deploy-prod.yml 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278
  1. name: "[Prod] Deploy"
  2. on:
  3. workflow_dispatch:
  4. inputs:
  5. tag:
  6. description: "Version tag to deploy (e.g. v1.2.39). Leave blank to deploy the newest published release. 'latest' is refused."
  7. required: false
  8. default: ""
  9. jobs:
  10. deploy:
  11. runs-on: [self-hosted, Linux, X64, hetzner]
  12. steps:
  13. - name: Resolve version tag
  14. id: tag
  15. run: |
  16. TAG="${{ inputs.tag }}"
  17. if [ -z "$TAG" ]; then
  18. # Newest published release (drafts are excluded, so publishing the
  19. # release in GitHub is the gate that makes it deployable).
  20. TAG=$(curl -fsSL -H "Authorization: Bearer ${{ secrets.GITHUB_TOKEN }}" \
  21. "https://api.github.com/repos/${{ github.repository }}/releases/latest" \
  22. | grep -m1 '"tag_name"' | cut -d'"' -f4)
  23. echo "Resolved newest published release: $TAG"
  24. fi
  25. if [ -z "$TAG" ] || [ "$TAG" = "latest" ]; then
  26. echo "::error::No deployable version tag ('latest' is refused; rollback needs a named target)."
  27. exit 1
  28. fi
  29. echo "value=$TAG" >> "$GITHUB_OUTPUT"
  30. - name: Log in to GitHub Container Registry
  31. run: echo "${{ secrets.GHCR_TOKEN }}" | docker login ghcr.io -u ${{ github.actor }} --password-stdin
  32. - name: Create deployment files
  33. env:
  34. VIRTUAL_HOST: ${{ secrets.CLOUD_VIRTUAL_HOST }}
  35. DATABASE_URL: ${{ secrets.CLOUD_DATABASE_URL }}
  36. BETTER_AUTH_SECRET: ${{ secrets.CLOUD_BETTER_AUTH_SECRET }}
  37. APP_URL: ${{ secrets.CLOUD_APP_URL }}
  38. # Plans are sold on torqvoice.com. This secret is the one torqvoice.com
  39. # holds as TORQVOICE_SERVICE_SECRET; the app signs checkout handoffs
  40. # with it and authenticates billing calls with it. The Stripe keys
  41. # themselves live in this app's database and are edited from the
  42. # torqvoice.com admin, so they are no longer passed here.
  43. TORQVOICE_SERVICE_SECRET: ${{ secrets.TORQVOICE_SERVICE_SECRET }}
  44. # Signed by torqvoice.com for APP_URL (scripts/mint-cloud-token.mjs there).
  45. # Without it the app ignores TORQVOICE_MODE=cloud and runs self-hosted.
  46. TORQVOICE_CLOUD_TOKEN: ${{ secrets.CLOUD_TORQVOICE_CLOUD_TOKEN }}
  47. DATA_PATH: ${{ secrets.DATA_PATH }}
  48. POSTHOG_HOST: ${{ secrets.NEXT_PUBLIC_POSTHOG_HOST }}
  49. POSTHOG_KEY: ${{ secrets.NEXT_PUBLIC_POSTHOG_KEY }}
  50. BACKUP_HEARTBEAT_TOKEN: ${{ secrets.BACKUP_HEARTBEAT_TOKEN }}
  51. INTEGRATIONS_ENCRYPTION_KEY: ${{ secrets.CLOUD_INTEGRATIONS_ENCRYPTION_KEY }}
  52. # Customer replies by email (Messages > email channel). The domain SES
  53. # receives mail for, the bucket its receipt rule writes into, the
  54. # region of both, and the topic whose deliveries the webhook accepts.
  55. # All four or none: with any missing the channel stays off and
  56. # customer mail goes out without a reply address, as before.
  57. INBOUND_MAIL_DOMAIN: ${{ secrets.CLOUD_INBOUND_MAIL_DOMAIN }}
  58. INBOUND_MAIL_BUCKET: ${{ secrets.CLOUD_INBOUND_MAIL_BUCKET }}
  59. INBOUND_MAIL_REGION: ${{ secrets.CLOUD_INBOUND_MAIL_REGION }}
  60. INBOUND_MAIL_TOPIC_ARN: ${{ secrets.CLOUD_INBOUND_MAIL_TOPIC_ARN }}
  61. # Backups by Torqvoice: one restic repository per Pro workshop in a bucket of ours.
