phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-10-04 12:07:34 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix(pr specific): passes pre-commit

Browse source
This commit is contained in:
Emilio Garcia 2025-10-03 12:35:09 -04:00
parent 4aa2dc110d
commit 2b7a765d02
20 changed files with 547 additions and 516 deletions
Download patch file Download diff file Expand all files Collapse all files

3
llama_stack/core/library_client.py
View file

@ -32,7 +32,7 @@ from termcolor import cprint
from llama_stack.core.build import print_pip_install_help
from llama_stack.core.configure import parse_and_maybe_upgrade_config
from llama_stack.core.datatypes import Api, BuildConfig, BuildProvider, DistributionSpec
from llama_stack.core.datatypes import BuildConfig, BuildProvider, DistributionSpec
from llama_stack.core.request_headers import (
PROVIDER_DATA_VAR,
request_provider_data_context,
@ -49,7 +49,6 @@ from llama_stack.core.utils.context import preserve_contexts_async_generator
from llama_stack.core.utils.exec import in_notebook
from llama_stack.log import get_logger
logger = get_logger(name=__name__, category="core")
T = TypeVar("T")

7
llama_stack/core/server/server.py
View file

@ -63,7 +63,6 @@ from llama_stack.core.utils.context import preserve_contexts_async_generator
from llama_stack.log import get_logger
from llama_stack.providers.datatypes import Api
from .auth import AuthenticationMiddleware
from .quota import QuotaMiddleware
@ -236,9 +235,7 @@ def create_dynamic_typed_route(func: Any, method: str, route: str) -> Callable:
try:
if is_streaming:
gen = preserve_contexts_async_generator(
sse_generator(func(**kwargs)), [PROVIDER_DATA_VAR]
)
gen = preserve_contexts_async_generator(sse_generator(func(**kwargs)), [PROVIDER_DATA_VAR])
return StreamingResponse(gen, media_type="text/event-stream")
else:
value = func(**kwargs)
@ -282,7 +279,7 @@ def create_dynamic_typed_route(func: Any, method: str, route: str) -> Callable:
]
)
setattr(route_handler, "__signature__", sig.replace(parameters=new_params))
route_handler.__signature__ = sig.replace(parameters=new_params)
return route_handler

1
llama_stack/core/stack.py
View file

@ -359,7 +359,6 @@ class Stack:
await refresh_registry_once(impls)
self.impls = impls
# safely access impls without raising an exception
def get_impls(self) -> dict[Api, Any]:
if self.impls is None:

8
llama_stack/core/telemetry/__init__.py
View file

@ -1,4 +1,10 @@
# 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.
# the root directory of this source tree.
# All rights reserved.
#
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.

41
llama_stack/core/telemetry/telemetry.py
View file

@ -4,14 +4,14 @@
# This source code is licensed under the terms described in the LICENSE file in
# the root directory of this source tree.
from abc import abstractmethod
from fastapi import FastAPI
from pydantic import BaseModel
from opentelemetry.trace import Tracer
from fastapi import FastAPI
from opentelemetry.metrics import Meter
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.metrics import MeterProvider
from opentelemetry.sdk.resources import Attributes
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.trace import Tracer
from pydantic import BaseModel
from sqlalchemy import Engine
@ -19,39 +19,44 @@ class TelemetryProvider(BaseModel):
"""
TelemetryProvider standardizes how telemetry is provided to the application.
"""
@abstractmethod
def fastapi_middleware(self, app: FastAPI, *args, **kwargs):
"""
Injects FastAPI middleware that instruments the application for telemetry.
"""
...
@abstractmethod
def sqlalchemy_instrumentation(self, engine: Engine | None = None):
"""
Injects SQLAlchemy instrumentation that instruments the application for telemetry.
"""
...
@abstractmethod
def get_tracer(self,
instrumenting_module_name: str,
instrumenting_library_version: str | None = None,
tracer_provider: TracerProvider | None = None,
schema_url: str | None = None,
attributes: Attributes | None = None
def get_tracer(
self,
instrumenting_module_name: str,
instrumenting_library_version: str | None = None,
tracer_provider: TracerProvider | None = None,
schema_url: str | None = None,
attributes: Attributes | None = None,
) -> Tracer:
"""
Gets a tracer.
"""
...
@abstractmethod
def get_meter(self, name: str,
version: str = "",
meter_provider: MeterProvider | None = None,
schema_url: str | None = None,
attributes: Attributes | None = None) -> Meter:
def get_meter(
self,
name: str,
version: str = "",
meter_provider: MeterProvider | None = None,
schema_url: str | None = None,
attributes: Attributes | None = None,
) -> Meter:
"""
Gets a meter.
"""

