glamperd / indrav2

Indra 2.0

Connext's new & improved state channel network!

TL;DR

Make sure the following tools are installed: make, jq, and docker and then run make start. This will build everything & launch a pre-configured payment hub + a simple UI, play with it from your browser at 127.0.0.1:3000.

Run bash ops/logs.sh node to see the node server's logs or replace "node" with one of the other service names, list services with: make dls (short for Docker List).

To run e2e UI tests use: make watch-ui, to run unit tests for the hub use: make test-hub. Browse the Makefile for all available shortcut commands and the ops/ folder for all available helper scripts.

When you're done testing it out, shut the whole thing down with make stop.

Contents

• Deploy for local development
• Deploy to production
• How to interact with the Hub
• Debugging & Troubleshooting

If you encounter any problems, check out the debugging guide at the bottom of this doc. For any unanswered questions, open an issue or reach out on our Discord channel & we'll be happy to help.

Discord Invitation: https://discord.gg/SmMSFf

Deploy for local development

Prerequisites

• make: Probably already installed, otherwise install w brew install make or apt install make or similar.
• jq: Probably not installed yet, install w brew install jq or apt install jq or similar.
• docker: sadly, Docker is kinda annoying to install. See website for instructions.

To download this repo, build, and deploy in dev-mode, run the following:

git clone https://github.com/ConnextProject/indra.git
cd indra
make start

Beware! The first time make start is run, it will take a very long time (maybe as long as 5 minutes depending on your internet speed) but have no fear: downloads will be cached & most build steps won't ever need to be repeated again so subsequent make start runs will go much more quickly. Get this started asap & browse the rest of the README while the first build/deploy completes.

Useful Commands

• make start: Builds everything & then starts the app
• make stop: Stop the app once it's been started
• make restart: Stop the app & start it again, rebuilding anything that's out of date
• make clean: Stops the app & deletes all build artifacts eg transpiled typescript.
• make reset: Stops the app & removes all persistent data (eg database & chaindata)
• make restart-prod: Restarts the app in production-mode
• make dls: Show all running services (groups of containers) plus list all running containers.
• bash ops/db.sh: Opens a console attached to the running app's database. You can also run npm run db '\d+' to run a single PostgreSQL query (eg \d+ to list table details) and then exit.
• bash ops/logs.sh hub: Monitor the hub's logs. Similar commands can be run to monitor logs for the proxy, chainsaw, ethprovider (for migrations output), ganache (for log of rpc calls to ganache), database, redis, etc.

Running Unit Tests

• make watch-ui: Start a test-optimized browser that will use cypress to run automated e2e tests.
• make watch-node: start a test watcher that will re-run node server's unit tests when source code changes
• make test-ui or make test-node: run either one-off ui-based e2e tests or node unit tests that will exit once tests are finished.

Deploying to Production

TL;DR

1. Push to staging, make sure CI passes & the staging deployment looks healthy
2. If there are contracts that have changed, delete their addresses from address-book.json and redeploy them with bash ops/deploy-contracts.sh <network>
3. If any npm packages have changed, run bash ops/npm-publish.sh from staging
4. Run bash ops/deploy-indra.sh from staging to merge to master & trigger the deployment to prod

First, setup CircleCI Environment Variables

Once per CircleCI account or organization

Run ssh-keygen -t rsa -b 4096 -C "circleci" -m pem -f .ssh/circleci to generate a new ssh key pair. Load the private key (.ssh/circleci) into CircleCI -> Settings -> Permissions -> SSH Permissions.

Go to CircleCI -> Settings -> Build Settings -> Environment Variables

• DOCKER_USER & DOCKER_PASSWORD: Login credentials for someone with push access to the docker repository specified by the registry vars at the top of the Makefile & ops/start-prod.sh.
• STAGING_DOMAINNAME & RINKEBY_DOMAINNAME & MAINNET_DOMAINNAME: The URLs from which the Indra application will be served.
• RINKEBY_ETH_PROVIDER & MAINNET_ETH_PROVIDER: Alchemy or Infura URLs that let us read/write to the blockchain

Note: If STAGING_URL=staging.example.com then

• DNS needs to be properly configured so that staging.example.com will resolve to the IP address of your staging server
• The admin should have ssh access via ssh root@$STAGING_URL or ssh ubuntu@$STAGING_URL after completing the next step.
• The application will be accessible from https://staging.example.com after deploying.

Second, setup the production server

Once per server

1. Create a new server via AWS or DigitalOcean or whichever is your favorite cloud provider. Note it's IP address.
2. Set up DNS so that eg the $RINKEBY_DOMAINNAME you specified in CI env vars points to this server.
3. Copy your indra hub's mnemonic to your clipboard. You can generate a new random mnemonic from a node console with ethers by doing something like this: require('ethers').Wallet.createRandom()
4. Run the following script (for best results, run it with a $SERVER_IP that points to a fresh Ubuntu VM)

bash ops/setup-ubuntu.sh $SERVER_IP rinkeby

To run the setup script, we need to be able to ssh into this server via either root@$SERVER_IP or ubuntu@$SERVER_IP. If root, this script will setup the ubuntu user and disable root login for security.