  62. # Endpoint, bucket, access key, secret key and master key go together, or the schedule
  63. # stays off. The master key derives every repository password: keep it in the DR vault.
  64. # Region is optional and defaults to auto, right for Cloudflare R2; set it for a
  65. # regional bucket anywhere else. The alert email is optional.
  66. PLATFORM_BACKUP_S3_ENDPOINT: ${{ secrets.CLOUD_PLATFORM_BACKUP_S3_ENDPOINT }}
  67. PLATFORM_BACKUP_S3_BUCKET: ${{ secrets.CLOUD_PLATFORM_BACKUP_S3_BUCKET }}
  68. PLATFORM_BACKUP_S3_REGION: ${{ secrets.CLOUD_PLATFORM_BACKUP_S3_REGION }}
  69. PLATFORM_BACKUP_S3_ACCESS_KEY: ${{ secrets.CLOUD_PLATFORM_BACKUP_S3_ACCESS_KEY }}
  70. PLATFORM_BACKUP_S3_SECRET_KEY: ${{ secrets.CLOUD_PLATFORM_BACKUP_S3_SECRET_KEY }}
  71. PLATFORM_BACKUP_MASTER_KEY: ${{ secrets.CLOUD_PLATFORM_BACKUP_MASTER_KEY }}
  72. PLATFORM_BACKUP_ALERT_EMAIL: ${{ secrets.CLOUD_PLATFORM_BACKUP_ALERT_EMAIL }}
  73. # Optional, a repository variable rather than a secret: the largest backup data
  74. # file a restore reads into memory, in MiB (128 unless set, see .env.example).
  75. BACKUP_MAX_DATA_FILE_MB: ${{ vars.CLOUD_BACKUP_MAX_DATA_FILE_MB }}
  76. # Optional. Next encrypts what a server action closes over with this
  77. # key; left unset, the key the image was built with is used. It does
  78. # not keep open tabs working across a deploy: action IDs are fixed
  79. # when the image is built, from the build's key (see the Dockerfile).
  80. # 32 random bytes, base64: `openssl rand -base64 32`.
  81. NEXT_SERVER_ACTIONS_ENCRYPTION_KEY: ${{ secrets.CLOUD_NEXT_SERVER_ACTIONS_ENCRYPTION_KEY }}
  82. GOOGLE_INTEGRATION_CLIENT_ID: ${{ secrets.CLOUD_GOOGLE_INTEGRATION_CLIENT_ID }}
  83. GOOGLE_INTEGRATION_CLIENT_SECRET: ${{ secrets.CLOUD_GOOGLE_INTEGRATION_CLIENT_SECRET }}
  84. GOOGLE_AUTH_CLIENT_ID: ${{ secrets.CLOUD_GOOGLE_AUTH_CLIENT_ID }}
  85. GOOGLE_AUTH_CLIENT_SECRET: ${{ secrets.CLOUD_GOOGLE_AUTH_CLIENT_SECRET }}
  86. MICROSOFT_INTEGRATION_CLIENT_ID: ${{ secrets.CLOUD_MICROSOFT_INTEGRATION_CLIENT_ID }}
  87. MICROSOFT_INTEGRATION_CLIENT_SECRET: ${{ secrets.CLOUD_MICROSOFT_INTEGRATION_CLIENT_SECRET }}
  88. ZOOM_INTEGRATION_CLIENT_ID: ${{ secrets.CLOUD_ZOOM_INTEGRATION_CLIENT_ID }}
  89. ZOOM_INTEGRATION_CLIENT_SECRET: ${{ secrets.CLOUD_ZOOM_INTEGRATION_CLIENT_SECRET }}
  90. QUICKBOOKS_INTEGRATION_CLIENT_ID: ${{ secrets.CLOUD_QUICKBOOKS_INTEGRATION_CLIENT_ID }}
  91. QUICKBOOKS_INTEGRATION_CLIENT_SECRET: ${{ secrets.CLOUD_QUICKBOOKS_INTEGRATION_CLIENT_SECRET }}
  92. FIKEN_INTEGRATION_CLIENT_ID: ${{ secrets.CLOUD_FIKEN_INTEGRATION_CLIENT_ID }}
  93. FIKEN_INTEGRATION_CLIENT_SECRET: ${{ secrets.CLOUD_FIKEN_INTEGRATION_CLIENT_SECRET }}
  94. # Torqvoice Codes, the fault code service, run beside the app. The address
  95. # is the service's on the server's Docker network (http://torqvoice-codes:8090
  96. # when it joins the proxy network), and the key is one key with read:codes,
  97. # read:fixes and write:fixes (sharing fixes needs the last two), minted
  98. # in the torqvoice.com admin under Code service. With both set every workshop
  99. # is connected with our key and never sees a key field; with either empty
  100. # the fault code lookups stay off.
