xlambda-postgres

Managed PostgreSQL: provisioning, sub-users, backups, and a get_pool() convenience.

pip install xlambda-postgres          # provisioning only
pip install "xlambda-postgres[pool]"  # plus psycopg, for get_pool()

Python port of @xlambda-tech/postgres.

Provisioning

from xlambda.postgres import PostgresClient

client = PostgresClient(api_key=os.environ["XLAMBDA_API_KEY"])

db = client.databases.create(project_id="proj_123", name="app", tier="starter")
db["connectionString"]   # shown here and on rotate/reveal only

create(), rotate_password(), and reveal_connection() return a DatabaseHandle — a plain dict subclass, so every response field reads normally, with pool constructors added on top:

pool = db.get_pool()                  # port 6432, PgBouncer transaction pooling
migrations = db.get_session_pool()    # port 6433, direct session

async_pool = db.get_async_pool()      # psycopg_pool.AsyncConnectionPool

Which port matters. 6432 is transaction-pooled and is what steady-state app traffic should use; its connection string carries ?pgbouncer=true, which disables client-side prepared statements because transaction pooling breaks them. Use 6433 for migrations, DDL, advisory locks, and anything else needing a real session.

reveal_connection() discloses the current password rather than changing it. It 404s for instances that predate the feature or were created without CONSOLE_CREDENTIAL_ENCRYPTION_KEY configured — rotate to recover in that case.

Sub-users and backups

user = client.databases.users(db["id"]).create(permission="readonly", expires_at="2027-01-01T00:00:00Z")
user["password"]     # shown once

backups = client.databases.backups(db["id"])
backups.create()
backups.restore("bkp_123")   # overwrites all current data; cannot be undone

What this doesn't wrap

The dashboard's own SQL console. Running queries isn't something the REST API mediates — there's no app-facing /v1/databases/:id/query route — so once a database is provisioned you talk to it with psycopg through the handle above. get_pool() is a thin ergonomic wrapper around psycopg_pool, not a reimplementation of it, and psycopg stays an optional extra so provisioning-only callers don't install a driver they never use.

Async

from xlambda.postgres import AsyncPostgresClient

async with AsyncPostgresClient(api_key=...) as client:
    db = await client.databases.create(project_id="proj_123", name="app")

Conventions

Arguments are snake_case; responses are the API's own camelCase, returned as plain dicts typed by TypedDict. Byte counts that are bigints server-side (storageQuotaBytes, storageBytes, backup sizeBytes) arrive as strings rather than losing precision in JSON. See xlambda-core.