11
llama_stack/providers/inline/telemetry/meta_reference/middleware.py
View file

@ -1,15 +1,22 @@
from aiohttp import hdrs
# 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.
from typing import Any
from aiohttp import hdrs
from llama_stack.apis.datatypes import Api
from llama_stack.core.external import ExternalApiSpec
from llama_stack.core.server.routes import find_matching_route, initialize_route_impls
from llama_stack.log import get_logger
from llama_stack.providers.utils.telemetry.tracing import end_trace, start_trace
logger = get_logger(name=__name__, category="telemetry::meta_reference")
class TracingMiddleware:
def __init__(
self,

11
llama_stack/providers/inline/telemetry/meta_reference/telemetry.py
View file

@ -10,7 +10,6 @@ import threading
from typing import Any, cast
from fastapi import FastAPI
from opentelemetry import metrics, trace
from opentelemetry.exporter.otlp.proto.http.metric_exporter import OTLPMetricExporter
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
@ -23,11 +22,6 @@ from opentelemetry.semconv.attributes import service_attributes
from opentelemetry.trace.propagation.tracecontext import TraceContextTextMapPropagator
from opentelemetry.util.types import Attributes
from llama_stack.core.external import ExternalApiSpec
from llama_stack.core.server.tracing import TelemetryProvider
from llama_stack.providers.inline.telemetry.meta_reference.middleware import TracingMiddleware
from llama_stack.apis.telemetry import (
Event,
MetricEvent,
@ -47,10 +41,13 @@ from llama_stack.apis.telemetry import (
UnstructuredLogEvent,
)
from llama_stack.core.datatypes import Api
from llama_stack.core.external import ExternalApiSpec
from llama_stack.core.server.tracing import TelemetryProvider
from llama_stack.log import get_logger
from llama_stack.providers.inline.telemetry.meta_reference.console_span_processor import (
ConsoleSpanProcessor,
)
from llama_stack.providers.inline.telemetry.meta_reference.middleware import TracingMiddleware
from llama_stack.providers.inline.telemetry.meta_reference.sqlite_span_processor import (
SQLiteSpanProcessor,
)
@ -381,7 +378,7 @@ class TelemetryAdapter(TelemetryDatasetMixin, Telemetry, TelemetryProvider):
max_depth=max_depth,
)
)
def fastapi_middleware(
self,
app: FastAPI,

5
llama_stack/providers/inline/telemetry/otel/__init__.py
View file

@ -12,13 +12,12 @@ __all__ = ["OTelTelemetryConfig"]
async def get_provider_impl(config: OTelTelemetryConfig, deps):
"""
Get the OTel telemetry provider implementation.
This function is called by the Llama Stack registry to instantiate
the provider.
"""
from .otel import OTelTelemetryProvider
# The provider is synchronously initialized via Pydantic model_post_init
# No async initialization needed
return OTelTelemetryProvider(config=config)

