Files
accounted/.github/workflows/test-pg-real.yml
T
Jakob Wennberg 325c827322 test(mcp): run the tools against a real PostgREST, not a fake supabase (#1983)
All 100 files in extensions/general/mcp-server/__tests__ fake supabase.
query-journal.test.ts says out loud that its query chain is "exercised by the
live MCP smoke test", and no such test exists in CI. So the PostgREST grammar
of 157 tools, every .select() column string, every resource embed, every
or=(...) form, is gated by nothing and fails first in production.

pg-real cannot cover this: it holds a pg Pool and writes SQL, and none of that
grammar is resolved by Postgres. It is resolved by PostgREST at request time.

Adds a tool-pg vitest project, a docker-compose stack, a reset script that
replays every migration the way the pg-real CI job does, and a CI job.

The first sweep covers 74 read tools and finds no malformed query, across 87
real requests. That number is honest rather than impressive: with an empty
argument set many tools bail before querying. Per-tool fixtures are what
deepen it, and this harness is what makes writing them worth the effort.

Includes a self-test that injects a bad column and asserts the harness detects
it. That is not ceremony. It caught this file passing green while exercising
nothing, twice: once locally where supabase-js prefixes /rest/v1 onto a bare
PostgREST that does not serve it, and once on CI where Node 20 has no native
WebSocket, so every client construction threw and was swallowed by the
per-tool catch as a domain refusal. The client is now built once outside that
catch, the proof-of-life assertion counts real requests instead of being
trivially satisfiable, and realtime gets an inert transport.

Also excludes .next from all three vitest projects. These projects override
vitest's default excludes, so a local `npm run build` leaves a traced copy of
the repo that gets collected as a second set of test files.

Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-27 18:11:41 +02:00

300 lines
12 KiB
YAML

name: pg-real tests
on: [pull_request]
# Neither job writes anything back: they read the repo, stand up a throwaway
# Postgres, and run tests. Without this block both inherit the repository's
# default token permissions, which are broader than that.
permissions:
contents: read
concurrency:
group: pg-real-${{ github.ref }}
cancel-in-progress: true
jobs:
coverage-gate:
# Enforces the database.md rule: a migration touching a trigger/RPC/RLS/
# DEFERRABLE must come with a *.pg.test.ts change. Previously instruction-
# only. Escape hatch: `-- pg-test: covered-by <path>` / `-- pg-test: skip
# (<reason>)` comments inside the migration.
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
# Full history so the merge-base with the PR base branch exists.
fetch-depth: 0
persist-credentials: false
- uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
with:
node-version: 20
- name: Require pg-real coverage for trigger/RPC/RLS migrations
env:
PG_GATE_BASE: origin/${{ github.base_ref }}
run: node scripts/check-pg-test-coverage.mjs
# Proves that an EXISTING database survives the PR's migrations, which the
# pg-real job below cannot: it applies every migration to an empty database,
# so a NOT NULL, a CHECK, a unique index or a backfill passes trivially
# against zero rows and can still fail (or silently corrupt) on production.
#
# Shape: schema at the merge-base -> seed real rows -> apply ONLY the new
# migrations -> assert the rows are intact. A PR that adds no migration skips
# straight past the apply step and costs one cheap no-op run.
pg-upgrade:
runs-on: ubuntu-latest
services:
postgres:
image: supabase/postgres:15.8.1.060
env:
POSTGRES_PASSWORD: postgres
ports:
- 5432:5432
options: >-
--health-cmd "pg_isready -U postgres"
--health-interval 5s
--health-timeout 5s
--health-retries 20
env:
DATABASE_URL: postgresql://postgres:postgres@localhost:5432/postgres
PGPASSWORD: postgres
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
# Full history so the merge-base with the PR base branch exists.
fetch-depth: 0
persist-credentials: false
- name: Install psql client
run: sudo apt-get update && sudo apt-get install -y --no-install-recommends postgresql-client
- name: Identify the migrations this PR adds
id: newmig
env:
BASE_REF: origin/${{ github.base_ref }}
run: |
set -euo pipefail
MERGE_BASE=$(git merge-base "$BASE_REF" HEAD)
echo "merge_base=$MERGE_BASE" >> "$GITHUB_OUTPUT"
echo "Merge base: $MERGE_BASE"
# Migrations present at HEAD but not at the merge base. Uses the git
# tree, not the filesystem, so a rebase or a merge commit cannot make
# an already-shipped migration look new.
git ls-tree -r --name-only HEAD -- supabase/migrations \
| grep '\.sql$' | sort > /tmp/head-migrations.txt
git ls-tree -r --name-only "$MERGE_BASE" -- supabase/migrations \
| grep '\.sql$' | sort > /tmp/base-migrations.txt
comm -23 /tmp/head-migrations.txt /tmp/base-migrations.txt > /tmp/new-migrations.txt
COUNT=$(wc -l < /tmp/new-migrations.txt | tr -d ' ')
echo "count=$COUNT" >> "$GITHUB_OUTPUT"
echo "New migrations ($COUNT):"
cat /tmp/new-migrations.txt
- name: Bootstrap storage schema
if: steps.newmig.outputs.count != '0'
run: psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -q -f tests/pg/bootstrap.sql
- name: Apply the schema as it stands at the merge base
if: steps.newmig.outputs.count != '0'
env:
MERGE_BASE: ${{ steps.newmig.outputs.merge_base }}
run: |
set -euo pipefail
# Read each migration out of the merge-base tree rather than the
# working tree: a PR that EDITS a shipped migration (forbidden, but
# this job must not be the thing that hides it) still gets the
# original applied here, so the edit shows up as a failure below.
while read -r f; do
echo "Applying (base) $f"
git show "$MERGE_BASE:$f" | psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -q
done < /tmp/base-migrations.txt
- name: Seed a real company with posted verifikat
if: steps.newmig.outputs.count != '0'
run: psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -q -f tests/pg/upgrade/seed.sql
- name: Apply ONLY the new migrations, against existing data
if: steps.newmig.outputs.count != '0'
run: |
set -euo pipefail
while read -r f; do
echo "Applying (new) $f"
psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -q -f "$f"
done < /tmp/new-migrations.txt
- name: Assert the existing data survived
if: steps.newmig.outputs.count != '0'
run: psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -f tests/pg/upgrade/assert.sql
- name: No migrations in this PR
if: steps.newmig.outputs.count == '0'
run: echo "This PR adds no migration; nothing to upgrade-test."
pg-real:
runs-on: ubuntu-latest
services:
postgres:
# Supabase image ships the auth schema, auth.uid(), and the extensions
# (uuid-ossp, pg_cron, btree_gist, vector) this repo's migrations need.
# Plain postgres:15 would require manual bootstrap SQL.
image: supabase/postgres:15.8.1.060
env:
POSTGRES_PASSWORD: postgres
ports:
- 5432:5432
options: >-
--health-cmd "pg_isready -U postgres"
--health-interval 5s
--health-timeout 5s
