Add HTTP Health Check Server

README.md

@@ -2,13 +2,11 @@
 
 [![Build Status](https://travis-ci.com/ChainSafe/ChainBridge.svg?branch=master)](https://travis-ci.com/ChainSafe/ChainBridge)
 
-<h3><b>[WIP]</b></h3>
 
 # Contents
 
 - [Installation](#installation)
 - [Configuration](#configuration)
-- [Running](#running)
 - [Chain Implementations](#chain-implementations)
 - [Testing](#testing)
 - [Simulations](#simulations)
@@ -97,6 +95,12 @@ To import private keys as keystores, use `chainbridge account import --privateKe
 
 For testing purposes, chainbridge provides 5 test keys. The can be used with `--testkey <name>`, where `name` is one of `Alice`, `Bob`, `Charlie`, `Dave`, or `Eve`. 
 
+## Metrics
+
+A basic health status check can be enabled with the `--metrics` flag (default port `8001`, use `--metricsPort` to specify).
+
+The endpoint `/health` will return the current block height and a timestamp of when it was processed. If the timestamp is at least 120 seconds old an error will be returned.
+
 # Chain Implementations
 
 - Ethereum (Solidity): [chainbridge-solidity](https://github.com/ChainSafe/chainbridge-solidity) 
@@ -109,7 +113,7 @@ For testing purposes, chainbridge provides 5 test keys. The can be used with `--
 
     A substrate pallet that can be integrated into a chain, as well as an example pallet to demonstrate chain integration.
 
-# MKdocs
+# Docs
 
 MKdocs will generate static HTML files for Chainsafe markdown files located in `Chainbridge/docs/`
 

chains/ethereum/chain.go

@@ -34,6 +34,7 @@ import (
 	"github.com/ChainSafe/ChainBridge/crypto/secp256k1"
 	"github.com/ChainSafe/ChainBridge/keystore"
 	msg "github.com/ChainSafe/ChainBridge/message"
+	metrics "github.com/ChainSafe/ChainBridge/metrics/types"
 	"github.com/ChainSafe/ChainBridge/router"
 	"github.com/ChainSafe/log15"
 	"github.com/ethereum/go-ethereum/accounts/abi/bind"
@@ -205,6 +206,10 @@ func (c *Chain) Name() string {
 	return c.cfg.Name
 }
 
+func (c *Chain) LatestBlock() metrics.LatestBlock {
+	return c.listener.latestBlock
+}
+
 // Stop signals to any running routines to exit
 func (c *Chain) Stop() {
 	close(c.stop)

chains/ethereum/listener.go

@@ -17,6 +17,7 @@ import (
 	"github.com/ChainSafe/ChainBridge/blockstore"
 	"github.com/ChainSafe/ChainBridge/chains"
 	msg "github.com/ChainSafe/ChainBridge/message"
+	metrics "github.com/ChainSafe/ChainBridge/metrics/types"
 	utils "github.com/ChainSafe/ChainBridge/shared/ethereum"
 	"github.com/ChainSafe/log15"
 	eth "github.com/ethereum/go-ethereum"
@@ -41,17 +42,19 @@ type listener struct {
 	blockstore             blockstore.Blockstorer
 	stop                   <-chan int
 	sysErr                 chan<- error // Reports fatal error to core
+	latestBlock            metrics.LatestBlock
 }
 
 // NewListener creates and returns a listener
 func NewListener(conn Connection, cfg *Config, log log15.Logger, bs blockstore.Blockstorer, stop <-chan int, sysErr chan<- error) *listener {
 	return &listener{
-		cfg:        *cfg,
-		conn:       conn,
-		log:        log,
-		blockstore: bs,
-		stop:       stop,
-		sysErr:     sysErr,
+		cfg:         *cfg,
+		conn:        conn,
+		log:         log,
+		blockstore:  bs,
+		stop:        stop,
+		sysErr:      sysErr,
+		latestBlock: metrics.LatestBlock{LastUpdated: time.Now()},
 	}
 }
 
@@ -132,6 +135,8 @@ func (l *listener) pollBlocks() error {
 
 			// Goto next block and reset retry counter
 			currentBlock.Add(currentBlock, big.NewInt(1))
+			l.latestBlock.Height = big.NewInt(0).Set(latestBlock)
+			l.latestBlock.LastUpdated = time.Now()
 			retry = BlockRetryLimit
 		}
 	}

chains/substrate/chain.go

@@ -29,6 +29,7 @@ import (
 	"github.com/ChainSafe/ChainBridge/crypto/sr25519"
 	"github.com/ChainSafe/ChainBridge/keystore"
 	msg "github.com/ChainSafe/ChainBridge/message"
+	metrics "github.com/ChainSafe/ChainBridge/metrics/types"
 	"github.com/ChainSafe/ChainBridge/router"
 	"github.com/ChainSafe/log15"
 )
@@ -132,6 +133,10 @@ func (c *Chain) SetRouter(r *router.Router) {
 	c.listener.setRouter(r)
 }
 
