From 1136e2cfeed5cc06a0811d39c2560c06f9dd74a2 Mon Sep 17 00:00:00 2001 From: Matthew Holt Date: Sat, 4 May 2019 10:49:50 -0600 Subject: Add reverse proxy --- modules/caddyhttp/reverseproxy/healthchecker.go | 64 +++ modules/caddyhttp/reverseproxy/module.go | 46 ++ modules/caddyhttp/reverseproxy/reverseproxy.go | 562 ++++++++++++++++++++++++ modules/caddyhttp/reverseproxy/upstream.go | 413 +++++++++++++++++ 4 files changed, 1085 insertions(+) create mode 100755 modules/caddyhttp/reverseproxy/healthchecker.go create mode 100755 modules/caddyhttp/reverseproxy/module.go create mode 100755 modules/caddyhttp/reverseproxy/reverseproxy.go create mode 100755 modules/caddyhttp/reverseproxy/upstream.go (limited to 'modules') diff --git a/modules/caddyhttp/reverseproxy/healthchecker.go b/modules/caddyhttp/reverseproxy/healthchecker.go new file mode 100755 index 0000000..add3aa0 --- /dev/null +++ b/modules/caddyhttp/reverseproxy/healthchecker.go @@ -0,0 +1,64 @@ +package reverseproxy + +import ( + "net/http" + "time" +) + +// Upstream represents the interface that must be satisfied to use the healthchecker. +type Upstream interface { + SetHealthiness(bool) +} + +// HealthChecker represents a worker that periodically evaluates if proxy upstream host is healthy. +type HealthChecker struct { + upstream Upstream + Ticker *time.Ticker + HTTPClient *http.Client +} + +// ScheduleChecks periodically runs health checks against an upstream host. +func (h *HealthChecker) ScheduleChecks(url string) { + // check if a host is healthy on start vs waiting for timer + h.upstream.SetHealthiness(h.IsHealthy(url)) + + for { + select { + case <-h.Ticker.C: + h.upstream.SetHealthiness(h.IsHealthy(url)) + } + } +} + +// Stop stops the healthchecker from makeing further requests. +func (h *HealthChecker) Stop() { + h.Ticker.Stop() +} + +// IsHealthy attempts to check if a upstream host is healthy. +func (h *HealthChecker) IsHealthy(url string) bool { + req, err := http.NewRequest(http.MethodGet, url, nil) + if err != nil { + return false + } + + resp, err := h.HTTPClient.Do(req) + if err != nil { + return false + } + + if resp.StatusCode < 200 || resp.StatusCode >= 400 { + return false + } + + return true +} + +// NewHealthCheckWorker returns a new instance of a HealthChecker. +func NewHealthCheckWorker(u Upstream, interval time.Duration, client *http.Client) *HealthChecker { + return &HealthChecker{ + upstream: u, + Ticker: time.NewTicker(interval), + HTTPClient: client, + } +} diff --git a/modules/caddyhttp/reverseproxy/module.go b/modules/caddyhttp/reverseproxy/module.go new file mode 100755 index 0000000..cc53bf5 --- /dev/null +++ b/modules/caddyhttp/reverseproxy/module.go @@ -0,0 +1,46 @@ +package reverseproxy + +import ( + "fmt" + + "bitbucket.org/lightcodelabs/caddy2" +) + +// Register caddy module. +func init() { + caddy2.RegisterModule(caddy2.Module{ + Name: "http.responders.reverse_proxy", + New: func() (interface{}, error) { return new(LoadBalanced), nil }, + OnLoad: func(instances []interface{}, _ interface{}) (interface{}, error) { + // we don't need to do anything with prior state because healthcheckers are + // cleaned up in OnUnload. + s := &State{ + HealthCheckers: []*HealthChecker{}, + } + + for _, i := range instances { + lb := i.(*LoadBalanced) + + err := NewLoadBalancedReverseProxy(lb, s) + if err != nil { + return nil, err + } + } + + return s, nil + }, + OnUnload: func(state interface{}) error { + s, ok := state.(*State) + if !ok { + return fmt.Errorf("proxy OnLoad: prior state not expected proxy.State type") + } + + // cleanup old healthcheckers + for _, hc := range s.HealthCheckers { + hc.Stop() + } + + return nil + }, + }) +} diff --git a/modules/caddyhttp/reverseproxy/reverseproxy.go b/modules/caddyhttp/reverseproxy/reverseproxy.go new file mode 100755 index 0000000..e5020e2 --- /dev/null +++ b/modules/caddyhttp/reverseproxy/reverseproxy.go @@ -0,0 +1,562 @@ +package reverseproxy + +import ( + "context" + "fmt" + "io" + "log" + "net" + "net/http" + "net/url" + "strings" + "sync" + "time" + + "golang.org/x/net/http/httpguts" +) + +// ReverseProxy is an HTTP Handler that takes an incoming request and +// sends it to another server, proxying the response back to the +// client. +type