grpc-server-kit

Batteries-optional async gRPC server toolkit (app facade, builder, TLS, lifecycle, health, interceptors, observability)

grpc-server-kit gives you a clean, production-grade foundation for building grpc.aio servers: a one-object GrpcApp facade, validated channel options, TLS/mTLS credential loading, graceful signal-driven shutdown, streaming-aware interceptors, and gRPC health checking — with optional extras for reflection, channelz, Prometheus metrics, OpenTelemetry tracing, Sentry, Dishka DI, and typed pydantic settings.

The core depends only on grpcio. Every integration is an opt-in extra, so you install exactly what you use.

Installation

pip install grpc-server-kit                      # core (grpcio only)
pip install "grpc-server-kit[settings]"          # + pydantic settings models
pip install "grpc-server-kit[health]"            # + gRPC health checking
pip install "grpc-server-kit[metrics,tracing]"   # + Prometheus + OpenTelemetry
pip install "grpc-server-kit[all]"               # everything

Requirements: Python 3.12+

Quick start

import asyncio

from grpc_server_kit import GrpcApp

app = GrpcApp(port=50051)
app.add_servicer(MyServicer(), add_MyServiceServicer_to_server)

asyncio.run(app.run())   # build + bind + SIGINT/SIGTERM + graceful shutdown

That's the whole server. GrpcApp builds the grpc.aio server, binds the port (TLS-aware), installs signal handlers, and serves until termination. More knobs when you need them:

from grpc_server_kit import GrpcApp, GrpcServerConfig
from grpc_server_kit.aio.interceptors import (
    AsyncContextInterceptor, AsyncExceptionHandlerInterceptor,
    AsyncMetricsInterceptor, AsyncRequestLoggerInterceptor, HeaderConfig,
)
from grpc_server_kit.observability.metrics import get_grpc_server_metrics

config = GrpcServerConfig(host="[::]", port=50051, grace_period=10.0)   # stdlib dataclass
app = GrpcApp(config, interceptors=[
    AsyncMetricsInterceptor(metrics=get_grpc_server_metrics(), service_name="my.pkg.MyService"),
    AsyncContextInterceptor([HeaderConfig("x-request-id", "request_id")]),
    AsyncRequestLoggerInterceptor(),
    AsyncExceptionHandlerInterceptor(MY_ERROR_STATUS_MAP),   # domain error → gRPC status
])
app.add_servicer(MyServicer(), add_MyServiceServicer_to_server)
app.enable_health(checkers=[DatabaseHealthChecker(session_maker)])   # [health] extra
app.enable_reflection(["my.pkg.MyService"])                          # [reflection] extra

asyncio.run(app.run())

Prefer pydantic-validated, env-friendly settings? Use grpc_server_kit.settings.BaseGrpcServerSettings (the [settings] extra) — GrpcServerConfig and BaseGrpcServerSettings are interchangeable everywhere the kit accepts settings.

Streaming-aware interceptors

Every interceptor wraps the whole RPC for all four call kinds — unary-unary, unary-stream, stream-unary, stream-stream. Durations cover the full stream, errors raised mid-stream are mapped to proper gRPC statuses, and cleanup always runs. Write your own by implementing a single async-generator hook:

from grpc_server_kit.aio.interceptors import AsyncServerInterceptor, RpcCall

class TimingInterceptor(AsyncServerInterceptor):
    async def around_call(self, call: RpcCall):
        start = time.perf_counter()
        try:
            yield                      # the RPC (or the full response stream) runs here
        finally:
            log.info("%s took %.3fs", call.method_name, time.perf_counter() - start)

Canonical chain order (outermost first): metrics → context → logger → tracing → exception handler → sentry. The exception handler maps exception types to gRPC statuses over the MRO and sends safe, non-leaking details; deliberate context.abort() calls pass through untouched. Sentry sits inside the exception handler so it observes raw handler exceptions before they become aborts (outside it would never capture anything), with a capture filter that reports only server-class errors.