This setup script expects to find the private key for ssh access to the server at ~/.ssh/connext-aws & CircleCI's public key at ~/.ssh/circleci.pub.

By default, this script will load your hub's rinkeby private key into a docker secret stored on the server. To setup a server for another network (eg mainnet) add a network arg to the end, eg: bash ops/setup-ubuntu.sh $SERVER_IP mainnet

You can remove the server's private key like this:

(Make sure that this server doesn't have a hub running on it before removing it's key)

ssh -t -i ~/.ssh/connext-aws ubuntu@$SERVER_IP docker secret rm hub_key_mainnet

And add a new private key by re-running the setup-ubuntu script. Note: this script is idempotent ie you can run it over and over again w/out causing any problems.

Third, deploy the contracts

Once per update to smart contracts

To deploy the ChannelManager contract & dependencies to Rinkeby:

export ETH_PROVIDER="https://rinkeby.infura.io/abc123"
bash ops/deploy.contracts.sh rinkeby

This script will prompt you to paste in the deployment address's private key if one called hub_key_rinkeby hasn't already been saved to the secret store, this address will be used to deploy contracts. See saved secrets with docker secret ls.

The contract deployment script will save the addresses of your deployed contracts in modules/contracts/ops/address-book.json. This file is automatically generated and you probably won't need to mess with it. One exception: if you want to redeploy some contract(s), then delete their addresses from the address book & re-run the above deployment script.

You can upload a custom address book to your prod server's project root like this:

scp -i ~/.ssh/connext-aws address-book.json ubuntu@$SERVER_IP:~/indra/

Fourth, deploy a new version of the connext npm packages

Once per update to npm packages eg messaging, types, or client

To publish a new version of the client:

bash ops/npm-publish.sh

This script will prompt you for a new version number. Heuristics:

• Is this minor bug fix? Then increment the minor version eg 1.0.0 -> 1.0.1
• Did you add a new, backwards-compatible feature? Then increment the middle version eg 1.0.0 -> 1.1.0
• Did you add a new, backwards-incompatible feature? Then increment the major version eg 1.0.0 -> 2.0.0

Once you specify the version, it will automatically:

• update modules/{packages-to-publish}/package.json with the new version
• run npm publish
• update modules/{packages-that-depend-on-newly-published-packages}/package.json to import the new version of the package
• create & push a new git commit
• create & push a new git tag

Lastly, deploy a new Indra hub

There is a long-lived staging branch that is the intermediate step between Indra development and production. All merges and PRs should be made from a feature branch onto staging. The master branch is dealt with automatically so you shouldn't need to manually commit or merge to master unless you're updating the readme or something that doesn't affect any of the actual source code.

Updating origin/staging will kick off the first round of CI and, if all tests pass, it will deploy the changes to the $STAGING_URL configured by your circle ci env.

git checkout staging && git mege feature-branch # alternatively, do a code review & merge via a GitHub PR

Once the staging branch's CI runs and deploys, check out your staging env. Does everything look good? Seem ready to deploy to production?

To deploy a new Indra hub to production, run:

bash ops/deploy-indra.sh

This script will prompt you for a new version number. See the previous step for versioning heuristics. Once you specify the version, it will automatically:

• merge staging into master
• update the project root's package.json with the version you provided & amend this change to the merge commit
• push this commit to origin/master
• create & push a new git tag

Pushing to origin/master will trigger another CI run that will deploy a new Indra hub to production if no tests fail.

Monitor the prod hub's logs with

ssh -i ~/.ssh/connext-aws ubuntu@SERVER_IP bash indra/ops/logs.sh hub

The ChannelManager contract needs collateral to keep doing its thing. Make sure the hub's wallet has enough funds before deploying. Funds can be moved from the hub's wallet to the contract manually via:

ssh -i ~/.ssh/connext-aws ubuntu@SERVER_IP bash indra/ops/collateralize.sh 3.14 eth
# To move Eth, or to move tokens:
ssh -i ~/.ssh/connext-aws ubuntu@SERVER_IP bash indra/ops/collateralize.sh 1000 token

How to interact with an Indra hub

A prod-mode indra hub exposes the following API (source):

• /api/hub is the prefix for the hub's api
• /api/hub/config returns the hub's config for example
• /api/hub/subscribe connects to the hub's websocket server for real-time exchange rate & gas price updates
• /api/eth connects to the hub's eth provider
• anything else, redirects the user to a daicard client

..from a dai card

Dai card in production runs a proxy with endpoints:

