docs: ADR-012 — raw supabase-py over Prisma/SQLAlchemy for portfolio data layer

Co-authored-by: aguzererler <6199053+aguzererler@users.noreply.github.com>
This commit is contained in:
copilot-swe-agent[bot] 2026-03-20 10:48:40 +00:00
parent f1cabe7a4a
commit 7ea9866d1d
3 changed files with 126 additions and 2 deletions

View File

@ -0,0 +1,92 @@
---
type: decision
status: active
date: 2026-03-20
agent_author: "claude"
tags: [portfolio, database, supabase, orm, prisma]
related_files:
- tradingagents/portfolio/supabase_client.py
- tradingagents/portfolio/repository.py
- tradingagents/portfolio/migrations/001_initial_schema.sql
---
## Context
When designing the Portfolio Manager data layer (Phase 1), the question arose:
should we use an ORM (specifically **Prisma**) or keep the raw `supabase-py`
client that the scaffolding already plans to use?
The options considered were:
| Option | Description |
|--------|-------------|
| **Raw `supabase-py`** (chosen) | Direct Supabase PostgREST client, builder-pattern API |
| **Prisma Python** (`prisma-client-py`) | Code-generated type-safe ORM backed by Node.js |
| **SQLAlchemy** | Full ORM with Core + ORM layers, Alembic migrations |
## The Decision
**Use raw `supabase-py` without an ORM for the portfolio data layer.**
The data access layer (`supabase_client.py`) wraps the Supabase client directly.
Our own `Portfolio`, `Holding`, `Trade`, and `PortfolioSnapshot` dataclasses
provide the type-safety layer; serialisation is handled by `to_dict()` /
`from_dict()` on each model.
## Why Not Prisma
1. **Node.js runtime dependency**`prisma-client-py` uses Prisma's Node.js
engine at code-generation time. This adds a non-Python runtime requirement
to a Python-only project.
2. **Conflicts with Supabase's migration tooling** — the project already uses
Supabase's SQL migration files (`migrations/001_initial_schema.sql`) and the
Supabase dashboard for schema changes. Prisma's `prisma migrate` maintains
its own shadow database and migration state, creating two competing systems.
3. **Code generation build step** — every schema change requires running
`prisma generate` before the Python code works. This complicates CI, local
setup, and agent-driven development.
4. **Overkill for 4 tables** — the portfolio schema has exactly 4 tables with
straightforward CRUD. Prisma's relationship traversal and complex query
features offer no benefit here.
## Why Not SQLAlchemy
1. **Not using a local database** — the database is managed by Supabase (hosted
PostgreSQL). SQLAlchemy's connection-pooling and engine management are
designed for direct database connections, which bypass Supabase's PostgREST
API and Row Level Security.
2. **Extra dependency** — SQLAlchemy + Alembic would be significant new
dependencies for a non-DB-heavy app.
## Why Raw `supabase-py` Is Sufficient
- `supabase-py` provides a clean builder-pattern API:
`client.table("holdings").select("*").eq("portfolio_id", id).execute()`
- Our dataclasses already provide compile-time type safety and lossless
serialisation; the client only handles transport.
- Migrations are plain SQL files — readable, versionable, Supabase-native.
- `SupabaseClient` is a thin singleton wrapper that translates HTTP errors into
domain exceptions — this gives us the ORM-like error-handling benefit without
the complexity.
## Constraints
- **Do not** add an ORM dependency (`prisma-client-py`, `sqlalchemy`, `tortoise-orm`)
to `pyproject.toml` without revisiting this decision.
- **Do not** bypass `SupabaseClient` by importing `supabase` directly in other
modules — always go through `PortfolioRepository`.
- If the schema grows beyond ~10 tables or requires complex multi-table joins,
revisit this decision and consider SQLAlchemy Core (not the ORM layer) with
direct `asyncpg` connections.
## Actionable Rules
- All DB access goes through `PortfolioRepository``SupabaseClient`.
- Migrations are `.sql` files in `tradingagents/portfolio/migrations/`, run via
the Supabase SQL Editor or `supabase db push`.
- Type safety comes from dataclass `to_dict()` / `from_dict()` — not from a
code-generated ORM schema.

View File

@ -72,6 +72,32 @@ investment portfolio end-to-end. It performs the following actions in sequence:
The `report_path` column in the `portfolios` table points to the daily portfolio
subdirectory on disk: `reports/daily/{date}/portfolio/`.
### Data Access Layer: raw `supabase-py` (no ORM)
The Python code talks to Supabase through the raw `supabase-py` client — **no
ORM** (Prisma, SQLAlchemy, etc.) is used.
**Why not Prisma?**
- `prisma-client-py` requires a Node.js runtime for code generation — an
extra non-Python dependency in a Python-only project.
- Prisma's `prisma migrate` conflicts with Supabase's own SQL migration tooling
(we use `.sql` files in `tradingagents/portfolio/migrations/`).
- 4 tables with straightforward CRUD don't benefit from a code-generated ORM.
**Why not SQLAlchemy?**
- Supabase is accessed via PostgREST (HTTP API), not a direct TCP database
connection. SQLAlchemy is designed for direct connections and would bypass
Supabase's Row Level Security.
- Extra dependency overhead for a non-DB-heavy feature.
**`supabase-py` is sufficient because:**
- Its builder-pattern API (`client.table("holdings").select("*").eq(...)`)
covers all needed queries cleanly.
- Our own dataclasses handle type-safety via `to_dict()` / `from_dict()`.
- Plain SQL migration files are readable, versionable, and Supabase-native.
> Full rationale: `docs/agent/decisions/012-portfolio-no-orm.md`
---
## 5-Phase Workflow

View File

@ -1,9 +1,15 @@
"""Supabase database client for the Portfolio Manager.
Thin wrapper around ``supabase-py`` that:
Thin wrapper around ``supabase-py`` (no ORM) that:
- Provides a singleton connection (one client per process)
- Translates Supabase / HTTP errors into domain exceptions
- Converts raw DB rows into typed model instances
- Converts raw DB rows into typed model instances via ``Model.from_dict()``
**No ORM is used here by design** see
``docs/agent/decisions/012-portfolio-no-orm.md`` for the full rationale.
In short: ``supabase-py``'s builder-pattern API is sufficient for 4 tables;
Prisma and SQLAlchemy add build-step / runtime complexity that isn't warranted
for this non-DB-heavy feature.
Usage::