--health-retries 20
env:
DATABASE_URL: postgresql://postgres:postgres@localhost:5432/postgres
PGPASSWORD: postgres
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
persist-credentials: false
# Deliberately no `cache: npm` here. This workflow runs on pull_request
# only, so the cache is never written on main, and GitHub scopes caches
# by ref: every PR uploaded its own 284 MB copy under an identical key
# that no other PR could ever restore. Fourteen dead copies (4 GB, 40% of
# the repo quota) accumulated in a single day. If this is ever worth
# caching again, it has to be actions/cache/save on main plus
# actions/cache/restore here, which is the only shape that gets hits.
- uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
with:
node-version: 20
- run: npm ci
- name: Install psql client
run: sudo apt-get update && sudo apt-get install -y --no-install-recommends postgresql-client
- name: Bootstrap storage schema
# The supabase/postgres image ships a partial storage schema; the rest
# is provisioned by the storage-api service at runtime, which we do
# not run in CI. This aligns the schema with what migrations expect.
run: psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -q -f tests/pg/bootstrap.sql
- name: Apply migrations
run: |
set -euo pipefail
shopt -s nullglob
files=(supabase/migrations/*.sql)
if [ ${#files[@]} -eq 0 ]; then
echo "No migration files found"
exit 1
fi
for f in "${files[@]}"; do
echo "Applying $f"
psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -q -f "$f"
done
- run: npm run test:pg
tool-pg:
# MCP tools driven through a REAL supabase-js client against a REAL
# PostgREST. This is NOT a duplicate of the pg-real job: that one holds a
# `pg` Pool and writes SQL, which cannot see the half of a tool that
# PostgREST resolves (the .select() column strings, the resource embeds,
# the or=(...) grammar). Before this job, all 100 files in
# extensions/general/mcp-server/__tests__ faked supabase and nothing in CI
# exercised that surface.
runs-on: ubuntu-latest
services:
postgres:
image: supabase/postgres:15.8.1.060
env:
POSTGRES_PASSWORD: postgres
ports:
- 5432:5432
options: >-
--health-cmd "pg_isready -U postgres"
--health-interval 5s
--health-timeout 5s
--health-retries 20
env:
DATABASE_URL: postgresql://postgres:postgres@localhost:5432/postgres
TOOL_PG_DATABASE_URL: postgresql://postgres:postgres@localhost:5432/postgres
TOOL_PG_REST_URL: http://127.0.0.1:3000
PGPASSWORD: postgres
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
persist-credentials: false
- uses: actions/setup-node@249970729cb0ef3589644e2896645e5dc5ba9c38 # v6
with:
node-version: 20
- run: npm ci
- name: Install psql client
run: sudo apt-get update && sudo apt-get install -y --no-install-recommends postgresql-client
- name: Bootstrap storage schema
run: psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -q -f tests/pg/bootstrap.sql
- name: Default privileges for the supabase roles
# Without these PostgREST answers every request with 42501, because the
# grants have to exist before the migrations create ~400 tables.
run: |
psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -q -c "
GRANT USAGE ON SCHEMA public TO postgres, anon, authenticated, service_role;
ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT ALL ON TABLES TO postgres, anon, authenticated, service_role;
ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT ALL ON ROUTINES TO postgres, anon, authenticated, service_role;
ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT ALL ON SEQUENCES TO postgres, anon, authenticated, service_role;
"
- name: Apply migrations
run: |
set -euo pipefail
shopt -s nullglob
files=(supabase/migrations/*.sql)
if [ ${#files[@]} -eq 0 ]; then echo "No migration files found"; exit 1; fi
for f in "${files[@]}"; do
psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -q -f "$f"
done
- name: Grant on everything the migrations created
run: |
psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -q -c "
GRANT ALL ON ALL TABLES IN SCHEMA public TO postgres, anon, authenticated, service_role;
GRANT ALL ON ALL ROUTINES IN SCHEMA public TO postgres, anon, authenticated, service_role;
GRANT ALL ON ALL SEQUENCES IN SCHEMA public TO postgres, anon, authenticated, service_role;
"
- name: Start PostgREST
# Deliberately `docker run --network host` rather than a service
# container. Service containers on a non-containerized job are reachable
# from the runner on localhost, but NOT from each other by name, and
# PostgREST has to reach Postgres. Host networking sidesteps that.
run: |
docker run -d --name postgrest --network host \
-e PGRST_DB_URI="postgres://postgres:postgres@127.0.0.1:5432/postgres" \
-e PGRST_DB_SCHEMAS=public \
-e PGRST_DB_ANON_ROLE=anon \
-e PGRST_JWT_SECRET="super-secret-jwt-token-with-at-least-32-characters-long" \
-e PGRST_DB_MAX_ROWS=100000 \
-e PGRST_SERVER_PORT=3000 \
postgrest/postgrest:v12.2.3
for _ in $(seq 1 60); do
if curl -sf -o /dev/null "http://127.0.0.1:3000/" ; then break; fi
sleep 1
done
curl -sf -o /dev/null "http://127.0.0.1:3000/" || (docker logs postgrest && exit 1)
- run: npm run test:tools