• /api/rinkeby/hub -> https://rinkeby.hub.connext.network/api/hub
• /api/rinkeby/eth -> https://rinkeby.hub.connext.network/api/eth
• /api/mainnet/hub -> https://hub.connext.network/api/hub
• /api/mainnet/eth -> https://hub.connext.network/api/eth
• anything else: serves the daicard html/css/js files

Hub API

1. AuthApiService

• GET /auth/status: returns success and address if a valid auth token is provided
• POST /auth/challenge: returns a challenge nonce
• POST /auth/response:
  • nonce: returned by /auth/challenge
  • address
  • origin
  • signature

2. ChannelsApiService

• POST /channel/:user/request-deposit:
  • depositWei
  • depositToken
  • lastChanTx
  • lastThreadUpdateId
• GET /channel/:user/sync
  • params
    • lastChanTx
    • lastThreadUpdateId

TODO: Complete this section

Debugging

If you encounter problems while the app is running, the first thing to do is check the logs of each component:

• bash ops/logs.sh node: Core hub logic logs
• bash ops/logs.sh database
• bash ops/logs.sh proxy

The container name "/indra_buidler" is already in use

Full error message:

docker: Error response from daemon: Conflict. The container name "/indra_buidler" is already in use by container "6d37b932d8047e16f4a8fdf58780fe6974e6beef58bf4cc5e48d00d3e94a67c3". You have to remove (or rename) that container to be able to reuse that name.

You probably started to build something and then stopped it with ctrl-c. It only looks like the build stopped: the builder process is still hanging out in the background wrapping up what it was working on. If you wait for a few seconds, this problem will usually go away as the builder finishes & exits.

To speed things up, run make stop to tell the builder to hurry up and finish.

Improperly installed dependencies

You'll notice this by an error that looks like this in some module's logs:

2019-03-04T15:13:46.213763000Z internal/modules/cjs/loader.js:718
2019-03-04T15:13:46.213801600Z   return process.dlopen(module, path.toNamespacedPath(filename));
2019-03-04T15:13:46.213822300Z                  ^
2019-03-04T15:13:46.213842600Z
2019-03-04T15:13:46.213862700Z Error: Error loading shared library /root/node_modules/scrypt/build/Release/scrypt.node: Exec format error
2019-03-04T15:13:46.213882900Z     at Object.Module._extensions..node (internal/modules/cjs/loader.js:718:18)
2019-03-04T15:13:46.213903000Z     at Module.load (internal/modules/cjs/loader.js:599:32)
2019-03-04T15:13:46.213923100Z     at tryModuleLoad (internal/modules/cjs/loader.js:538:12)
2019-03-04T15:13:46.213943100Z     at Function.Module._load (internal/modules/cjs/loader.js:530:3)
2019-03-04T15:13:46.213963100Z     at Module.require (internal/modules/cjs/loader.js:637:17)
2019-03-04T15:13:46.213983100Z     at require (internal/modules/cjs/helpers.js:22:18)
2019-03-04T15:13:46.214003200Z     at Object.<anonymous> (/root/node_modules/scrypt/index.js:3:20)
2019-03-04T15:13:46.214023700Z     at Module._compile (internal/modules/cjs/loader.js:689:30)

If you noticed this error in the node, for example, you can reinstall dependencies by running make clean && make start.

This happen when you run npm install manually and then try to deploy the app using docker. Some dependencies (eg scrypt) have pieces in C that need to be compiled. If they get compiled for your local machine, they won't work in docker & vice versa.

Ethprovider or Ganache not working

cat -> curleth.sh <<EOF
#!/bin/bash
url=$ETH_PROVIDER; [[ $url ]] || url=http://localhost:8545
echo "Sending $1 query to provider: $url"
curl -H "Content-Type: application/json" -X POST --data '{"id":31415,"jsonrpc":"2.0","method":"'$1'","params":'$2'}' $url
EOF

This lets us do a simple bash curleth.sh net_version '[]' as a sanity check to make sure the ethprovider is alive & listening. If not, curl might give more useful errors that direct you towards investigating either metamask or ganache.

One other sanity check is to run docker service ls and make sure that you see an ethprovider service that has port 8545 exposed.

You can also run docker exec -it indra_ethprovider.1.<containerId> bash to start a shell inside the docker container. Even if there are networking issues between the container & host, you can still ping localhost:8545 here to see if ganache is listening & run ps to see if it's even alive.

Ganache should dump its logs onto your host and you can print/follow them with: tail -f modules/contracts/ops/ganache.log as another way to make sure it's alive. Try deleting this file then running npm restart to see if it gets recreated & if so, check to see if there is anything suspicious there

Have you tried turning it off and back on again?

Restarting: the debugger's most useful tool.

Some problems will be fixed by just restarting the app so try this first: make restart

If this doesn't work, try resetting all persistent data (database + the ethprovider's chain data) and starting the app again: make reset && npm start. After doing this, you'll likely need to reset your MetaMask account to get your tx nonces synced up correctly.

If that doesn't work either, try rebuilding everything with make clean && make start.