  101. TORQVOICE_CODES_URL: ${{ secrets.CLOUD_TORQVOICE_CODES_URL }}
  102. TORQVOICE_CODES_API_TOKEN: ${{ secrets.CLOUD_TORQVOICE_CODES_API_TOKEN }}
  103. run: |
  104. mkdir -p $HOME/torqvoice-deploy/prod
  105. cd $HOME/torqvoice-deploy/prod
  106. env | grep -E '^(VIRTUAL_HOST|DATA_PATH|DATABASE_URL|BETTER_AUTH_SECRET|APP_URL|TORQVOICE_|POSTHOG_|BACKUP_|INTEGRATIONS_|GOOGLE_INTEGRATION_|GOOGLE_AUTH_|MICROSOFT_INTEGRATION_|ZOOM_INTEGRATION_|QUICKBOOKS_INTEGRATION_|FIKEN_INTEGRATION_|INBOUND_MAIL_|PLATFORM_BACKUP_|NEXT_SERVER_ACTIONS_)' > .env
  107. cat > docker-compose.yml << 'COMPOSE'
  108. networks:
  109. proxy:
  110. external: true
  111. name: proxy
  112. services:
  113. torqvoice-app:
  114. container_name: torqvoice-app
  115. image: ghcr.io/torqvoice/torqvoice:${APP_TAG:-latest}
  116. restart: unless-stopped
  117. volumes:
  118. - ${DATA_PATH}/torqvoice_data:/app/data
  119. expose:
  120. - "3000"
  121. environment:
  122. VIRTUAL_HOST: ${VIRTUAL_HOST}
  123. VIRTUAL_PORT: "3000"
  124. LETSENCRYPT_HOST: ${VIRTUAL_HOST}
  125. DATABASE_URL: ${DATABASE_URL}
  126. BETTER_AUTH_SECRET: ${BETTER_AUTH_SECRET}
  127. NEXT_PUBLIC_APP_URL: ${APP_URL}
  128. TORQVOICE_MODE: cloud
  129. # Production sits behind Cloudflare and the origin only accepts its
  130. # edges, so the visitor address is the one Cloudflare says it is.
  131. # Without this every visitor shares the edge's rate-limit bucket.
  132. TRUST_CF_CONNECTING_IP: "true"
  133. TORQVOICE_SERVICE_SECRET: ${TORQVOICE_SERVICE_SECRET}
  134. TORQVOICE_CLOUD_TOKEN: ${TORQVOICE_CLOUD_TOKEN}
  135. TORQVOICE_COM_URL: https://torqvoice.com
  136. POSTHOG_HOST: ${POSTHOG_HOST}
  137. POSTHOG_KEY: ${POSTHOG_KEY}
  138. BACKUP_HEARTBEAT_TOKEN: ${BACKUP_HEARTBEAT_TOKEN}
  139. INTEGRATIONS_ENCRYPTION_KEY: ${INTEGRATIONS_ENCRYPTION_KEY}
  140. INBOUND_MAIL_DOMAIN: ${INBOUND_MAIL_DOMAIN}
  141. INBOUND_MAIL_BUCKET: ${INBOUND_MAIL_BUCKET}
  142. INBOUND_MAIL_REGION: ${INBOUND_MAIL_REGION}
  143. INBOUND_MAIL_TOPIC_ARN: ${INBOUND_MAIL_TOPIC_ARN}
  144. PLATFORM_BACKUP_S3_ENDPOINT: ${PLATFORM_BACKUP_S3_ENDPOINT}
  145. PLATFORM_BACKUP_S3_BUCKET: ${PLATFORM_BACKUP_S3_BUCKET}
  146. PLATFORM_BACKUP_S3_REGION: ${PLATFORM_BACKUP_S3_REGION}
  147. PLATFORM_BACKUP_S3_ACCESS_KEY: ${PLATFORM_BACKUP_S3_ACCESS_KEY}
  148. PLATFORM_BACKUP_S3_SECRET_KEY: ${PLATFORM_BACKUP_S3_SECRET_KEY}
  149. PLATFORM_BACKUP_MASTER_KEY: ${PLATFORM_BACKUP_MASTER_KEY}
  150. PLATFORM_BACKUP_ALERT_EMAIL: ${PLATFORM_BACKUP_ALERT_EMAIL}
  151. BACKUP_MAX_DATA_FILE_MB: ${BACKUP_MAX_DATA_FILE_MB}
