Supabase

The Supabase store uses the @supabase/supabase-js client to write failures and detect patterns via PostgREST.

Free tier: 500 MB database, unlimited projects (paused after 1 week of inactivity).

Installation

bun add -D @flaky-tests/store-supabase
# or: npm install -D @flaky-tests/store-supabase

Setup

1. Create a Supabase project at supabase.com

2. Create the tables — go to the SQL editor and run:

   create table if not exists runs (
     id text primary key,
     started_at timestamptz not null,
     finished_at timestamptz,
     exit_code integer
   );
   
   create table if not exists failures (
     id text primary key,
     run_id text not null references runs(id),
     test_file text not null,
     test_name text not null,
     failure_kind text not null,
     error_message text,
     error_stack text,
     failed_at timestamptz not null
   );

3. Get your project URL and anon key from Settings → API

4. Add to your environment

   SUPABASE_URL=https://abcdefghijklmnop.supabase.co
   SUPABASE_KEY=eyJhbGci...

Usage

import { SupabaseStore } from '@flaky-tests/store-supabase'

const store = new SupabaseStore({
  url: process.env.SUPABASE_URL!,
  key: process.env.SUPABASE_KEY!,
})

Constructor options

interface SupabaseStoreOptions {
  /** Your Supabase project URL */
  url: string
  /** Your anon or service role key */
  key: string
  /** Prefix for `{prefix}_runs` / `{prefix}_failures` table names. Default: `flaky_test`. */
  tablePrefix?: string
  /** Retry tuning for read methods. Defaults: 3 attempts, 100ms base delay. */
  retry?: { attempts?: number; baseMs?: number }
}

Error handling, retries, and cancellation

Driver errors are wrapped in StoreError with cause preserved. Read methods auto-retry transient failures (network transport errors, HTTP 5xx) with exponential backoff and jitter; writes are not retried (no idempotency key). Read methods accept an optional signal?: AbortSignal, forwarded to postgrest-js via native .abortSignal() for mid-flight cancellation.

Note

Supabase’s JS client does not support DDL. Tables must be created manually via the SQL editor (see setup above). store.migrate() performs a connectivity check only and throws StoreError if the configured tables are missing. Table names honor tablePrefix: if you set tablePrefix: 'my_app', create my_app_runs and my_app_failures instead of the defaults shown in setup.

CLI usage

FLAKY_TESTS_STORE=supabase \
  SUPABASE_URL=https://... \
  SUPABASE_KEY=eyJhbGci... \
  bunx @flaky-tests/core

Limitations

Caution

Pattern detection makes two round-trip fetches. PostgREST (the layer Supabase uses for the JS client) doesn’t support the conditional aggregates that getNewPatterns needs. The store fetches recent and prior failures separately and groups them in JavaScript. For typical failure volumes this is fine, but it’s less efficient than a direct Postgres connection.

If performance is a concern, use the Postgres store with a Supabase connection string instead.

Using the service role key

The anon key works with default RLS policies disabled. If you have RLS enabled on your tables, use the service role key instead (keep it in server-side secrets only, never expose to browsers).

SUPABASE_KEY=your-service-role-key  # from Settings → API → service_role