ReverseProxy struct { + // Director must be a function which modifies + // the request into a new request to be sent + // using Transport. Its response is then copied + // back to the original client unmodified. + // Director must not access the provided Request + // after returning. + Director func(*http.Request) + + // The transport used to perform proxy requests. + // If nil, http.DefaultTransport is used. + Transport http.RoundTripper + + // FlushInterval specifies the flush interval + // to flush to the client while copying the + // response body. + // If zero, no periodic flushing is done. + // A negative value means to flush immediately + // after each write to the client. + // The FlushInterval is ignored when ReverseProxy + // recognizes a response as a streaming response; + // for such responses, writes are flushed to the client + // immediately. + FlushInterval time.Duration + + // ErrorLog specifies an optional logger for errors + // that occur when attempting to proxy the request. + // If nil, logging goes to os.Stderr via the log package's + // standard logger. + ErrorLog *log.Logger + + // BufferPool optionally specifies a buffer pool to + // get byte slices for use by io.CopyBuffer when + // copying HTTP response bodies. + BufferPool BufferPool + + // ModifyResponse is an optional function that modifies the + // Response from the backend. It is called if the backend + // returns a response at all, with any HTTP status code. + // If the backend is unreachable, the optional ErrorHandler is + // called without any call to ModifyResponse. + // + // If ModifyResponse returns an error, ErrorHandler is called + // with its error value. If ErrorHandler is nil, its default + // implementation is used. + ModifyResponse func(*http.Response) error + + // ErrorHandler is an optional function that handles errors + // reaching the backend or errors from ModifyResponse. + // + // If nil, the default is to log the provided error and return + // a 502 Status Bad Gateway response. + ErrorHandler func(http.ResponseWriter, *http.Request, error) +} + +// A BufferPool is an interface for getting and returning temporary +// byte slices for use by io.CopyBuffer. +type BufferPool interface { + Get() []byte + Put([]byte) +} + +func singleJoiningSlash(a, b string) string { + aslash := strings.HasSuffix(a, "/") + bslash := strings.HasPrefix(b, "/") + switch { + case aslash && bslash: + return a + b[1:] + case !aslash && !bslash: + return a + "/" + b + } + return a + b +} + +// NewSingleHostReverseProxy returns a new ReverseProxy that routes +// URLs to the scheme, host, and base path provided in target. If the +// target's path is "/base" and the incoming request was for "/dir", +// the target request will be for /base/dir. +// NewSingleHostReverseProxy does not rewrite the Host header. +// To rewrite Host headers, use ReverseProxy directly with a custom +// Director policy. +func NewSingleHostReverseProxy(target *url.URL) *ReverseProxy { + targetQuery := target.RawQuery + director := func(req *http.Request) { + req.URL.Scheme = target.Scheme + req.URL.Host = target.Host + req.URL.Path = singleJoiningSlash(target.Path, req.URL.Path) + if targetQuery == "" || req.URL.RawQuery == "" { + req.URL.RawQuery = targetQuery + req.URL.RawQuery + } else { + req.URL.RawQuery = targetQuery + "&" + req.URL.RawQuery + } + if _, ok := req.Header["User-Agent"]; !ok { + // explicitly disable User-Agent so it's not set to default value + req.Header.Set("User-Agent", "") + } + } + return &ReverseProxy{Director: director} +} + +func copyHeader(dst, src http.Header) { + for k, vv := range src { + for _, v := range vv { + dst.Add(k, v) + } + } +} + +func cloneHeader(h http.Header) http.Header { + h2 := make(http.Header, len(h)) + for k, vv := range h { + vv2 := make([]string, len(vv)) + copy(vv2, vv) + h2[k] = vv2 + } + return h2 +} + +// Hop-by-hop headers. These are removed when sent to the backend. +// As of RFC 7230, hop-by-hop headers are required to appear in the +// Connection header field. These are the headers defined by the +// obsoleted RFC 2616 (section 13.5.1) and are used for backward +// compatibility. +var hopHeaders = []string{ + "Connection", + "Proxy-Connection", // non-standard but still