Exception Handlers

Without a handler, a Tortoise error reaching the top of a handler is a 500. The four below map the common ones onto the status codes they actually mean.

from sillo.record.exceptions import register_db_exception_handlers

app = SilloApp()
register_db_exception_handlers(app)

The four

Exception	Status	Body
DoesNotExist	404	{"error": "Not Found", "detail": "…"}
IntegrityError	409	{"error": "Conflict", "detail": "…"}
ValidationError	422	{"error": "Validation Error", "detail": "…"}
OperationalError	503	{"error": "Service Unavailable", "detail": "Database unavailable"}

DoesNotExist → 404

Raised by .get() when nothing matches.

post = await Post.get(id=post_id)     # 404 if there is no such post

Which makes the handler-level shape pleasantly short, no if post is None branch in every endpoint.

The alternative is get_or_none() and an explicit branch. Use that when absence is not a 404. When you want to return an empty object, or a different message.

IntegrityError → 409

Unique constraint violations, null constraint violations, foreign key violations.

409 is right for the duplicate case: the request was well-formed, and conflicts with the current state. It is less right for a null or FK violation, which is usually a bug rather than a conflict, but the exception type does not distinguish them, and 409 is the least-wrong single answer.

ValidationError → 422

Tortoise’s own field validation, a value too long for its column, a bad choice.

Most validation should never reach here. Request validation runs before the handler and produces field-level errors; this is the backstop for what gets past it.

OperationalError → 503

Connection refused, timeout, network partition.

503 tells a load balancer and a monitoring system that the instance is unhealthy rather than that the request was bad, which is what you want when the database is down: shed load, do not retry the same broken request as if it were the client’s fault.

Note this one does not include the exception text. A connection error can name hosts, ports and users, and none of that belongs in a public response body.

The other three do leak detail

{ "error": "Conflict", "detail": "UNIQUE constraint failed: users.email" }

That is the driver’s message, and it names your table and column. Sometimes useful in development, and on a public API it hands out schema details for free, and for a login or sign-up endpoint, a unique-violation message confirms whether an address is registered.

Override them in production:

from tortoise.exceptions import IntegrityError


async def handle_conflict(request, response, exc):
    logger.warning("integrity error", exc_info=exc)
    return response.json(
        {"error": "Conflict", "detail": "That value is already in use."},
        status_code=409,
    )


app.add_exception_handler(IntegrityError, handle_conflict)

Log the real message; return a generic one. register_db_exception_handlers is a sensible default, not a security review.

Registering selectively

The bundled function registers all four. To pick:

from tortoise.exceptions import DoesNotExist, OperationalError
from sillo.record.exceptions import handle_does_not_exist, handle_operational_error

app.add_exception_handler(DoesNotExist, handle_does_not_exist)
app.add_exception_handler(OperationalError, handle_operational_error)

Each handler is a plain async function taking (request, response, exc), the same signature as any other exception handler, so they compose with your own and can be wrapped.

Ordering

A handler registered for a subclass wins over one for its parent, so a specific handler added later does not need the general one removed:

app.add_exception_handler(IntegrityError, handle_conflict)
app.add_exception_handler(MyDuplicateEmailError, handle_duplicate_email)

What is not covered

• DoesNotExist from a legitimate lookup inside a service. A 404 is right for “the URL names a row that does not exist”; it is wrong when a background job cannot find a related record. Catch it there.
• Deadlocks and serialisation failures. These arrive as OperationalError and become a 503, when the correct answer is usually a retry.
• Timeouts under load, which look identical to a database being down.

The handlers are a floor. Anything with specific meaning in your application deserves its own.