fix: normalize async Postgres URLs for alembic

[lkshrk]

HUMAN:

Reproduced against a self-hosted deployment on Kubernetes with a CloudNativePG Postgres and AUTOMATION_DB_URL=postgresql+asyncpg://…: alembic upgrade head fails with MissingGreenlet, and hand-rewriting the URL to postgresql+pg8000:// makes it pass. Ran uv run pytest tests/test_db.py and pre-commit locally on this branch.

AGENT:

Problem

AUTOMATION_DB_URL is the documented way to point a self-hosted deployment at a
database, and the Postgres form of it is postgresql+asyncpg://… — that is what
the agent-canvas Helm chart sets, and what db.py expects for the application
engine.

Alembic cannot use it. migrations/env.py get_engine() reads
AUTOMATION_DB_URL from the environment and passes it straight to the sync
sqlalchemy.create_engine(), normalizing only the SQLite prefix:

if DB_URL:
    url = DB_URL
    if url.startswith("sqlite+aiosqlite"):
        url = url.replace("sqlite+aiosqlite", "sqlite", 1)
    return create_engine(url, pool_pre_ping=True)

With an asyncpg URL, alembic upgrade head fails:

sqlalchemy.exc.MissingGreenlet: greenlet_spawn has not been called;
can't call await_only() here.

Rewriting the URL by hand to postgresql+pg8000://… makes the same command
succeed, which is the shape of the fix: pg8000 is already a dependency and is
already the driver env.py uses for its host/port and Cloud SQL paths.

Fix

• db.py: normalize_sqlite_url_for_alembic() becomes
  normalize_url_for_alembic(), driven by a small ALEMBIC_SYNC_DRIVERS map —
  sqlite+aiosqlite → sqlite (unchanged behaviour) and
  postgresql+asyncpg → postgresql+pg8000. Only a leading driver prefix is
  rewritten, so a URL whose password happens to contain the driver name is left
  alone. Every other URL passes through untouched.
• migrations/env.py: get_engine() and run_migrations_offline() both run a
  configured AUTOMATION_DB_URL through the normalizer. A URL that resolves to
  pg8000 now also gets _build_pg8000_connect_args(DB_SSL_MODE), so
  AUTOMATION_DB_SSL_MODE finally applies to the URL path the same way it
  already applies to the host/port path.
• app.py: uses the renamed helper when it sets sqlalchemy.url.

The helper is renamed rather than aliased, because an alias could not stay
backward compatible: the old name's contract was "asyncpg URLs are unchanged",
which is precisely the bug. Both call sites and the tests are updated; nothing
outside this repo imports it.

Verification

• uv run pytest tests/test_db.py — 30 passed. The normalizer suite gained
  cases for asyncpg → pg8000, prefix-only rewriting, a non-async Postgres
  driver, and the empty string.
• Full unit suite and pre-commit (ruff format, ruff lint, pycodestyle,
  pyright) clean on the changed files.
• AUTOMATION_DB_URL=postgresql+asyncpg://… alembic upgrade head --sql renders
  Postgres DDL, and the previously failing online path now builds a pg8000
  engine instead of an asyncpg one.

migrations/env.py executes run_migrations_online() at import time, so it has
no direct unit test; the behaviour is covered through the shared helper.

No schema change, no new dependency, no behaviour change for SQLite or for
deployments configured via AUTOMATION_DB_HOST/AUTOMATION_DB_PORT.

[all-hands-bot]

👋 This PR needs a couple of things fixed before OpenHands can review it:

• the PR description's HUMAN: section needs at least 20 characters describing what you tested, not just the template placeholder

Push an update once this is addressed and this check re-runs automatically.

This is an automated check - no AI was used to generate this comment.