sent by libcurl and rejected by e.g. google + "Keep-Alive", + "Proxy-Authenticate", + "Proxy-Authorization", + "Te", // canonicalized version of "TE" + "Trailer", // not Trailers per URL above; https://www.rfc-editor.org/errata_search.php?eid=4522 + "Transfer-Encoding", + "Upgrade", +} + +func (p *ReverseProxy) defaultErrorHandler(rw http.ResponseWriter, req *http.Request, err error) { + p.logf("http: proxy error: %v", err) + rw.WriteHeader(http.StatusBadGateway) +} + +func (p *ReverseProxy) getErrorHandler() func(http.ResponseWriter, *http.Request, error) { + if p.ErrorHandler != nil { + return p.ErrorHandler + } + return p.defaultErrorHandler +} + +// modifyResponse conditionally runs the optional ModifyResponse hook +// and reports whether the request should proceed. +func (p *ReverseProxy) modifyResponse(rw http.ResponseWriter, res *http.Response, req *http.Request) bool { + if p.ModifyResponse == nil { + return true + } + if err := p.ModifyResponse(res); err != nil { + res.Body.Close() + p.getErrorHandler()(rw, req, err) + return false + } + return true +} + +func (p *ReverseProxy) ServeHTTP(rw http.ResponseWriter, req *http.Request) (*http.Response, error) { + transport := p.Transport + if transport == nil { + transport = http.DefaultTransport + } + + ctx := req.Context() + if cn, ok := rw.(http.CloseNotifier); ok { + var cancel context.CancelFunc + ctx, cancel = context.WithCancel(ctx) + defer cancel() + notifyChan := cn.CloseNotify() + go func() { + select { + case <-notifyChan: + cancel() + case <-ctx.Done(): + } + }() + } + + outreq := req.WithContext(ctx) // includes shallow copies of maps, but okay + if req.ContentLength == 0 { + outreq.Body = nil // Issue 16036: nil Body for http.Transport retries + } + + outreq.Header = cloneHeader(req.Header) + + p.Director(outreq) + outreq.Close = false + + reqUpType := upgradeType(outreq.Header) + removeConnectionHeaders(outreq.Header) + + // Remove hop-by-hop headers to the backend. Especially + // important is "Connection" because we want a persistent + // connection, regardless of what the client sent to us. + for _, h := range hopHeaders { + hv := outreq.Header.Get(h) + if hv == "" { + continue + } + if h == "Te" && hv == "trailers" { + // Issue 21096: tell backend applications that + // care about trailer support that we support + // trailers. (We do, but we don't go out of + // our way to advertise that unless the + // incoming client request thought it was + // worth mentioning) + continue + } + outreq.Header.Del(h) + } + + // After stripping all the hop-by-hop connection headers above, add back any + // necessary for protocol upgrades, such as for websockets. + if reqUpType != "" { + outreq.Header.Set("Connection", "Upgrade") + outreq.Header.Set("Upgrade", reqUpType) + } + + if clientIP, _, err := net.SplitHostPort(req.RemoteAddr); err == nil { + // If we aren't the first proxy retain prior + // X-Forwarded-For information as a comma+space + // separated list and fold multiple headers into one. + if prior, ok := outreq.Header["X-Forwarded-For"]; ok { + clientIP = strings.Join(prior, ", ") + ", " + clientIP + } + outreq.Header.Set("X-Forwarded-For", clientIP) + } + + res, err := transport.RoundTrip(outreq) + if err != nil { + p.getErrorHandler()(rw, outreq, err) + return nil, err + } + + // Deal with 101 Switching Protocols responses: (WebSocket, h2c, etc) + if res.StatusCode == http.StatusSwitchingProtocols { + if !p.modifyResponse(rw, res, outreq) { + return res, nil + } + + p.handleUpgradeResponse(rw, outreq, res) + return res, nil + } + + removeConnectionHeaders(res.Header) + + for _, h := range hopHeaders { + res.Header.Del(h) + } + + if !p.modifyResponse(rw, res, outreq) { + return res, nil + } + + copyHeader(rw.Header(), res.Header) + + // The "Trailer" header isn't included in the Transport's response, + // at least for *http.Transport. Build it up from Trailer. + announcedTrailers := len(res.Trailer) + if announcedTrailers > 0 { + trailerKeys := make([]string, 0, len(res.Trailer)) + for k := range res.Trailer { + trailerKeys = append(trailerKeys, k) + } + rw.Header().Add("Trailer", strings.Join(trailerKeys, ", ")) + } + + rw.WriteHeader(res.StatusCode) + + err = p.copyResponse(rw, res.Body, p.flushInterval(req, res)) + if