refactor(api): move from Postgres+RLS to single-engine SQLite
CI / Repo hygiene (pull_request) Successful in 2s
CI / Web (lint, typecheck, build) (pull_request) Successful in 13s
CI / Migrations reversible (pull_request) Successful in 6s
CI / API (lint, types, tests) (pull_request) Successful in 52s

Reverses a shipped, tested, merged decision (D4/PR #2) rather than building
on it — see docs/DECISIONS.md D15 for the full record: what was rejected
(Postgres as a second container; Postgres+PostGIS bundled inside the single
container via a supervisor), what this costs (no database-level RLS, no
PostGIS, procrastinate needs replacing — all stated as a concern before this
was decided, and reaffirmed anyway, which is the user's call to make about
their own instance).

The one invariant-critical consequence: isolation between users now rests
entirely on the repository-layer scope (db.py's `Scope.select()`), not two
layers. CLAUDE.md's invariant #4 is revised accordingly. This is not a
downgrade-and-hope — `Scope` is built so an unfiltered query against a
user-owned table is structurally harder to write than a scoped one (there is
no method on `Scope` that returns one), and
tests/test_auth.py::test_scoped_session_blocks_cross_user_reads replaces the
old RLS proof with the same empirical standard: it doesn't trust the query
builder filters correctly because the code reads correctly, it registers two
real users and checks.
test_unscoped_session_can_see_every_user_when_misused is the deliberately
alarming companion — it demonstrates exactly what a reviewer must now catch,
since nothing else will.

Six real, non-obvious SQLite behaviours found and fixed by actually running
this against a real file, not assumed from docs:

- Foreign keys, ON DELETE CASCADE included, are OFF by default per
  connection — deleting a user silently left orphaned sessions/api_tokens,
  no error either way. Fixed with PRAGMA foreign_keys=ON on every connect.
- Transactions default to DEFERRED, which only takes a write lock on the
  first actual write — a real check-then-act race for invite redemption
  (two concurrent redemptions could both read used_count < max_uses as true
  before either commits). Fixed by disabling the driver's implicit BEGIN and
  issuing BEGIN IMMEDIATE ourselves — SQLAlchemy's own documented recipe for
  this, not improvised.
- DateTime(timezone=True) does NOT round-trip tzinfo on SQLite — a tz-aware
  datetime goes in, a naive one comes back out, and every
  `expires_at < datetime.now(UTC)` comparison in auth/service.py then raises
  TypeError. Fixed once at the Base level with a UTCDateTime TypeDecorator
  rather than per-column.
- Uuid(as_uuid=True) stores as 32-char hex with NO hyphens on SQLite, not
  str(uuid)'s hyphenated form. A test fixture that raw-inserted the
  hyphenated form left rows the ORM's own later UPDATE (via
  invite.used_count += 1's autoflush) could never match by primary key,
  updating zero rows and raising StaleDataError. Fixed by using .hex to
  match exactly what the ORM itself writes.
- BEGIN IMMEDIATE applies to every transaction, reads included — a
  long-lived test fixture that autobegins a transaction via a bare read and
  never explicitly closes it holds SQLite's exclusive write lock for the
  rest of the test, and a later scoped_session() call fails with "database
  is locked". Not an app-code bug (every real session block closes cleanly
  on exit), but real enough to document since the next person writing a
  test against the db_auth fixture will hit it too.
- Python's sqlite3 module deprecates its own implicit datetime adapter as of
  3.12 — silent today, warns on every raw-SQL datetime bind. Only ever hit
  test fixture code (the ORM path never uses it, confirmed by running the
  ORM-only health test with warnings promoted to errors and it stayed
  clean); fixed there with an explicit .isoformat() rather than left for a
  future Python version to turn into a real failure.

Also, since with_for_update() silently no-ops on SQLite (confirmed —
SQLAlchemy emits no SQL for it, no error either) rather than actually
locking anything: removed it from register()'s invite-redemption query and
corrected the comment to attribute the concurrency guarantee to BEGIN
IMMEDIATE, where it now actually lives.

One PR, not several, for the same reason PR #2 was: the migration, the
models, db.py, and the docs recording why are five views of one decision —
splitting them wouldn't make review easier, just disconnected. 552
insertions / 548 deletions across 17 files, most of it necessarily touching
what PR #2 shipped rather than net-new code.

