Skip to content

Run the devnet

Layout and prerequisites

Clone sibling repositories under one workspace, for example:

computechain/
  computechain/   # blockchain
  monitoring/    # Prometheus + Grafana
  docs/          # this guide

Linux, Python 3.12, pip, and a running Docker Engine with Compose are required for the complete stack. Install local chain tools once:

cd /root/computechain/computechain
python3 scripts/setup_comet.py

The setup installs into the parent workspace's .tools/, not globally. The toolchain is already installed on the current development host.

Start and stop

./start_test.sh
./start_test.sh status
./cleanup.sh

Start runs 4 equal-power validators + 1 full node and monitoring. Repeated start preserves a healthy running stand. Stop preserves keys, chain data, signing state, logs and monitoring volumes. A partial startup requires inspection, then safe stop/start; the script does not replace arbitrary processes.

Default data directory: /root/computechain/.runtime/comet-staking-devnet/. Never delete or roll back validator signing state to restart. For an independent experiment, choose a new --dir.

Without Docker: ./start_test.sh up --no-monitoring --no-docs --no-web. Add monitoring later with ./start_test.sh monitoring-up.

Open monitoring over LAN

On the current host:

No SSH tunnel is needed. The launcher detects/saves the private LAN IPv4 address; override it with --monitoring-host 192.168.0.100. Grafana login is admin; its generated password is in /root/computechain/.runtime/comet-staking-devnet/monitoring/monitoring.env (0600).

Grafana/Prometheus are available to the trusted LAN only by binding that address. Grafana requires login; Prometheus has no LAN authentication. Do not port-forward these HTTP services to the Internet. NAT does not make LAN clients trusted. ABCI, node RPC, P2P and metrics exporters remain loopback.

Generate signed transfers

./start_test.sh load --mode medium --duration 60
./start_test.sh load-stop

Modes target 3/25/100 TPS for low/medium/high. Custom: ./start_test.sh load --tps 50 --duration 600 --accounts 16 --window 8. Target is not guaranteed throughput: a short local high run observed about 36 confirmed TPS, not a capacity or 24-hour stability benchmark.

The load generator sends TRANSFER only. Separate wallets live in <devnet>/load-wallets/; the personal keystore is untouched. Funding precedes the sending duration; shutdown drains pending transactions for up to 30 seconds. One generator/faucet writer is allowed at a time. Nodes remain running after load finishes.

Status and troubleshooting

./start_test.sh status
./start_test.sh monitoring-status
tail -f ../.runtime/comet-staking-devnet/load.log

load-latest.json separates submitted, confirmed, rejected, execution failures, RPC uncertainty and unresolved transactions. Admission to mempool is not success. Logs are appN.log, engineN.log, exporter.log and load.log. Monitoring should show 7 healthy targets: 5 Comet nodes, exporter, Prometheus.

For another stand use --dir, --base-port, --grafana-port and --prometheus-port; reserve the whole blockchain range, base through base+145. Existing directories use their saved network/monitoring settings. For implementation details see the blockchain repository's COMETBFT.md.

Local interfaces

Normal up starts all three. Each has docs-up, website-up, explorer-up and matching -down, -status, -logs commands; these do not restart the chain. Run docs-up to rebuild after editing Markdown. cleanup.sh stops all interfaces and preserves data. --no-docs / --no-web skip startup; --docs-port, --website-port, --explorer-port select UI ports. Only read-only UI gateways bind LAN; explorer processes and native RPC remain loopback. Explorer is an observer, not independent proof: indexed-history totals and account state may have different heights, clearly shown in its banner.

Bootstrap a fresh state-sync follower

From the core repository, while the controlled local network is running:

./start_test.sh checkpoint --checkpoint ../.runtime/comet-staking-devnet/checkpoint-01.json --witnesses 0 1 2
./start_test.sh state-sync --node 5 --checkpoint ../.runtime/comet-staking-devnet/checkpoint-01.json --witnesses 0 1 2

Use a new filename: export never overwrites an anchor. The 30-second test trust window is intentionally short; expired anchors require a fresh trusted file, not a larger trust period. Only fresh non-genesis followers are accepted; existing history/signing state is preserved. Failed partial bootstrap requires inspection, not automatic reset. Normal offline nodes use restart/catch-up. State-sync reports are <devnet>/state-sync-nodeN.json; agreement alone is not checkpoint authenticity.

Stake and delegate (local test funds)

./start_test.sh stake --validator-node 4 --amount 6000000000000000000000
./start_test.sh delegate --validator-node 4 --amount 100000000000000000000
./start_test.sh undelegate --validator-node 4 --amount 100000000000000000000
./start_test.sh unstake --validator-node 4 --amount 6000000000000000000000

Amounts are base units: these examples use 6,000 and 100 CPC. Stake uses the node's local owner key and automatically funds it from the devnet faucet; delegation uses the faucet. Stop load before these commands. Native updates activate at H+2. Withdrawals return only after H+2+100 blocks AND 60 seconds. Old v2 data is not migrated: stop the old stand explicitly with ./cleanup.sh --dir /root/computechain/.runtime/comet-devnet before starting v3 on the same ports. Keys, data and old monitoring volumes remain intact.