Bancer

Bancer is a load balancer written in Node.js.

⚠ Disclaimer ⚠

This is a weekend project not meant to be used in production. I decided to take the load balancer as a subject to better understand the Node.js runtime and the JavaScript event loop.

Features

SSL Offloading

Bancer has built in support for TLS termination. Set the ssl field in the json config file and Bancer will spin up a HTTPS server. Bancer supports offloading standard 1024-bit and 2048-bit SSL keys.

Health Check and Retry

When a target server is unavailable or abruptly killed while requests are ditributed to the problematic server, Bancer will attempt to retry the failed connections to other available servers.

Configuration

{
  "id": "backend-service-id",
  "algorithm": "round-robin",
  "port": 4040,
  "workers": 4,
  "ssl": {
    "cert": "~path/to/server.crt",
    "key": "~path/to/server.key"
  },
  "servers": [
    {
      "id": "server-1",
      "host": "localhost",
      "port": 3001,
      "weight": 2,
      "livenessProbe": {
        "path": "/ping",
        "periodSeconds": 5
      }
    },
    {
      "id": "server-2",
      "host": "localhost",
      "port": 3002,
      "weight": 3,
      "livenessProbe": {
        "path": "/ping",
        "periodSeconds": 5
      }
    }
  ]
}

id -> string

Identifier for running multiple load balancer on the same thread

algorithm -> round-robin | least-connection | random [default: round-robin]

Defaults to round-robin.

port -> number

The port your load balancer is listening on.

workers -> number | "auto" [default: 1]

Built-in support for Node.js cluster mode. Defaults to 1. Maximum is number of available CPUs. If you set a number higher than there are available CPUs, bancer will gracefully fallback to the number of CPUs and show a warning.

servers.id -> string | undefined

Identifier for each target server. Must be unique across different targets. If you set the same id on 2 or more servers, bancer will throw an error and quit it's process.

servers.host -> number

Host the target server is running on.

servers.port -> number

Port number the target server is running on. Must be unique across target servers.

servers.weight -> numnber [default: 1]

The ratio of traffic routed to the target servers. Higher weight means higher number of request.

servers.livenessProbe

Liveness probe to know if a serve is ready to receive load. Periodical check will ensure traffic is directed at healthy servers only.

servers.livenessProbe.path

servers.livenessProbe.periodSeconds

Todo

• HTTP to HTTPS redirection
• Monitoring and Configuration UI
• Queue and retry if no servers available
• HTTP/2 Support
• Native IPv6 Support
• Content-based Routing (Host, Path, Headers, Method, Source IP, Device etc.)
• WebSocket Support
• Sticky Sessions
• Logging
• Slow Start Mode
• Fixed Response
• Data Compression