Deliberately deferred, not solved here: PostGIS's replacement for spatial
storage, procrastinate's replacement for background jobs, and the
EXCLUDE USING gist constraint's replacement for component_installs — none
of those tables exist yet (Phase 1-2), so none of it is broken, and
docs/DECISIONS.md D15 records exactly what each future phase needs to
decide before it can be built. .gitea/workflows/deploy pipeline (PR #4,
built for the old 3-container Postgres compose stack) was closed as
superseded rather than merged; the single-container image build is
follow-up work, not part of this change.

Verified: ruff check, ruff format --check, and mypy --strict all clean.
13/13 pytest passing against a real SQLite file, including with
DeprecationWarning promoted to an error (confirms the sqlite3 adapter
deprecation fix actually holds, not just that it's quiet by default). Full
alembic upgrade -> downgrade -1 -> upgrade cycle run clean. alembic check
clean with no include_object filter needed at all now (SQLite starts with
nothing but what our own migrations create — no PostGIS/TIGER noise to
filter out in the first place). CI's exact migration command sequence
reproduced locally end to end before touching the workflow file.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
2026-09-21 15:42:44 -04:00
co-authored by Claude Opus 5
parent 1832059e03
commit e7392a5723
17 changed files with 548 additions and 544 deletions
+22 -155
View File
@@ -1,21 +1,18 @@
"""Baseline: identity tables, two runtime DB roles, RLS policies.
"""Baseline: identity tables (users, invites, sessions, api_tokens).
This migration creates the schema AND the two application-facing Postgres roles it depends on
(velodrome_app, velodrome_auth) — see db.py and auth/service.py module docstrings for the full
rationale. It runs as a privileged/owner connection (VELODROME_DATABASE_URL_MIGRATE), which is
why it's able to CREATE ROLE and GRANT at all; neither of the two roles it creates could do this
to itself.
No roles, no RLS, no GRANTs — SQLite has none of those concepts. Isolation between users is
enforced entirely at the application layer now; see db.py and CLAUDE.md's invariant #4, and
docs/DECISIONS.md D15 for the full story of why this migration looks nothing like the Postgres
version it replaced.
Revision ID: 0001
Revises:
Create Date: 2026-09-21
"""
import os
from collections.abc import Sequence
import sqlalchemy as sa
from sqlalchemy.dialects import postgresql as pg
from alembic import op
@@ -25,77 +22,10 @@ branch_labels: str | Sequence[str] | None = None
depends_on: str | Sequence[str] | None = None
def _require_password(env_var: str) -> str:
value = os.environ.get(env_var)
if not value:
raise RuntimeError(
f"{env_var} must be set before running this migration — see deploy/.env.example. "
"There is no default: these are the passwords for real runtime database roles."
)
return value
def _dollar_quoted(password: str) -> str:
"""Safely embed a password literal in DDL.
CREATE ROLE's PASSWORD clause is DDL, not DML — it does NOT accept bind parameters over the
wire (Postgres rejects `PASSWORD $1` with a syntax error; ask how we know). Dollar-quoting
sidesteps manual escaping entirely rather than hand-rolling quote-doubling, which is easy to
get subtly wrong for a password an operator chose.
"""
tag = "$velodrome_pw$"
if tag in password:
raise RuntimeError(
f"password must not contain the literal sequence {tag!r} — pick a different one"
)
return f"{tag}{password}{tag}"
def upgrade() -> None:
app_password = _require_password("VELODROME_DB_APP_PASSWORD")
auth_password = _require_password("VELODROME_DB_AUTH_PASSWORD")