24
llama_stack/providers/inline/telemetry/otel/config.py
View file

@ -1,8 +1,13 @@
# 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.
from typing import Any, Literal
from pydantic import BaseModel, Field
type BatchSpanProcessor = Literal["batch"]
type SimpleSpanProcessor = Literal["simple"]
@ -13,26 +18,27 @@ class OTelTelemetryConfig(BaseModel):
Most configuration is set using environment variables.
See https://opentelemetry.io/docs/specs/otel/configuration/sdk-configuration-variables/ for more information.
"""
service_name: str = Field(
description="""The name of the service to be monitored.
description="""The name of the service to be monitored.
Is overridden by the OTEL_SERVICE_NAME or OTEL_RESOURCE_ATTRIBUTES environment variables.""",
)
service_version: str | None = Field(
default=None,
description="""The version of the service to be monitored.
Is overriden by the OTEL_RESOURCE_ATTRIBUTES environment variable."""
description="""The version of the service to be monitored.
Is overriden by the OTEL_RESOURCE_ATTRIBUTES environment variable.""",
)
deployment_environment: str | None = Field(
default=None,
description="""The name of the environment of the service to be monitored.
Is overriden by the OTEL_RESOURCE_ATTRIBUTES environment variable."""
description="""The name of the environment of the service to be monitored.
Is overriden by the OTEL_RESOURCE_ATTRIBUTES environment variable.""",
)
span_processor: BatchSpanProcessor | SimpleSpanProcessor | None = Field(
description="""The span processor to use.
description="""The span processor to use.
Is overriden by the OTEL_SPAN_PROCESSOR environment variable.""",
default="batch"
default="batch",
)
@classmethod
def sample_run_config(cls, __distro_dir__: str = "") -> dict[str, Any]:
"""Sample configuration for use in distributions."""

