Migrate from FastAPI (Python) to GoFr

Mental model

GoFr handlers are synchronous functions, but each one runs in its own goroutine — you don't decorate handlers with async because the runtime already gives you concurrency for free. Where FastAPI uses async def + await to avoid blocking the event loop, GoFr blocks the goroutine and lets the Go scheduler interleave others. The result is the same shape of code as a sync FastAPI route, with throughput closer to async.

Pydantic's runtime validation becomes compile-time struct typing plus tag-based validation on Bind. FastAPI's Depends() injection is replaced by passing dependencies through constructors or accessing datasources via *gofr.Context.

Side-by-side: FastAPI handler ↔ GoFr handler

FastAPI:

from fastapi import FastAPI
from pydantic import BaseModel

class CreateUser(BaseModel):
    name: str
    email: str

app = FastAPI()

@app.post("/users")
async def create_user(payload: CreateUser):
    user = await db.create(payload.dict())
    return user

GoFr:

package main

import "gofr.dev/pkg/gofr"

type CreateUser struct {
    Name  string `json:"name"`
    Email string `json:"email"`
}

func main() {
    app := gofr.New()
    app.POST("/users", func(c *gofr.Context) (any, error) {
        var input CreateUser
        if err := c.Bind(&input); err != nil {
            return nil, err
        }
        return createUser(c, input)
    })
    app.Run()
}

Concurrency: async/await → goroutines

A typical FastAPI deployment runs uvicorn workers, each one running its own event loop with cooperative async tasks. A GoFr service is a single binary; each request runs as a goroutine, and I/O calls block the goroutine without blocking the OS thread. There is no await keyword in user code — the framework, drivers, and HTTP/SQL clients propagate cancellation via context.Context (which *gofr.Context embeds).

If you previously offloaded CPU work via run_in_threadpool, in Go you simply call the function: the scheduler will move blocked goroutines off the worker threads.

Validation and OpenAPI

GoFr ships a Swagger UI that renders any openapi.json you place in the static directory — see the Swagger documentation guide.

Dependency injection

FastAPI's Depends() is replaced by either:

• Constructor passing — build a struct holding your dependencies and use methods as handlers.
• *gofr.Context — datasources (SQL, Redis, Mongo, Pub/Sub) are accessed through the request context, so per-request injection of those is automatic.

Datasources

FastAPI users typically reach for SQLAlchemy / Tortoise / Motor. In GoFr, SQL and Redis are auto-initialized from environment variables — set DB_DIALECT, DB_HOST, DB_PORT, DB_USER, DB_PASSWORD, DB_NAME (or REDIS_HOST, REDIS_PORT) in configs/.env and gofr.New() wires the connection for you. Other clients are registered explicitly with a provider, e.g.:

app.AddMongo(mongo.New(mongo.Config{/* ... */}))

Then access them inside the handler via c.SQL, c.Redis, c.Mongo. GoFr supports SQL (MySQL/Postgres/Oracle/SQLite/SQL Server), MongoDB, Redis, Cassandra, ScyllaDB, Couchbase, ArangoDB, Dgraph and SurrealDB. SQL/Mongo/Redis/Dgraph migrations are first-class — see the datasources reference.

Observability

FastAPI users typically wire opentelemetry-instrumentation-fastapi and prometheus-fastapi-instrumentator themselves. GoFr emits OpenTelemetry traces, Prometheus metrics at /metrics, and structured JSON logs (with trace IDs) by default. Health is exposed at /.well-known/health. Log levels are changeable at runtime via the remote log-level endpoint.

Gradual adoption

Run your FastAPI service alongside a new GoFr microservice and call it from GoFr using the built-in HTTP client with circuit breaker + retry + rate limiting:

app.AddHTTPService("legacy-api", "http://legacy-fastapi:8000")

Move endpoints over progressively, repointing your gateway/load balancer until the old service can be retired.