ci: gate build on tests, deploy waits for container health
Some checks failed
Build and Deploy to Production / build (push) Blocked by required conditions
Build and Deploy to Production / deploy (push) Blocked by required conditions
Build and Deploy to Production / test (push) Failing after 11m37s

This commit is contained in:
Sergei Poljanski 2026-07-02 18:32:53 +04:00
commit 67263cb75c
Signed by: asxpi
GPG key ID: 4F8851660FA4121B
3 changed files with 56 additions and 7 deletions

View file

@ -6,8 +6,44 @@ on:
- main
jobs:
# Same job as in test.yaml (which covers non-main branches); duplicated here
# so a red suite blocks the build+deploy — Forgejo has no cross-workflow needs.
test:
runs-on: nix-latest
services:
postgres:
image: postgres:17-alpine
env:
POSTGRES_USER: postgres
POSTGRES_HOST_AUTH_METHOD: trust
POSTGRES_DB: asxpio_test
steps:
- name: Prepare container for actions
run: |
echo "experimental-features = nix-command flakes" >> /etc/nix/nix.conf
# node for JS actions (checkout); git-minimal is already in the image
nix-env -iA nixpkgs.nodejs_22
- name: Checkout
uses: actions/checkout@v4
- name: Wait for Postgres
run: |
for i in $(seq 1 30); do
nix develop -c pg_isready -h postgres -U postgres && exit 0
sleep 2
done
echo "Postgres service never became ready" >&2
exit 1
- name: Run test suite
env:
TEST_DATABASE_URL: postgres://postgres@postgres:5432/asxpio_test
run: nix develop -c bundle exec rake test
build:
runs-on: arch-latest
needs: test
container:
image: gcr.io/kaniko-project/executor:debug
outputs:
@ -115,7 +151,16 @@ jobs:
docker pull $NEW_IMAGE
docker compose up -d
sleep 5
docker ps | grep asxpio || exit 1
echo "Deployment successful!"
# Wait for the image HEALTHCHECK (GET /healthz) to report healthy;
# `docker ps | grep` passed even while the app crash-looped.
for i in $(seq 1 18); do
status=$(docker inspect --format '{{.State.Health.Status}}' asxpio 2>/dev/null || echo missing)
if [ "$status" = "healthy" ]; then
echo "Deployment successful!"
exit 0
fi
sleep 5
done
echo "asxpio never became healthy (last status: $status)" >&2
docker logs --tail 50 asxpio
exit 1

View file

@ -1,9 +1,11 @@
name: Tests
# main is covered by the test job inside deploy.yaml (which gates the build);
# this workflow covers every other branch.
on:
push:
branches:
- '**'
branches-ignore:
- main
jobs:
test:

View file

@ -84,6 +84,7 @@ Public:
- `GET /i/:uuid` — landing page: client name, number, total, status badge, download button.
- `GET /i/:uuid/pdf` — 302 to a 5-minute MinIO presigned URL (Content-Disposition: attachment).
- `GET /healthz` — liveness (plus `SELECT 1` when invoicing is configured); used by the Docker HEALTHCHECK and the deploy pipeline's post-deploy wait.
### Storage model
@ -138,10 +139,11 @@ The hedgehog is the user's pre-IE personal mark; once the IE has its own logo, t
- **Run tests:** `nix develop -c bin/test` — full suite (minitest + rack-test) against an ephemeral Postgres it provisions and tears down itself. `bundle exec rake test` runs without a DB, skipping DB-backed tests. Mail uses `Mail::TestMailer`; S3 uses aws-sdk stubbed responses; tests never touch real services, and `.env` is deliberately not loaded when `RACK_ENV=test`.
- **Local image build (sanity check):** `docker compose build`
- **Production deploy:** automatic via `.forgejo/workflows/deploy.yaml` on push to `main`. Pipeline:
0. Test suite runs on the `nix-latest` runner label (postgres service container); a red suite blocks the build. Non-main branches get the same job from `test.yaml`.
1. Kaniko builds the image and pushes it to the Forgejo registry.
2. SSH copies `docker-compose.yml` to the deploy directory on the prod host.
3. SSH writes `.env` (mode 600) from Forgejo secrets. The full set of invoicing envs (`DATABASE_URL`, `S3_*`, `ADMIN_*`) is built here; the per-app `ASXPIO_DB_PASSWORD` / `ASXPIO_S3_*` values must match the storage repo's Forgejo secrets, since storage's init containers provision the corresponding role + MinIO user.
4. SSH `sed`-substitutes the image tag in `docker-compose.yml` and runs `docker compose up -d`.
4. SSH `sed`-substitutes the image tag in `docker-compose.yml`, runs `docker compose up -d`, and waits (up to 90s) for the image `HEALTHCHECK` (`GET /healthz`) to report healthy — a crash-looping container fails the deploy.
## Where the secrets live