141
llama_stack/providers/inline/telemetry/otel/otel.py
View file

@ -1,24 +1,28 @@
# 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 os
from opentelemetry import trace, metrics
from fastapi import FastAPI
from opentelemetry import metrics, trace
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor
from opentelemetry.instrumentation.sqlalchemy import SQLAlchemyInstrumentor
from opentelemetry.metrics import Meter
from opentelemetry.sdk.metrics import MeterProvider
from opentelemetry.sdk.resources import Attributes, Resource
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor, SimpleSpanProcessor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk.metrics import MeterProvider
from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor
from opentelemetry.trace import Tracer
from opentelemetry.metrics import Meter
from opentelemetry.instrumentation.sqlalchemy import SQLAlchemyInstrumentor
from sqlalchemy import Engine
from llama_stack.core.telemetry.telemetry import TelemetryProvider
from llama_stack.log import get_logger
from sqlalchemy import Engine
from .config import OTelTelemetryConfig
from fastapi import FastAPI
logger = get_logger(name=__name__, category="telemetry::otel")
@ -27,6 +31,7 @@ class OTelTelemetryProvider(TelemetryProvider):
"""
A simple Open Telemetry native telemetry provider.
"""
config: OTelTelemetryConfig
def model_post_init(self, __context):
@ -56,66 +61,66 @@ class OTelTelemetryProvider(TelemetryProvider):
tracer_provider.add_span_processor(BatchSpanProcessor(otlp_span_exporter))
elif self.config.span_processor == "simple":
tracer_provider.add_span_processor(SimpleSpanProcessor(otlp_span_exporter))
meter_provider = MeterProvider(resource=resource)
metrics.set_meter_provider(meter_provider)
# Do not fail the application, but warn the user if the endpoints are not set properly.
if not os.environ.get("OTEL_EXPORTER_OTLP_ENDPOINT"):
if not os.environ.get("OTEL_EXPORTER_OTLP_TRACES_ENDPOINT"):
logger.warning("OTEL_EXPORTER_OTLP_ENDPOINT or OTEL_EXPORTER_OTLP_TRACES_ENDPOINT is not set. Traces will not be exported.")
logger.warning(
"OTEL_EXPORTER_OTLP_ENDPOINT or OTEL_EXPORTER_OTLP_TRACES_ENDPOINT is not set. Traces will not be exported."
)
if not os.environ.get("OTEL_EXPORTER_OTLP_METRICS_ENDPOINT"):
logger.warning("OTEL_EXPORTER_OTLP_ENDPOINT or OTEL_EXPORTER_OTLP_METRICS_ENDPOINT is not set. Metrics will not be exported.")
logger.warning(
"OTEL_EXPORTER_OTLP_ENDPOINT or OTEL_EXPORTER_OTLP_METRICS_ENDPOINT is not set. Metrics will not be exported."
)
def fastapi_middleware(self, app: FastAPI):
"""
Instrument FastAPI with OTel for automatic tracing and metrics.
Captures:
- Distributed traces for all HTTP requests (via FastAPIInstrumentor)
- HTTP metrics following semantic conventions (custom middleware)
"""
# Enable automatic tracing
FastAPIInstrumentor.instrument_app(app)
# Add custom middleware for HTTP metrics
meter = self.get_meter("llama_stack.http.server")
# Create HTTP metrics following semantic conventions
# https://opentelemetry.io/docs/specs/semconv/http/http-metrics/
request_duration = meter.create_histogram(
"http.server.request.duration",
unit="ms",
description="Duration of HTTP server requests"
"http.server.request.duration", unit="ms", description="Duration of HTTP server requests"
)
active_requests = meter.create_up_down_counter(
"http.server.active_requests",
unit="requests",
description="Number of active HTTP server requests"
"http.server.active_requests", unit="requests", description="Number of active HTTP server requests"
)
request_count = meter.create_counter(
"http.server.request.count",
unit="requests",
description="Total number of HTTP server requests"
"http.server.request.count", unit="requests", description="Total number of HTTP server requests"
)
# Add middleware to record metrics
@app.middleware("http") # type: ignore[misc]
async def http_metrics_middleware(request, call_next):
import time
# Record active request
active_requests.add(1, {
"http.method": request.method,
"http.route": request.url.path,
})
active_requests.add(
1,
{
"http.method": request.method,
"http.route": request.url.path,
},
)
start_time = time.time()
status_code = 500 # Default to error
try:
response = await call_next(request)
status_code = response.status_code
@ -124,22 +129,24 @@ class OTelTelemetryProvider(TelemetryProvider):
finally:
# Record metrics
duration_ms = (time.time() - start_time) * 1000
attributes = {
"http.method": request.method,
"http.route": request.url.path,
"http.status_code": status_code,
}
request_duration.record(duration_ms, attributes)
request_count.add(1, attributes)
active_requests.add(-1, {
"http.method": request.method,
"http.route": request.url.path,
})
return response
active_requests.add(
-1,
{
"http.method": request.method,
"http.route": request.url.path,
},
)
return response
def sqlalchemy_instrumentation(self, engine: Engine | None = None):
kwargs = {}
@ -147,34 +154,30 @@ class OTelTelemetryProvider(TelemetryProvider):
kwargs["engine"] = engine
SQLAlchemyInstrumentor().instrument(**kwargs)
def get_tracer(self,
instrumenting_module_name: str,
instrumenting_library_version: str | None = None,
tracer_provider: TracerProvider | None = None,
schema_url: str | None = None,
attributes: Attributes | None = None
def get_tracer(
self,
instrumenting_module_name: str,
instrumenting_library_version: str | None = None,
tracer_provider: TracerProvider | None = None,
schema_url: str | None = None,
attributes: Attributes | None = None,
) -> Tracer:
return trace.get_tracer(
instrumenting_module_name=instrumenting_module_name,
instrumenting_library_version=instrumenting_library_version,
tracer_provider=tracer_provider,
schema_url=schema_url,
attributes=attributes
instrumenting_module_name=instrumenting_module_name,
instrumenting_library_version=instrumenting_library_version,
tracer_provider=tracer_provider,
schema_url=schema_url,
attributes=attributes,
)
def get_meter(self,
name: str,
version: str = "",
meter_provider: MeterProvider | None = None,
schema_url: str | None = None,
attributes: Attributes | None = None
def get_meter(
self,
name: str,
version: str = "",
meter_provider: MeterProvider | None = None,
schema_url: str | None = None,
attributes: Attributes | None = None,
) -> Meter:
return metrics.get_meter(
name=name,
version=version,
meter_provider=meter_provider,
schema_url=schema_url,
attributes=attributes
)
name=name, version=version, meter_provider=meter_provider, schema_url=schema_url, attributes=attributes
)