URL、ヘッダー、メディアタイプによるバージョニング
バージョニング方式を比較し、クライアントが破壊的変更なくアップグレードできる、整理されたルートグループを実装します。
「URL、ヘッダー、メディアタイプによるバージョニング」はCoddyKit上の無料FastAPI Backend Development Bootcampレッスンです。 これはレッスン1/4です。 下記で完全なレッスンを無料で読むことができます。その後、ブラウザ内の組み込みコードエディタと24時間対応のAIチューターでハンズオン演習できます。 これはFastAPI Backend Development Bootcamp学習パスの一部であり、ウェブとCoddyKitアプリ全体で進捗が同期されます。 FastAPI Backend Development Bootcampコースには全4レッスンが含まれています。
このレッスンの一部はまだ翻訳されておらず、英語で表示されています。
Why Version an API at All?
Once real clients depend on your API, you can no longer change a response shape freely. Renaming a field, removing a key, or changing a status code can break production apps you do not control.
Versioning lets you ship breaking changes under a new label while the old behavior keeps working. Clients upgrade on their own schedule.
- Non-breaking change (add an optional field) usually needs no new version.
- Breaking change (remove/rename a field, change types, change semantics) needs a version boundary.
This lesson compares the three classic strategies: URL path, Header, and Media-Type versioning.
Strategy 1: URL Path Versioning
The most common and most visible approach: put the version directly in the path, e.g. /v1/users and /v2/users.
- Pros: obvious in logs and browsers, trivial to route, easy to cache, simple to document.
- Cons: the version leaks into every URL; technically a URL should identify a resource, not a representation.
In FastAPI you express this with a prefix on an APIRouter. Each version gets its own router and its own prefix.
from fastapi import APIRouter, FastAPI
app = FastAPI()
v1 = APIRouter(prefix="/v1", tags=["v1"])
v2 = APIRouter(prefix="/v2", tags=["v2"])
@v1.get("/users/{user_id}")
def get_user_v1(user_id: int):
return {"id": user_id, "name": "Ada Lovelace"}
@v2.get("/users/{user_id}")
def get_user_v2(user_id: int):
# v2 splits name into first/last
return {"id": user_id, "first_name": "Ada", "last_name": "Lovelace"}
app.include_router(v1)
app.include_router(v2)Clean Route Grouping with Sub-Routers
Do not stack every endpoint onto one giant version router. Group by resource, then mount each resource router under the version router so structure stays clean as the API grows.
Here users_router is built independently and then included into v1. The combined path becomes /v1/users/....
from fastapi import APIRouter, FastAPI
app = FastAPI()
users_router = APIRouter(prefix="/users", tags=["users"])
@users_router.get("/{user_id}")
def get_user(user_id: int):
return {"id": user_id, "name": "Grace Hopper"}
v1 = APIRouter(prefix="/v1")
v1.include_router(users_router) # -> /v1/users/{user_id}
app.include_router(v1)Strategy 2: Header Versioning
Here the URL stays clean (/users/42) and the client signals the version through a custom request header, commonly X-API-Version: 2.
- Pros: URLs are version-free and stable; the resource path never changes.
- Cons: harder to test in a browser, easy to forget the header, and caching layers must be told to vary on it.
In FastAPI you read the header with a typed parameter and branch (or dispatch) on it.
from fastapi import FastAPI, Header
app = FastAPI()
@app.get("/users/{user_id}")
def get_user(user_id: int, x_api_version: int = Header(default=1)):
if x_api_version >= 2:
return {"id": user_id, "first_name": "Ada", "last_name": "Lovelace"}
return {"id": user_id, "name": "Ada Lovelace"}A Dependency to Resolve the Version
Branching inside every endpoint gets messy. Extract the version logic into a reusable dependency that validates the header once and rejects unsupported versions with a clean 406.
Any endpoint can now depend on api_version and trust it is one of the supported values.
from fastapi import FastAPI, Header, HTTPException, Depends
app = FastAPI()
SUPPORTED = {1, 2}
def api_version(x_api_version: int = Header(default=1)) -> int:
if x_api_version not in SUPPORTED:
raise HTTPException(
status_code=406,
detail=f"Unsupported API version {x_api_version}",
)
return x_api_version
@app.get("/users/{user_id}")
def get_user(user_id: int, version: int = Depends(api_version)):
return {"id": user_id, "version": version}Strategy 3: Media-Type (Content Negotiation) Versioning
The most RESTful but least common approach. The client asks for a specific representation via the Accept header using a vendor media type:
Accept: application/vnd.myapp.v1+jsonAccept: application/vnd.myapp.v2+json
The URL identifies the resource; the media type identifies the representation/version. This is true HTTP content negotiation.
- Pros: URLs are clean and semantically pure; you version the representation, not the resource.
- Cons: verbose, hard for casual clients, weak tooling support, easy to typo.
Parsing the Vendor Media Type
The core of media-type versioning is parsing the vendor string into a version number. That parsing is pure Python and easy to unit-test on its own, independent of any framework.
