ci(mariadb): self-hosted fan-out CI (ARC runners, datadir bake, 8 shards)

This commit is contained in:
Mihir Kandoi
2026-06-24 11:19:58 +05:30
parent ead694c9cb
commit 3b8674a4a9
4 changed files with 581 additions and 135 deletions

View File

@@ -31,51 +31,49 @@ on:
permissions:
contents: read
packages: read
concurrency:
group: server-mariadb-develop-${{ github.event_name }}-${{ github.event.number || github.event_name == 'workflow_dispatch' && github.run_id || '' }}
cancel-in-progress: true
# Shared across both jobs. Both run in the SAME CI image so the bench lives at the identical
# path (/home/ci/frappe-bench) on the setup runner and the test shards — that's what makes the
# packaged Python venv portable between them.
env:
TZ: 'Asia/Kolkata'
DEBIAN_FRONTEND: noninteractive
NODE_ENV: "production"
WITH_COVERAGE: ${{ github.event_name != 'pull_request' }}
ERPNEXT_CI_USER: ci
PIP_CACHE_DIR: /home/ci/.cache/pip
npm_config_cache: /home/ci/.cache/npm
YARN_CACHE_FOLDER: /home/ci/.cache/yarn
UV_CACHE_DIR: /home/ci/.cache/uv
jobs:
test:
runs-on: ubuntu-latest
timeout-minutes: 60
env:
TZ: 'Asia/Kolkata'
NODE_ENV: "production"
WITH_COVERAGE: ${{ github.event_name != 'pull_request' }}
strategy:
fail-fast: false
matrix:
container: [1, 2, 3, 4]
name: Python Unit Tests
services:
mysql:
image: mariadb:10.6
env:
TZ: 'Asia/Kolkata'
MARIADB_ROOT_PASSWORD: 'root'
# Disable durability guarantees that are unnecessary in a throwaway CI container.
# innodb_flush_log_at_trx_commit=0 avoids an fsync on every commit (biggest win).
# sync_binlog=0 skips binary-log syncs; innodb_doublewrite=0 skips the doublewrite buffer.
MARIADB_EXTRA_FLAGS: --innodb-flush-log-at-trx-commit=0 --sync-binlog=0 --innodb-doublewrite=0
ports:
- 3306:3306
options: --health-cmd="mariadb-admin ping" --health-interval=5s --health-timeout=2s --health-retries=3
# Build the bench (clone + pip + yarn + assets) and reinstall test_site ONCE, on a free
# GitHub-hosted runner, then publish the whole bench (with a DB dump baked in) as an artifact.
# The expensive, non-parallelisable work happens here exactly once instead of on every shard.
setup:
name: Build & reinstall (setup)
# Dedicated scale set (fat cpu request) so the build+reinstall runs at full speed, uncontended
# by the many thin test shards. Same CI image + /home/ci path + 127.0.0.1 DB as the shards,
# so the packaged bench (and its venv) transplants cleanly.
runs-on: erpnext-arc-setup
timeout-minutes: 40
container:
image: ghcr.io/mihir-kandoi/erpnext-ci-mariadb:py3.14-node24-arc5
credentials:
username: ${{ secrets.GHCR_USERNAME || github.actor }}
password: ${{ secrets.GHCR_TOKEN || github.token }}
defaults:
run:
shell: bash
steps:
- name: Clone
uses: actions/checkout@v6
- name: Setup Python
uses: actions/setup-python@v6
with:
python-version: '3.14'
- name: Check for valid Python & Merge Conflicts
run: |
python -m compileall -fq "${GITHUB_WORKSPACE}"
@@ -84,53 +82,17 @@ jobs:
exit 1
fi
- name: Setup Node
uses: actions/setup-node@v6
with:
node-version: 24
check-latest: true
- name: Add to Hosts
run: echo "127.0.0.1 test_site" | sudo tee -a /etc/hosts
- name: Cache pip
uses: actions/cache@v4
with:
path: ~/.cache/pip
key: ${{ runner.os }}-pip-${{ hashFiles('**/*requirements.txt', '**/pyproject.toml') }}
restore-keys: |
${{ runner.os }}-pip-
${{ runner.os }}-
- name: Cache node modules
uses: actions/cache@v4
# MariaDB runs in-container on a datadir OUTSIDE the bench, because install.sh's next step
# does `rm -rf ~/frappe-bench`. After the reinstall, the datadir is moved into the bench so
# it ships in the artifact — test shards then start an already-loaded server (no restore).
- name: Start DB
run: bash ${GITHUB_WORKSPACE}/.github/helper/start-db.sh
env:
cache-name: cache-node-modules
with:
path: ~/.npm
key: ${{ runner.os }}-build-${{ env.cache-name }}-${{ hashFiles('**/package-lock.json') }}
restore-keys: |
${{ runner.os }}-build-${{ env.cache-name }}-
${{ runner.os }}-build-
${{ runner.os }}-
- name: Get yarn cache directory path
id: yarn-cache-dir-path
run: echo "dir=$(yarn cache dir)" >> $GITHUB_OUTPUT
- uses: actions/cache@v4
id: yarn-cache
with:
path: ${{ steps.yarn-cache-dir-path.outputs.dir }}
key: ${{ runner.os }}-yarn-${{ hashFiles('**/yarn.lock') }}
restore-keys: |