err != nil { + defer res.Body.Close() + // Since we're streaming the response, if we run into an error all we can do + // is abort the request. Issue 23643: ReverseProxy should use ErrAbortHandler + // on read error while copying body. + if !shouldPanicOnCopyError(req) { + p.logf("suppressing panic for copyResponse error in test; copy error: %v", err) + return nil, err + } + + panic(http.ErrAbortHandler) + } + res.Body.Close() // close now, instead of defer, to populate res.Trailer + + if len(res.Trailer) > 0 { + // Force chunking if we saw a response trailer. + // This prevents net/http from calculating the length for short + // bodies and adding a Content-Length. + if fl, ok := rw.(http.Flusher); ok { + fl.Flush() + } + } + + if len(res.Trailer) == announcedTrailers { + copyHeader(rw.Header(), res.Trailer) + return res, nil + } + + for k, vv := range res.Trailer { + k = http.TrailerPrefix + k + for _, v := range vv { + rw.Header().Add(k, v) + } + } + + return res, nil +} + +var inOurTests bool // whether we're in our own tests + +// shouldPanicOnCopyError reports whether the reverse proxy should +// panic with http.ErrAbortHandler. This is the right thing to do by +// default, but Go 1.10 and earlier did not, so existing unit tests +// weren't expecting panics. Only panic in our own tests, or when +// running under the HTTP server. +func shouldPanicOnCopyError(req *http.Request) bool { + if inOurTests { + // Our tests know to handle this panic. + return true + } + if req.Context().Value(http.ServerContextKey) != nil { + // We seem to be running under an HTTP server, so + // it'll recover the panic. + return true + } + // Otherwise act like Go 1.10 and earlier to not break + // existing tests. + return false +} + +// removeConnectionHeaders removes hop-by-hop headers listed in the "Connection" header of h. +// See RFC 7230, section 6.1 +func removeConnectionHeaders(h http.Header) { + if c := h.Get("Connection"); c != "" { + for _, f := range strings.Split(c, ",") { + if f = strings.TrimSpace(f); f != "" { + h.Del(f) + } + } + } +} + +// flushInterval returns the p.FlushInterval value, conditionally +// overriding its value for a specific request/response. +func (p *ReverseProxy) flushInterval(req *http.Request, res *http.Response) time.Duration { + resCT := res.Header.Get("Content-Type") + + // For Server-Sent Events responses, flush immediately. + // The MIME type is defined in https://www.w3.org/TR/eventsource/#text-event-stream + if resCT == "text/event-stream" { + return -1 // negative means immediately + } + + // TODO: more specific cases? e.g. res.ContentLength == -1? + return p.FlushInterval +} + +func (p *ReverseProxy) copyResponse(dst io.Writer, src io.Reader, flushInterval time.Duration) error { + if flushInterval != 0 { + if wf, ok := dst.(writeFlusher); ok { + mlw := &maxLatencyWriter{ + dst: wf, + latency: flushInterval, + } + defer mlw.stop() + dst = mlw + } + } + + var buf []byte + if p.BufferPool != nil { + buf = p.BufferPool.Get() + defer p.BufferPool.Put(buf) + } + _, err := p.copyBuffer(dst, src, buf) + return err +} + +// copyBuffer returns any write errors or non-EOF read errors, and the amount +// of bytes written. +func (p *ReverseProxy) copyBuffer(dst io.Writer, src io.Reader, buf []byte) (int64, error) { + if len(buf) == 0 { + buf = make([]byte, 32*1024) + } + var written int64 + for { + nr, rerr := src.Read(buf) + if rerr != nil && rerr != io.EOF && rerr != context.Canceled { + p.logf("httputil: ReverseProxy read error during body copy: %v", rerr) + } + if nr > 0 { + nw, werr := dst.Write(buf[:nr]) + if nw > 0 { + written += int64(nw) + } + if werr != nil { + return written, werr + } + if nr != nw { + return written, io.ErrShortWrite + } + } + if rerr != nil { + if rerr == io.EOF { + rerr = nil + } + return written, rerr + } + } +} + +func (p *ReverseProxy) logf(format string, args ...interface{}) { + if p.ErrorLog != nil { + p.ErrorLog.Printf(format, args...) + } else { + log.Printf(format, args...) + } +} + +type writeFlusher interface { + io.Writer + http.Flusher +} + +type maxLatencyWriter struct { + dst writeFlusher + latency time.Duration // non-zero; negative means to flush immediately + + mu sync.Mutex // protects