RPC Gateway

RPC Gateway acts as a failover proxy routing ETH RPC requests across configured RPC nodes. For every ETH RPC node(group) configured the RPC Gateway tracks its latency, current height and error rates. These are then used to determine whether or not to failover.

From a high level it simply looks like this:

sequenceDiagram
Alice->>RPC Gateway: eth_call
loop Healthcheck
    RPC Gateway->>Alchemy: Check health
    RPC Gateway->>Infura: Check health
end
Note right of RPC Gateway: Routes only to healthy targets
loop Configurable Retries
RPC Gateway->>Alchemy: eth_call?
Alchemy-->>RPC Gateway: ERROR
end
Note right of RPC Gateway: RPC Call is rerouted after failing retries
RPC Gateway->>Infura: eth_call?
Infura-->>RPC Gateway: {"result":[...]}
RPC Gateway-->>Alice: {"result":[...]}

Loading

The gateway assesses the health of the underlying RPC provider by:

• continuously (configurable how often) checking the blockNumber, if the request fails or timeouts it marks it as unhealthy (configurable thresholds)
• every request that fails will be rerouted to the next available healthy target after a configurable amount of retries
  • if it will be rerouted the current target will be "tainted"

Developing

Start dependent services

docker-compose up

Make sure the test pass

go test

To run the app locally

go run . --config ./example_config.yml

Running & Configuration

Build the binary:

go build

The statically linked rpc-gateway binary has one flag --config that defaults to ./config.yml simply run it by:

./rpc-gateway --config ~/.rpc-gateway/config.yml

Configuration

metrics:
  port: "9090" # port for prometheus metrics, served on /metrics and /

proxy:
  port: "3000" # port for RPC gateway
  upstreamTimeout: "1s" # when is a request considered timed out

healthChecks:
  interval: "5s" # how often to do healthchecks
  timeout: "1s" # when should the timeout occur and considered unhealthy
  failureThreshold: 2 # how many failed checks until marked as unhealthy
  successThreshold: 1 # how many successes to be marked as healthy again

targets: # the order here determines the failover order
  - name: "Cloudflare"
    connection:
      http: # ws is supported by default, it will be a sticky connection.
        url: "https://cloudflare-eth.com"
  - name: "Alchemy"
    connection:
      http: # ws is supported by default, it will be a sticky connection.
        url: "https://alchemy.com/rpc/<apikey>"

Websockets

Websockets are sticky and are handled transparently.

Taints

Taints are a way for the HealthcheckManager to mark a node as unhealthy even though it responds to RPC calls. Some reasons for that are:

• BlockNumber is way behind a "quorum".
• A number of proxied requests fail in a given time.

Currently taint clearing is not implemented yet.

Build Docker images locally

We should build multi-arch image so the image can be run in both arm64 and amd64 arch.

TAG="$(git rev-parse HEAD)"
docker buildx build --platform linux/amd64,linux/arm64 -t 883408475785.dkr.ecr.us-east-1.amazonaws.com/rpc-gateway:${TAG} --push .