# --- Runtime roles -------------------------------------------------------------------
# velodrome_app: NOBYPASSRLS — every request-scoped query after identity is established.
# velodrome_auth: BYPASSRLS — ONLY auth/service.py's pre-identity lookups. See db.py.
# Roles are cluster-wide in Postgres, not per-database — if this migration ever runs
# against a second database sharing the same cluster (exactly what a local dev + test setup
# commonly looks like), a plain CREATE ROLE fails with "role already exists" even though
# THIS database has never seen the migration before. Postgres has no CREATE ROLE IF NOT
# EXISTS, so the standard idiom is catching duplicate_object in a DO block. ALTER ROLE
# afterwards keeps the password in sync with the current env var either way, rather than
# silently keeping whatever password the role happened to be created with previously.
op.execute(
f"""
DO $$
BEGIN
CREATE ROLE velodrome_app LOGIN PASSWORD {_dollar_quoted(app_password)}
NOSUPERUSER NOCREATEDB NOCREATEROLE NOREPLICATION NOBYPASSRLS;
EXCEPTION WHEN duplicate_object THEN
ALTER ROLE velodrome_app WITH LOGIN PASSWORD {_dollar_quoted(app_password)}
NOSUPERUSER NOCREATEDB NOCREATEROLE NOREPLICATION NOBYPASSRLS;
END
$$;
"""
)
op.execute(
f"""
DO $$
BEGIN
CREATE ROLE velodrome_auth LOGIN PASSWORD {_dollar_quoted(auth_password)}
NOSUPERUSER NOCREATEDB NOCREATEROLE NOREPLICATION BYPASSRLS;
EXCEPTION WHEN duplicate_object THEN
ALTER ROLE velodrome_auth WITH LOGIN PASSWORD {_dollar_quoted(auth_password)}
NOSUPERUSER NOCREATEDB NOCREATEROLE NOREPLICATION BYPASSRLS;
END
$$;
"""
)
# --- Tables ----------------------------------------------------------------------------
op.create_table(
"users",
sa.Column("id", pg.UUID(as_uuid=True), primary_key=True),
sa.Column("id", sa.Uuid(as_uuid=True), primary_key=True),
sa.Column("email", sa.String(320), nullable=False, unique=True),
sa.Column("display_name", sa.String(200), nullable=False),
sa.Column("password_hash", sa.Text(), nullable=False),
@@ -103,18 +33,16 @@ def upgrade() -> None:
sa.Column("timezone", sa.String(64), nullable=False, server_default="UTC"),
sa.Column("unit_system", sa.String(10), nullable=False, server_default="imperial"),
sa.Column("is_active", sa.Boolean(), nullable=False, server_default=sa.true()),
sa.Column(
"created_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.func.now()
),
sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
)
op.create_table(
"invites",
sa.Column("id", pg.UUID(as_uuid=True), primary_key=True),
sa.Column("id", sa.Uuid(as_uuid=True), primary_key=True),
sa.Column("code_hash", sa.LargeBinary(32), nullable=False, unique=True),
sa.Column(
"created_by",
pg.UUID(as_uuid=True),
sa.Uuid(as_uuid=True),
sa.ForeignKey("users.id"),
nullable=False,
),
@@ -128,23 +56,20 @@ def upgrade() -> None:
op.create_table(
"sessions",
sa.Column("id", pg.UUID(as_uuid=True), primary_key=True),
sa.Column("id", sa.Uuid(as_uuid=True), primary_key=True),
sa.Column(
"user_id",
pg.UUID(as_uuid=True),
sa.Uuid(as_uuid=True),
sa.ForeignKey("users.id", ondelete="CASCADE"),
nullable=False,
),
sa.Column("token_hash", sa.LargeBinary(32), nullable=False, unique=True),
sa.Column("client", sa.String(20), nullable=False, server_default="web"),
sa.Column("user_agent", sa.Text(), nullable=True),
sa.Column("ip", pg.INET(), nullable=True),
sa.Column(
"created_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.func.now()
),
sa.Column(
"last_seen_at", sa.DateTime(timezone=True), nullable=False, server_default=sa.func.now()
),
# Plain string — SQLite has no INET type, and nothing queries the address's structure.
sa.Column("ip", sa.String(45), nullable=True),
sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
sa.Column("last_seen_at", sa.DateTime(timezone=True), nullable=False),
sa.Column("expires_at", sa.DateTime(timezone=True), nullable=False),
sa.Column("revoked_at", sa.DateTime(timezone=True), nullable=True),
)
@@ -152,87 +77,29 @@ def upgrade() -> None:
op.create_table(
"api_tokens",
sa.Column("id", pg.UUID(as_uuid=True), primary_key=True),
sa.Column("id", sa.Uuid(as_uuid=True), primary_key=True),
sa.Column(
"user_id",
pg.UUID(as_uuid=True),
sa.Uuid(as_uuid=True),
sa.ForeignKey("users.id", ondelete="CASCADE"),