+func (c *Chain) LatestBlock() metrics.LatestBlock {
+	return c.listener.latestBlock
+}
+
 func (c *Chain) Id() msg.ChainId {
 	return c.cfg.Id
 }

chains/substrate/listener.go

@@ -12,6 +12,7 @@ import (
 	"github.com/ChainSafe/ChainBridge/blockstore"
 	"github.com/ChainSafe/ChainBridge/chains"
 	msg "github.com/ChainSafe/ChainBridge/message"
+	metrics "github.com/ChainSafe/ChainBridge/metrics/types"
 	utils "github.com/ChainSafe/ChainBridge/shared/substrate"
 	"github.com/ChainSafe/log15"
 	"github.com/centrifuge/go-substrate-rpc-client/types"
@@ -28,6 +29,7 @@ type listener struct {
 	log           log15.Logger
 	stop          <-chan int
 	sysErr        chan<- error
+	latestBlock   metrics.LatestBlock
 }
 
 // Frequency of polling for a new block
@@ -45,6 +47,7 @@ func NewListener(conn *Connection, name string, id msg.ChainId, startBlock uint6
 		log:           log,
 		stop:          stop,
 		sysErr:        sysErr,
+		latestBlock:   metrics.LatestBlock{LastUpdated: time.Now()},
 	}
 }
 
@@ -159,6 +162,8 @@ func (l *listener) pollBlocks() error {
 			}
 
 			currentBlock++
+			l.latestBlock.Height = big.NewInt(0).SetUint64(currentBlock)
+			l.latestBlock.LastUpdated = time.Now()
 			retry = BlockRetryLimit
 		}
 	}

cmd/chainbridge/main.go

@@ -17,6 +17,7 @@ import (
 	"github.com/ChainSafe/ChainBridge/config"
 	"github.com/ChainSafe/ChainBridge/core"
 	msg "github.com/ChainSafe/ChainBridge/message"
+	"github.com/ChainSafe/ChainBridge/metrics/health"
 	log "github.com/ChainSafe/log15"
 	"github.com/urfave/cli/v2"
 )
@@ -30,6 +31,8 @@ var cliFlags = []cli.Flag{
 	config.BlockstorePathFlag,
 	config.FreshStartFlag,
 	config.LatestBlockFlag,
+	config.MetricsFlag,
+	config.MetricsPort,
 }
 
 var generateFlags = []cli.Flag{
@@ -188,6 +191,12 @@ func run(ctx *cli.Context) error {
 		c.AddChain(newChain)
 	}
 
+	// Start metrics server
+	if ctx.Bool(config.MetricsFlag.Name) {
+		port := ctx.Int(config.MetricsPort.Name)
+		go health.Start(port, c.Registry)
+	}
+
 	c.Start()
 
 	return nil

config/flags.go

@@ -43,6 +43,20 @@ var (
 	}
 )
 
+// Metrics flags
+var (
+	MetricsFlag = &cli.BoolFlag{
+		Name:  "metrics",
+		Usage: "Enables metric server",
+	}
+
+	MetricsPort = &cli.IntFlag{
+		Name:  "metricsPort",
+		Usage: "Port to serve metrics on",
+		Value: 8001,
+	}
+)
+
 // Generate subcommand flags
 var (
 	PasswordFlag = &cli.StringFlag{

core/chain.go

@@ -5,6 +5,7 @@ package core
 
 import (
 	msg "github.com/ChainSafe/ChainBridge/message"
+	metrics "github.com/ChainSafe/ChainBridge/metrics/types"
 	"github.com/ChainSafe/ChainBridge/router"
 )
 
@@ -13,6 +14,7 @@ type Chain interface {
 	SetRouter(*router.Router)
 	Id() msg.ChainId
 	Name() string
+	LatestBlock() metrics.LatestBlock
 	Stop()
 }
 

core/core.go

@@ -9,37 +9,36 @@ import (
 	"os/signal"
 	"syscall"
 
-	msg "github.com/ChainSafe/ChainBridge/message"
 	"github.com/ChainSafe/ChainBridge/router"
 
 	"github.com/ChainSafe/log15"
 )
 
 type Core struct {
-	registry map[msg.ChainId]Chain
+	Registry []Chain
 	route    *router.Router
 	log      log15.Logger
 	sysErr   <-chan error
 }
 
 func NewCore(sysErr <-chan error) *Core {
 	return &Core{
-		registry: make(map[msg.ChainId]Chain),
+		Registry: make([]Chain, 0),
 		route:    router.NewRouter(log15.New("system", "router")),
 		log:      log15.New("system", "core"),
 		sysErr:   sysErr,
 	}
 }
 
-// AddChain registers the chain in the registry and calls Chain.SetRouter()
+// AddChain registers the chain in the Registry and calls Chain.SetRouter()
 func (c *Core) AddChain(chain Chain) {
-	c.registry[chain.Id()] = chain
+	c.Registry = append(c.Registry, chain)
 	chain.SetRouter(c.route)
 }
 
 // Start will call all registered chains' Start methods and block forever (or until signal is received)
 func (c *Core) Start() {
-	for _, chain := range c.registry {
+	for _, chain := range c.Registry {
 		err := chain.Start()
 		if err != nil {
 			c.log.Error(
@@ -65,7 +64,7 @@ func (c *Core) Start() {
 	}
 
 	// Signal chains to shutdown
-	for _, chain := range c.registry {
+	for _, chain := range c.Registry {
 		chain.Stop()
 	}
 }