feat(server): Gumroad webhook receiver + Postmark email (PR 2)
Wires the second source-adapter (Gumroad) plus the email delivery
that lets the server fulfill a sale end-to-end without operator
intervention.
Auth model: Gumroad doesn't HMAC the body, so we use their
recommended URL-secret pattern (?secret=...). Wrong/missing secret
returns 404 — no signal to a prober that the endpoint exists.
Webhook flow (server/app/routes/webhooks.py):
1. audit-log the raw payload (gumroad_events row) BEFORE anything
else, so a later failure leaves us replayable
2. parse via GumroadAdapter (server/app/adapters/gumroad.py)
3. mint_from_sale — UNIQUE(source, source_order_id) dedups
duplicate webhook retries
4. send the license email
5. mark gumroad_events.processed = true
Always returns 200 once auth passes. Non-2xx would trigger Gumroad's
3-day retry storm; we'd rather record the failure on the audit row
and replay manually after fixing whatever surfaced.
Product → tier mapping is per-source YAML at
server/config/products.yaml (lru_cached). Adding a SKU = edit yaml,
restart api. Unmapped product_id is an error on the audit row, not
a crash.
EmailService (server/app/email.py): provider-agnostic interface with
Postmark as the first implementation. When POSTMARK_TOKEN is unset
the factory returns LoggingEmailService instead, so the webhook
exercises end-to-end before Postmark is provisioned.
48 unit tests (was 21) including:
- Gumroad secret verify with constant-time compare
- Sale parsing: amount-in-cents, name fallback from email,
test=true tagging, missing-required fields, offer codes
- Product mapping lookups
- Email rendering text + HTML, HTML-escapes user input
- Postmark client via httpx.MockTransport (success and 4xx)
- Webhook end-to-end: secret check, audit log, idempotency on
retry, unmapped product, email failure keeps license
Smoke test (server/scripts/smoke.sh) extended to POST a synthetic
Ping payload, verify the row + audit log, prove wrong-secret is
rejected, prove duplicate sale_id stays one row.
SQLite-test compatibility:
- BigInteger primary key uses with_variant(Integer, "sqlite") since
SQLite only autoincrements INTEGER PRIMARY KEY.
- python-multipart pulled in for FastAPI Form parsing.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
@@ -16,6 +16,7 @@ from sqlalchemy import (
|
||||
BigInteger,
|
||||
DateTime,
|
||||
Index,
|
||||
Integer,
|
||||
Numeric,
|
||||
String,
|
||||
UniqueConstraint,
|
||||
@@ -29,6 +30,11 @@ from sqlalchemy.orm import Mapped, mapped_column
|
||||
# (SQLite for tests). Same Python interface either way.
|
||||
_JSON_TYPE = JSON().with_variant(JSONB(), "postgresql")
|
||||
|
||||
# SQLite only auto-increments INTEGER PRIMARY KEY (not BIGINT).
|
||||
# Postgres can autoincrement either, so the variant keeps the
|
||||
# production migration on BigInteger while tests use Integer.
|
||||
_PK_TYPE = BigInteger().with_variant(Integer(), "sqlite")
|
||||
|
||||
from app.db import Base
|
||||
|
||||
|
||||
@@ -77,7 +83,7 @@ class GumroadEvent(Base):
|
||||
|
||||
__tablename__ = "gumroad_events"
|
||||
|
||||
id: Mapped[int] = mapped_column(BigInteger, primary_key=True, autoincrement=True)
|
||||
id: Mapped[int] = mapped_column(_PK_TYPE, primary_key=True, autoincrement=True)
|
||||
received_at: Mapped[datetime] = mapped_column(DateTime(timezone=True), nullable=False, server_default=func.now())
|
||||
event_type: Mapped[str] = mapped_column(String, nullable=False)
|
||||
order_id: Mapped[Optional[str]] = mapped_column(String, nullable=True)
|
||||
|
||||
Reference in New Issue
Block a user