  152. NEXT_SERVER_ACTIONS_ENCRYPTION_KEY: ${NEXT_SERVER_ACTIONS_ENCRYPTION_KEY}
  153. GOOGLE_INTEGRATION_CLIENT_ID: ${GOOGLE_INTEGRATION_CLIENT_ID}
  154. GOOGLE_INTEGRATION_CLIENT_SECRET: ${GOOGLE_INTEGRATION_CLIENT_SECRET}
  155. GOOGLE_AUTH_CLIENT_ID: ${GOOGLE_AUTH_CLIENT_ID}
  156. GOOGLE_AUTH_CLIENT_SECRET: ${GOOGLE_AUTH_CLIENT_SECRET}
  157. MICROSOFT_INTEGRATION_CLIENT_ID: ${MICROSOFT_INTEGRATION_CLIENT_ID}
  158. MICROSOFT_INTEGRATION_CLIENT_SECRET: ${MICROSOFT_INTEGRATION_CLIENT_SECRET}
  159. ZOOM_INTEGRATION_CLIENT_ID: ${ZOOM_INTEGRATION_CLIENT_ID}
  160. ZOOM_INTEGRATION_CLIENT_SECRET: ${ZOOM_INTEGRATION_CLIENT_SECRET}
  161. QUICKBOOKS_INTEGRATION_CLIENT_ID: ${QUICKBOOKS_INTEGRATION_CLIENT_ID}
  162. QUICKBOOKS_INTEGRATION_CLIENT_SECRET: ${QUICKBOOKS_INTEGRATION_CLIENT_SECRET}
  163. FIKEN_INTEGRATION_CLIENT_ID: ${FIKEN_INTEGRATION_CLIENT_ID}
  164. FIKEN_INTEGRATION_CLIENT_SECRET: ${FIKEN_INTEGRATION_CLIENT_SECRET}
  165. TORQVOICE_CODES_URL: ${TORQVOICE_CODES_URL}
  166. TORQVOICE_CODES_API_TOKEN: ${TORQVOICE_CODES_API_TOKEN}
  167. networks:
  168. - proxy
  169. COMPOSE
  170. # Without the cloud token the new container would come up as a
  171. # self-hosted install: one workshop for everybody, branding back on.
  172. # Stop here, before anything is replaced.
  173. if [ -z "$TORQVOICE_CLOUD_TOKEN" ]; then
  174. echo "::error title=Cloud token missing::CLOUD_TORQVOICE_CLOUD_TOKEN is empty. Mint one for $APP_URL with scripts/mint-cloud-token.mjs in the torqvoice.com repo."
  175. exit 1
  176. fi
  177. # The link to torqvoice.com is what the subscription pages hang on. A
  178. # wrong secret or an origin the site does not list only shows up as a
  179. # missing page, so the deploy asks the site itself and says so.
  180. if [ -n "$TORQVOICE_SERVICE_SECRET" ]; then
  181. status=$(curl -sS -o /dev/null -w '%{http_code}' -X POST https://torqvoice.com/api/app/subscription/ping \
  182. -H "Authorization: Bearer $TORQVOICE_SERVICE_SECRET" -H 'Content-Type: application/json' \
  183. -d "{\"appUrl\":\"$APP_URL\"}" || echo 000)
  184. if [ "$status" != "200" ]; then
  185. echo "::warning title=torqvoice.com link not confirmed::ping answered $status. Check TORQVOICE_SERVICE_SECRET on both sides and that $APP_URL is TORQVOICE_APP_URL or TORQVOICE_STAGING_APP_URL on torqvoice.com. The subscription page stays hidden until it is."
  186. else
  187. echo "torqvoice.com link confirmed for $APP_URL"
  188. fi
  189. else
  190. echo "::warning title=torqvoice.com link not configured::TORQVOICE_SERVICE_SECRET is empty, so plans cannot be bought from this deployment."