nullable=False,
),
sa.Column("name", sa.String(200), nullable=False),
sa.Column("token_hash", sa.LargeBinary(32), nullable=False, unique=True),
sa.Column(
"scopes",
pg.ARRAY(sa.String()),
nullable=False,
server_default="{}",
),
# JSON-encoded TEXT — SQLite has no array type; see models/identity.py.
sa.Column("scopes", sa.JSON(), nullable=False, server_default="[]"),
sa.Column("last_used_at", sa.DateTime(timezone=True), nullable=True),
sa.Column("expires_at", sa.DateTime(timezone=True), nullable=True),
sa.Column("revoked_at", sa.DateTime(timezone=True), nullable=True),
)
op.create_index("ix_api_tokens_user_id", "api_tokens", ["user_id"])
# --- Grants ------------------------------------------------------------------------------
# velodrome_auth needs full read/write on these four tables — it's the only thing that ever
# creates a user, a session, or redeems an invite. velodrome_app needs the same grants
# because RLS policies (below) restrict WHICH ROWS it sees, not whether the underlying
# privilege exists — GRANT and POLICY are two independent layers, both required.
# Deliberately NOT granting TRUNCATE: it's a whole-table operation that RLS cannot filter
# (Postgres RLS policies do not apply to TRUNCATE at all), so granting it to velodrome_app
# would let any bug in ordinary request-handling code wipe an entire table in one statement,
# defeating the isolation these policies exist to provide. Neither runtime role needs it —
# tests use DELETE for fixture cleanup instead (see tests/conftest.py).
for table in ("users", "invites", "sessions", "api_tokens"):
op.execute(f"GRANT SELECT, INSERT, UPDATE, DELETE ON {table} TO velodrome_app")
op.execute(f"GRANT SELECT, INSERT, UPDATE, DELETE ON {table} TO velodrome_auth")
# --- Row-level security ------------------------------------------------------------------
# Policies below apply to velodrome_app only — velodrome_auth has BYPASSRLS and ignores them
# entirely, by design (see module docstring). current_setting('app.user_id', true) returns
# NULL when unset, which makes every policy below deny-by-default for an unscoped connection.
op.execute("ALTER TABLE users ENABLE ROW LEVEL SECURITY")
op.execute(
"CREATE POLICY own_row ON users FOR ALL "
"USING (id = current_setting('app.user_id', true)::uuid) "
"WITH CHECK (id = current_setting('app.user_id', true)::uuid)"
)
op.execute("ALTER TABLE sessions ENABLE ROW LEVEL SECURITY")
op.execute(
"CREATE POLICY own_rows ON sessions FOR ALL "
"USING (user_id = current_setting('app.user_id', true)::uuid) "
"WITH CHECK (user_id = current_setting('app.user_id', true)::uuid)"
)
op.execute("ALTER TABLE api_tokens ENABLE ROW LEVEL SECURITY")
op.execute(
"CREATE POLICY own_rows ON api_tokens FOR ALL "
"USING (user_id = current_setting('app.user_id', true)::uuid) "
"WITH CHECK (user_id = current_setting('app.user_id', true)::uuid)"
)
op.execute("ALTER TABLE invites ENABLE ROW LEVEL SECURITY")
op.execute(
"CREATE POLICY own_rows ON invites FOR ALL "
"USING (created_by = current_setting('app.user_id', true)::uuid) "
"WITH CHECK (created_by = current_setting('app.user_id', true)::uuid)"
)
def downgrade() -> None:
for table in ("invites", "api_tokens", "sessions", "users"):
op.execute(f"DROP POLICY IF EXISTS own_rows ON {table}")
op.execute(f"DROP POLICY IF EXISTS own_row ON {table}")
# Children before parents (FK order).
# Children before parents (FK order) — matters even with ON DELETE CASCADE enforced (see
# db.py's PRAGMA foreign_keys=ON), since dropping a table Postgres-style doesn't rely on that
# at all; SQLite's DROP TABLE has no dependency ordering of its own to lean on.
op.drop_table("api_tokens")
op.drop_table("sessions")
op.drop_table("invites")
op.drop_table("users")
# Dropping the tables drops the GRANTs that referenced them; the roles themselves remain
# until explicitly dropped here.
op.execute("DROP ROLE IF EXISTS velodrome_app")
op.execute("DROP ROLE IF EXISTS velodrome_auth")