${{ runner.os }}-yarn-
- name: Cache wkhtmltopdf
uses: actions/cache@v4
with:
path: /tmp/wkhtmltox.deb
key: wkhtmltox-0.12.6.1-2-jammy-amd64
SKIP_SYSTEM_SETUP: "1"
CI_DB_DATADIR: /home/ci/db-data
- name: Install
run: bash ${GITHUB_WORKSPACE}/.github/helper/install.sh
@@ -138,21 +100,91 @@ jobs:
DB: mariadb
TYPE: server
FRAPPE_USER: ${{ github.event.inputs.user }}
FRAPPE_BRANCH: ${{ github.event.client_payload.sha || github.event.inputs.branch }}
FRAPPE_BRANCH: ${{ github.event.client_payload.sha || github.event.inputs.branch || 'develop' }}
DB_HOST: 127.0.0.1
DB_USER_HOST: '%'
WKHTMLTOX_DEB: /tmp/wkhtmltox.deb
SKIP_SYSTEM_SETUP: "1"
SKIP_WKHTMLTOX_SETUP: "1"
# Clean shutdown (consistent InnoDB datadir), then stage it inside the bench for packaging.
- name: Stop DB and stage datadir
run: |
mariadb-admin -h 127.0.0.1 -P 3306 -u root -proot shutdown || true
for _ in $(seq 1 30); do [ -f /home/ci/db-data/mysqld.pid ] || break; sleep 1; done
mv /home/ci/db-data /home/ci/frappe-bench/mariadb-data
# Package the whole bench (apps, venv, node_modules, sites, the DB dump, and hydrate.sh)
# into one artifact for the test shards to consume.
# Single-node hand-off: stage the bench on a node-local hostPath instead of round-tripping
# through GitHub artifact storage (~60s/shard). Setup and shards share the same disk, so
# the shards just untar it locally. NOTE: this assumes one node — a shard on a different
# node could not read this path (then you'd need GitHub artifacts or an NFS/RWX volume).
- name: Stage bench on node (hostPath)
run: |
cp "${GITHUB_WORKSPACE}/.github/helper/hydrate.sh" /home/ci/frappe-bench/hydrate.sh
cp "${GITHUB_WORKSPACE}/.github/helper/start-db.sh" /home/ci/frappe-bench/start-db.sh
mkdir -p /opt/ci-bench-staging
# self-clean: drop bench tars from runs older than 2h
find /opt/ci-bench-staging -maxdepth 1 -name '*.tar.gz' -mmin +120 -delete 2>/dev/null || true
# Exclude .git/node_modules; the mariadb-data datadir IS included (the pre-loaded DB).
tar czpf "/opt/ci-bench-staging/${GITHUB_RUN_ID}.tar.gz" -C /home/ci \
--exclude='.git' --exclude='node_modules' frappe-bench
ls -lh "/opt/ci-bench-staging/${GITHUB_RUN_ID}.tar.gz"
# Fan-out: each shard downloads the bench, untars it, restores the DB dump, and runs its
# slice of the suite. No clone, no build, no reinstall on the self-hosted boxes.
test:
name: Python Unit Tests (ARC)
needs: setup
runs-on: erpnext-arc
timeout-minutes: 60
container:
image: ghcr.io/mihir-kandoi/erpnext-ci-mariadb:py3.14-node24-arc5
credentials:
username: ${{ secrets.GHCR_USERNAME || github.actor }}
password: ${{ secrets.GHCR_TOKEN || github.token }}
defaults:
run:
shell: bash
strategy:
fail-fast: false
matrix:
container: [1, 2, 3, 4, 5, 6, 7, 8]
steps:
- name: Add to Hosts
run: echo "127.0.0.1 test_site" | sudo tee -a /etc/hosts
# Read the bench straight from the node-local hostPath the setup job staged it on — no
# GitHub download. -p preserves the ci (uid 1001) ownership so bench runs as ci cleanly.
- name: Untar bench from node (hostPath)
run: |
tar xzpf "/opt/ci-bench-staging/${GITHUB_RUN_ID}.tar.gz" -C /home/ci
ls -ld /home/ci/frappe-bench
- name: Hydrate (DB up + restore + bench start)
run: bash /home/ci/frappe-bench/hydrate.sh
env:
DB_HOST: 127.0.0.1
SKIP_SYSTEM_SETUP: "1"
CI_BASELINE_BACKUP: /home/ci/frappe-bench/test_site-db.sql.gz
- name: Run Tests
run: |
su -m "${ERPNEXT_CI_USER:-frappe}" -s /bin/bash <<'EOF'
cd ~/frappe-bench/
coverage_flag=""
if [ "$WITH_COVERAGE" = "true" ]; then coverage_flag="--with-coverage"; fi
bench --site test_site run-parallel-tests --lightmode --app erpnext \
--total-builds ${{ strategy.job-total }} \
--total-builds 8 \
--build-number ${{ matrix.container }} \
$coverage_flag
EOF
env:
TYPE: server
- name: Show bench output
if: ${{ always() }}
run: cat ~/frappe-bench/bench_start.log || true
@@ -162,12 +194,12 @@ jobs:
uses: actions/upload-artifact@v4
with:
name: coverage-${{ matrix.container }}
path: /home/runner/frappe-bench/sites/coverage.xml
path: /home/ci/frappe-bench/sites/coverage.xml
coverage:
name: Coverage Wrap Up
needs: test
if: ${{ github.event_name != 'pull_request' }}
needs: [test]
if: ${{ always() && github.event_name != 'pull_request' && !cancelled() && !failure() }}
runs-on: ubuntu-latest
steps:
- name: Clone