  191. fi
  192. - name: Ensure data directory permissions
  193. run: |
  194. DATA_DIR="${{ secrets.DATA_PATH }}/torqvoice_data"
  195. mkdir -p "$DATA_DIR"
  196. docker run --rm -v "$DATA_DIR":/data alpine chown -R 1001:1001 /data
  197. - name: Check free disk space
  198. run: |
  199. MIN_GB=5
  200. DOCKER_ROOT=$(docker info --format '{{.DockerRootDir}}' 2>/dev/null || echo /var/lib/docker)
  201. AVAIL_KB=$(df -Pk "$DOCKER_ROOT" | awk 'NR==2 {print $4}')
  202. AVAIL_GB=$(awk "BEGIN {printf \"%.1f\", $AVAIL_KB/1024/1024}")
  203. echo "Free space on $DOCKER_ROOT: ${AVAIL_GB} GB (minimum ${MIN_GB} GB)"
  204. if [ "$AVAIL_KB" -lt $((MIN_GB * 1024 * 1024)) ]; then
  205. echo "::error::Not enough disk space to pull image: ${AVAIL_GB} GB free, need ${MIN_GB} GB. Aborting deploy."
  206. exit 1
  207. fi
  208. - name: Back up production database
  209. env:
  210. DATABASE_URL: ${{ secrets.CLOUD_DATABASE_URL }}
  211. run: |
  212. set -euo pipefail
  213. DB_USER=$(echo "$DATABASE_URL" | sed -E 's|^[^:]+://([^:]+):.*|\1|')
  214. DB_NAME=$(echo "$DATABASE_URL" | sed -E 's|.*/([^/?]+)(\?.*)?$|\1|')
  215. BACKUP_DIR="${{ secrets.DATA_PATH }}/db-backups/$DB_NAME"
  216. # The volume is root-owned and the runner is not root. Binding the
  217. # path into a container makes Docker create it, and the chown hands
  218. # it to the runner so the dump redirect below can write there.
  219. docker run --rm -v "$BACKUP_DIR":/b alpine chown "$(id -u):$(id -g)" /b
  220. # The Postgres data directory lives on this filesystem: filling it
  221. # takes the database down, not just the backup.
  222. MIN_GB=2
  223. AVAIL_KB=$(df -Pk "$BACKUP_DIR" | awk 'NR==2 {print $4}')
  224. if [ "$AVAIL_KB" -lt $((MIN_GB * 1024 * 1024)) ]; then
  225. echo "::error::Only $(awk "BEGIN {printf \"%.1f\", $AVAIL_KB/1024/1024}") GB free where the database lives. Refusing to dump."
  226. exit 1
  227. fi
  228. DUMP="$BACKUP_DIR/${DB_NAME}-$(date +%Y%m%d-%H%M%S)-pre-${{ steps.tag.outputs.value }}.dump"
  229. echo "Dumping $DB_NAME to $DUMP"
  230. docker exec torqvoice-db pg_dump -U "$DB_USER" -d "$DB_NAME" -Fc > "$DUMP"
  231. # A dump nobody can read is worse than no dump. No backup, no deploy.
  232. docker exec -i torqvoice-db pg_restore --list < "$DUMP" > /dev/null
  233. echo "Verified $(du -h "$DUMP" | cut -f1) backup"
  234. # Keep the 10 most recent pre-deploy dumps; hourly dumps in the same
  235. # folder belong to the backup cron and age out there.
  236. ls -t "$BACKUP_DIR"/*-pre-*.dump 2>/dev/null | tail -n +11 | xargs -r rm --
  237. - name: Deploy
  238. run: |
  239. cd $HOME/torqvoice-deploy/prod
  240. APP_TAG=${{ steps.tag.outputs.value }} docker compose up -d --pull always
  241. - name: Health check
  242. run: |
  243. echo "Waiting for prod container to be ready..."
  244. for i in $(seq 1 12); do
  245. STATUS=$(docker inspect --format='{{.State.Status}}' torqvoice-app 2>/dev/null || echo "not found")
  246. if [ "$STATUS" = "running" ]; then
  247. echo "Deployed ${{ steps.tag.outputs.value }}."
  248. exit 0
  249. fi
  250. echo "Status: $STATUS, retrying in 5s..."
  251. sleep 5
  252. done
  253. echo "::error::Container failed to start"
  254. docker logs torqvoice-app --tail 50
  255. exit 1
  256. - name: Prune old images and build cache
  257. run: docker image prune -f && docker builder prune -f