annotation, Body(..., embed=True)])
for param in new_params[1:]
]
endpoint.__signature__ = sig.replace(parameters=new_params)
return endpoint
def main(
yaml_config: str = "llamastack-run.yaml",
port: int = 5000,
disable_ipv6: bool = False,
):
with open(yaml_config, "r") as fp:
config = StackRunConfig(**yaml.safe_load(fp))
app = FastAPI()
impls = asyncio.run(resolve_impls_with_routing(config))
if Api.telemetry in impls:
setup_logger(impls[Api.telemetry])
all_endpoints = get_all_api_endpoints()
if config.apis:
apis_to_serve = set(config.apis)
else:
apis_to_serve = set(impls.keys())
for inf in builtin_automatically_routed_apis():
apis_to_serve.add(inf.routing_table_api.value)
apis_to_serve.add("inspect")
for api_str in apis_to_serve:
api = Api(api_str)
endpoints = all_endpoints[api]
impl = impls[api]
if is_passthrough(impl.__provider_spec__):
for endpoint in endpoints:
url = impl.__provider_config__.url.rstrip("/") + endpoint.route
getattr(app, endpoint.method)(endpoint.route)(
create_dynamic_passthrough(url)
)
else:
for endpoint in endpoints:
if not hasattr(impl, endpoint.name):
# ideally this should be a typing violation already
raise ValueError(
f"Could not find method {endpoint.name} on {impl}!!"
)
impl_method = getattr(impl, endpoint.name)
getattr(app, endpoint.method)(endpoint.route, response_model=None)(
create_dynamic_typed_route(
impl_method,
endpoint.method,
)
)
cprint(f"Serving API {api_str}", "white", attrs=["bold"])
for endpoint in endpoints:
cprint(f" {endpoint.method.upper()} {endpoint.route}", "white")
print("")
app.exception_handler(RequestValidationError)(global_exception_handler)
app.exception_handler(Exception)(global_exception_handler)
signal.signal(signal.SIGINT, functools.partial(handle_sigint, app))
app.__llama_stack_impls__ = impls
import uvicorn
# FYI this does not do hot-reloads
listen_host = "::" if not disable_ipv6 else "0.0.0.0"
print(f"Listening on {listen_host}:{port}")
uvicorn.run(app, host=listen_host, port=port)
if __name__ == "__main__":
fire.Fire(main)