Below we extract the vN token from an Accept value, defaulting to v1 when it is missing or malformed.
import re
PATTERN = re.compile(r"application/vnd\.myapp\.v(\d+)\+json")
def parse_version(accept: str, default: int = 1) -> int:
match = PATTERN.search(accept or "")
return int(match.group(1)) if match else default
# A few quick checks
print(parse_version("application/vnd.myapp.v2+json")) # 2
print(parse_version("application/json")) # 1 (default)
print(parse_version("application/vnd.myapp.v10+json")) # 10Wiring Media-Type Versioning into FastAPI
With the parser ready, plug it into a dependency that reads the Accept header. Endpoints stay clean and resource-focused while the version comes from content negotiation.
import re
from fastapi import FastAPI, Header, Depends
app = FastAPI()
PATTERN = re.compile(r"application/vnd\.myapp\.v(\d+)\+json")
def accept_version(accept: str = Header(default="")) -> int:
m = PATTERN.search(accept)
return int(m.group(1)) if m else 1
@app.get("/users/{user_id}")
def get_user(user_id: int, version: int = Depends(accept_version)):
if version >= 2:
return {"id": user_id, "first_name": "Ada", "last_name": "Lovelace"}
return {"id": user_id, "name": "Ada Lovelace"}Keeping Versions DRY with Transformers
Duplicating business logic per version rots fast. A cleaner pattern: compute the data once in a canonical internal shape, then run a small per-version transformer that adapts it to the contract each version promised.
This isolates contract differences in tiny, testable functions instead of forking your whole handler.
def canonical_user(user_id: int) -> dict:
return {"id": user_id, "first": "Ada", "last": "Lovelace"}
def to_v1(u: dict) -> dict:
return {"id": u["id"], "name": f"{u['first']} {u['last']}"}
def to_v2(u: dict) -> dict:
return {"id": u["id"], "first_name": u["first"], "last_name": u["last"]}
VERSIONS = {1: to_v1, 2: to_v2}
def render(user_id: int, version: int) -> dict:
return VERSIONS[version](canonical_user(user_id))
print(render(42, 1)) # {'id': 42, 'name': 'Ada Lovelace'}
print(render(42, 2)) # {'id': 42, 'first_name': 'Ada', 'last_name': 'Lovelace'}Caching, Vary, and Documentation Pitfalls
Header and media-type versioning have a sharp edge: caches. If a proxy caches /users/42 without knowing about your version header, a v1 client may receive a cached v2 body.
- Always send
Vary: X-API-Version(orVary: Acceptfor media-type versioning) so caches key on it. - URL versioning sidesteps this entirely because each version has a distinct URL.
- Header/media-type versions are also harder to see in the auto-generated
/docsschema, since the path looks identical across versions.
Choosing a Strategy and Deprecating Cleanly
Practical guidance for the bootcamp:
- URL versioning is the default for public REST APIs: visible, cache-friendly, easy to onboard.
- Header versioning suits internal services that want stable URLs and control both client and server.
- Media-type versioning fits hypermedia/REST purists; rare in practice.
Whatever you pick, ship a deprecation plan: announce a sunset date, return a Deprecation/Sunset header on old versions, and keep them alive long enough for clients to migrate.
from fastapi import APIRouter
from fastapi.responses import JSONResponse
v1 = APIRouter(prefix="/v1")
@v1.get("/users/{user_id}")
def get_user_v1(user_id: int):
body = {"id": user_id, "name": "Ada Lovelace"}
headers = {
"Deprecation": "true",
"Sunset": "Wed, 31 Dec 2025 23:59:59 GMT",
"Link": '</v2/users>; rel="successor-version"',
}
return JSONResponse(content=body, headers=headers)Quick Check: Picking the Right Versioning Approach
Apply what you learned about the trade-offs between the three strategies.
Recap: Versioning Without Breakage
You compared three ways to version a FastAPI API:
- URL path (
/v1/users): visible, cache-friendly, easy to route via per-versionAPIRouterprefixes. The default for public APIs. - Header (
X-API-Version): clean URLs, resolved with a dependency; rememberVaryfor caches. - Media-type (
Accept: application/vnd.myapp.vN+json): purest REST, parsed from the Accept header; rare and verbose.
Keep routes grouped by resource and mounted under version routers, isolate contract differences in small per-version transformers, and always pair a new version with a clear deprecation/sunset plan so clients upgrade smoothly.
よくある質問
「URL、ヘッダー、メディアタイプによるバージョニング」レッスンは無料ですか?
はい。「URL、ヘッダー、メディアタイプによるバージョニング」の完全なテキストはこのウェブで無料で読めます。インタラクティブに演習し(組み込みコードエディタと24時間対応のAIチューター)、FastAPI Backend Development Bootcampコースの残りをアンロックするには、CoddyKit PROにアップグレードしてください。 FastAPI Backend Development Bootcampコースには全4レッスンが含まれています。
「URL、ヘッダー、メディアタイプによるバージョニング」で何を学びますか?
バージョニング方式を比較し、クライアントが破壊的変更なくアップグレードできる、整理されたルートグループを実装します。 ブラウザで直接実行するハンズオンコードでFastAPI Backend Development Bootcampを演習し、24時間対応のAIチューターがレッスンを進める中での質問に答えます。
FastAPI Backend Development Bootcampを始めるのに経験は必要ですか?
事前経験は必要ありません。CoddyKitのFastAPI Backend Development Bootcampは初級者から上級者向けに構成されているため、ここから始めるか最初から始めて、自分のペースで進むことができます。 これはレッスン1/4です。
「URL、ヘッダー、メディアタイプによるバージョニング」レッスンにはどのくらい時間がかかりますか?
ほとんどのCoddyKitレッスンは約5~10分かかります。各レッスンはコンパクトでインタラクティブなので、着実に進歩し、ウェブとアプリ全体で正確に前回の場所から再開できます。
このFastAPI Backend Development Bootcampレッスンでコードを書いて実行できますか?
はい。すべてのFastAPI Backend Development Bootcampレッスンに組み込みコードエディタが含まれているため、ブラウザでリアルコードを書いて実行し、即座のAIフィードバックを取得できます。ローカル設定は不要です。
このコースのすべてのレッスン
- URL、ヘッダー、メディアタイプによるバージョニング
- 大規模環境におけるCursorとOffsetページネーション
- 動的なフィルタリングとソートパラメーター
- 安定したレスポンスエンベロープの設計