Skip to content

REDEPLOY.md

Latest commit

 

History

History
131 lines (93 loc) · 5.23 KB

REDEPLOY.md

File metadata and controls

131 lines (93 loc) · 5.23 KB
Raw

Redeploy x402.sentinel.co

The latest image is built and pushed. The host needs to pull and restart.

What changed (commits since the running build)

SHA Title
3175833 server: structured JSON errors + pre-payment validation
7099f15 server: fix manifest URLs (x402.blue → x402.sentinel.co, repo links)
3d14045 docs: fix dead x402.blue URL, add USDC bootstrap section (docs-only, no code)
1b1fdb9 treewide: rename blue-agent-connect → blue-js-sdk/ai-path
4f04835 server: add GET /manifest + /nodes for AI-agent discovery
a10db35 dashboard: restructure on EventEmitter-based SSE pattern (dashboard-only)

The running build is 47+ hours old and predates all of these.

Why this matters

Currently the server's 200 response from POST /vpn/connect/<tier> ships:

"instructions": "import { connect } from 'blue-agent-connect'; ..."

That package name is dead — it was renamed to blue-js-sdk/ai-path. The corrected source is in 7099f15. The full flow works because the docs (HTML + llms.txt) teach the correct import, but agents that read the response's instructions field get bad copy-paste.

Also the manifest at GET /manifest currently 404s (route added in 4f04835) — the discovery surface AI agents are supposed to hit first.

Image

GHCR built and pushed by .github/workflows/build.yml on push to master.

If ghcr.io/sentinel-bluebuilder/x402 is private, the host may need a docker login ghcr.io with a token that has read:packages.

Redeploy

Whatever orchestrator is running it (compose / systemd / k8s / bare docker), the recipe is:

docker pull ghcr.io/sentinel-bluebuilder/x402:latest
docker stop x402 && docker rm x402         # adjust container name
docker run -d \
  --name x402 \
  --restart unless-stopped \
  -p 4020:4020 \
  -p 4021:4021 \
  --env-file /path/to/server/.env \
  ghcr.io/sentinel-bluebuilder/x402:latest

If they use docker compose:

docker compose pull
docker compose up -d

Verify after restart

# Uptime should reset to seconds, not hours
curl -s https://x402.sentinel.co/health
# {"status":"ok","uptime":12.345}

# Manifest must respond (currently 404 on the old build)
curl -s https://x402.sentinel.co/manifest | head -c 200

# Nodes route must respond (currently 404 on the old build)
curl -s https://x402.sentinel.co/nodes

# Error surface must be structured JSON (currently empty/HTML on the old build)
curl -s -X POST https://x402.sentinel.co/vpn/connect/forever -H 'Content-Type: application/json' -d '{}'
# Expect: {"code":"UNKNOWN_TIER","message":"Tier 'forever' does not exist...",...}

curl -s -X POST https://x402.sentinel.co/vpn/connect/1day -H 'Content-Type: application/json' -d '{"sentinelAddr":"hello"}'
# Expect: {"code":"INVALID_SENTINEL_ADDR","message":"...","expectedPattern":"^sent1[0-9a-z]{38}$",...}

# A real provisioning call's instructions string must reference blue-js-sdk/ai-path
# (no easy curl — running the end-to-end test from fresh-test/connect-docs-driven.mjs
#  is the cleanest check; verifies instructions field has the correct package name)

Env vars (no changes from current .env)

Nothing new is required. The image reads the same variables it always has:

Required
OPERATOR_ADDRESS EVM wallet receiving USDC on Base
FACILITATOR_PRIVATE_KEY EVM key with ETH on Base for facilitator gas
SENTINEL_OPERATOR_MNEMONIC 12-word Sentinel mnemonic with P2P
SENTINEL_PLAN_ID Sentinel plan ID with leased nodes (currently 243)
Optional (defaults shown)
PORT=4020 Server port
FACILITATOR_PORT=4021 Self-hosted facilitator port
BASE_NETWORK=eip155:8453 Base mainnet
SENTINEL_RPC_URL=https://rpc.sentinel.co:443
SENTINEL_LCD_URL=https://lcd.sentinel.co LCD fallback

Rollback

If anything misbehaves, pin to the previous good SHA:

docker pull ghcr.io/sentinel-bluebuilder/x402:1b1fdb9...   # last known-good SHA
# or any earlier tag from https://github.com/Sentinel-Bluebuilder/x402/pkgs/container/x402

The currently running build corresponds to a SHA from ~2026-05-19 — we don't know exactly which, but :1b1fdb9... is a safe rollback (it's behind by only 2 commits, both narrow scope).

Risk

  • Low. The server's TS build is clean. Routes added (/manifest, /nodes) are additive. The instructions string change is a pure string. No schema changes, no DB migrations, no env-var changes.
  • Tested: end-to-end test against the current server passes today (Romania node, 24.9s, zero friction). The new image only differs in: (1) instructions field content, (2) two new GET endpoints, (3) manifest URLs in the manifest body.

Contact for questions

The end-to-end test that validates everything is:

cd fresh-test
node connect-docs-driven.mjs

It costs ~$0.033 USDC (1-day tier) on Base, connects through a real Sentinel node, and disconnects cleanly. Run it before and after the redeploy to compare.