Advanced API

GrpcApp is sugar over small, composable pieces you can use directly — for DI containers, custom lifecycles, or embedding:

from grpc_server_kit import bind_server_port
from grpc_server_kit.aio import AsyncGrpcServerBuilder, run_async_grpc_server

server = (
    AsyncGrpcServerBuilder(settings)
    .with_interceptors(interceptors)
    .with_servicers(lambda s: add_MyServiceServicer_to_server(MyServicer(), s))
    .with_reflection(["my.pkg.MyService"])
    .build()
)
bind_server_port(server, settings)            # honors settings.ssl_enabled (TLS/mTLS)
await run_async_grpc_server(server, address=f"{settings.host}:{settings.port}")

Dishka DI providers (cut the per-service boilerplate)

If your service uses Dishka, the kit ships ready-made providers so you register a handful of providers instead of hand-writing the interceptor list, settings aliasing, and a metrics provider. grpc_server_providers(...) bundles the standard set in one call:

from dishka import make_async_container
from grpc_server_kit.dishka import GrpcioProvider, FromDishka, inject
from grpc_server_kit.aio.dishka import grpc_server_providers
from grpc_server_kit.aio.server import AsyncServer
from grpc_server_kit.aio import run_async_grpc_server
from grpc_server_kit import bind_server_port

container = make_async_container(
    *grpc_server_providers(                       # settings + metrics + sentry/tracer seams
        settings.grpc,                            # + the canonical interceptor chain + AsyncServer
        service_name="my.pkg.MyService",
        error_status_map=MY_ERROR_STATUS_MAP,     # service-specific domain map (optional)
        metrics_enabled=settings.grpc.metrics_enabled,
    ),
    GrpcioProvider(),                             # request-scoped ServicerContext for FromDishka
    MyDomainProvider(),                           # your use-cases / repos
)

server = await container.get(AsyncServer)         # interceptors + options already wired
add_MyServiceServicer_to_server(MyServicer(), server)   # AsyncServer works directly
bind_server_port(server, settings.grpc)
await run_async_grpc_server(server, address=f"{settings.grpc.host}:{settings.grpc.port}")

The interceptor chain is built in the canonical order (metrics → context → logger → tracing → exception → sentry → DishkaAioInterceptor), with include_* toggles and extra_outer / extra_inner slots on GrpcServerInterceptorsProvider. Providers degrade gracefully: the metrics / Sentry / tracer seams resolve to None when disabled or their extra is absent, and the interceptors no-op. SDK initialization (sentry_sdk.init, OpenTelemetry setup, the /metrics HTTP server) is a process-wide concern and is not done by these providers. Requires [dishka] (+ [metrics]/[sentry]/[tracing] for those seams).

Granular providers (GrpcServerSettingsProvider, PrometheusGrpcServerMetricsProvider, SentryAdapterProvider, TracerAdapterProvider, GrpcServerInterceptorsProvider, AsyncGrpcServerProvider) are available for finer control.

What's inside

Layout convention: all async code lives under a single grpc_server_kit.aio package (mirroring grpc.aio), organized by domain (aio.health, aio.interceptors, aio.observability, aio.dishka). Modules that work for both sync and async stay at the top level — either directly (config, options, credentials, server, protocols, signals, settings) or as shared domain packages (interceptors → constants/utils, observability → protocols/metrics/sentry, dishka → providers).

Optional dependencies

Examples

Runnable, self-contained scripts in examples/ — each prints what it's doing and exits on its own:

• minimal_server.py — the smallest real server: GrpcApp, one servicer, asyncio.run(app.run()).
• observability_chain.py — the canonical interceptor chain (metrics → context → logger → tracing → exception handler → sentry) wired by hand.
• custom_interceptor.py — writing your own streaming-aware interceptor with AsyncServerInterceptor.around_call.
• health_checks.py — app.enable_health() with a custom dependency checker, driven through the real grpc_health.v1 stubs.

Documentation

Full documentation at bedrock-python.github.io/grpc-server-kit.

• For AI agents — the whole API surface, the rules that break a server when broken and a map of the rest, on one page to hand to a coding assistant

License

Apache 2.0 — see LICENSE.