t, flushPending, and dst.Flush + t *time.Timer + flushPending bool +} + +func (m *maxLatencyWriter) Write(p []byte) (n int, err error) { + m.mu.Lock() + defer m.mu.Unlock() + n, err = m.dst.Write(p) + if m.latency < 0 { + m.dst.Flush() + return + } + if m.flushPending { + return + } + if m.t == nil { + m.t = time.AfterFunc(m.latency, m.delayedFlush) + } else { + m.t.Reset(m.latency) + } + m.flushPending = true + return +} + +func (m *maxLatencyWriter) delayedFlush() { + m.mu.Lock() + defer m.mu.Unlock() + if !m.flushPending { // if stop was called but AfterFunc already started this goroutine + return + } + m.dst.Flush() + m.flushPending = false +} + +func (m *maxLatencyWriter) stop() { + m.mu.Lock() + defer m.mu.Unlock() + m.flushPending = false + if m.t != nil { + m.t.Stop() + } +} + +func upgradeType(h http.Header) string { + if !httpguts.HeaderValuesContainsToken(h["Connection"], "Upgrade") { + return "" + } + return strings.ToLower(h.Get("Upgrade")) +} + +func (p *ReverseProxy) handleUpgradeResponse(rw http.ResponseWriter, req *http.Request, res *http.Response) { + reqUpType := upgradeType(req.Header) + resUpType := upgradeType(res.Header) + if reqUpType != resUpType { + p.getErrorHandler()(rw, req, fmt.Errorf("backend tried to switch protocol %q when %q was requested", resUpType, reqUpType)) + return + } + + copyHeader(res.Header, rw.Header()) + + hj, ok := rw.(http.Hijacker) + if !ok { + p.getErrorHandler()(rw, req, fmt.Errorf("can't switch protocols using non-Hijacker ResponseWriter type %T", rw)) + return + } + backConn, ok := res.Body.(io.ReadWriteCloser) + if !ok { + p.getErrorHandler()(rw, req, fmt.Errorf("internal error: 101 switching protocols response with non-writable body")) + return + } + defer backConn.Close() + conn, brw, err := hj.Hijack() + if err != nil { + p.getErrorHandler()(rw, req, fmt.Errorf("Hijack failed on protocol switch: %v", err)) + return + } + defer conn.Close() + res.Body = nil // so res.Write only writes the headers; we have res.Body in backConn above + if err := res.Write(brw); err != nil { + p.getErrorHandler()(rw, req, fmt.Errorf("response write: %v", err)) + return + } + if err := brw.Flush(); err != nil { + p.getErrorHandler()(rw, req, fmt.Errorf("response flush: %v", err)) + return + } + errc := make(chan error, 1) + spc := switchProtocolCopier{user: conn, backend: backConn} + go spc.copyToBackend(errc) + go spc.copyFromBackend(errc) + <-errc + return +} + +// switchProtocolCopier exists so goroutines proxying data back and +// forth have nice names in stacks. +type switchProtocolCopier struct { + user, backend io.ReadWriter +} + +func (c switchProtocolCopier) copyFromBackend(errc chan<- error) { + _, err := io.Copy(c.user, c.backend) + errc <- err +} + +func (c switchProtocolCopier) copyToBackend(errc chan<- error) { + _, err := io.Copy(c.backend, c.user) + errc <- err +} diff --git a/modules/caddyhttp/reverseproxy/upstream.go b/modules/caddyhttp/reverseproxy/upstream.go new file mode 100755 index 0000000..b521d46 --- /dev/null +++ b/modules/caddyhttp/reverseproxy/upstream.go @@ -0,0 +1,413 @@ +// Package reverseproxy implements a load-balanced reverse proxy. +package reverseproxy + +import ( + "context" + "encoding/json" + "fmt" + "math/rand" + "net" + "net/http" + "net/url" + "strings" + "sync" + "sync/atomic" + "time" + + "bitbucket.org/lightcodelabs/caddy2" +) + +// State represents the global state of a loadbalancer. It is used to store +// references to health checkers. +type State struct { + HealthCheckers []*HealthChecker +} + +// CircuitBreaker defines the functionality of a circuit breaker module. +type CircuitBreaker interface { + Ok() bool + RecordMetric(statusCode int, latency time.Duration) +} + +type noopCircuitBreaker struct{} + +func (ncb noopCircuitBreaker) RecordMetric(statusCode int, latency time.Duration) {} +func (ncb noopCircuitBreaker) Ok() bool { + return true +} + +const ( + // TypeBalanceRoundRobin represents the value to use for configuring a load balanced reverse proxy to use round robin load balancing. + TypeBalanceRoundRobin = iota + + // TypeBalanceRandom represents the value to use for configuring a load balanced reverse proxy to use random load balancing. + TypeBalanceRandom + + // msgNoHealthyUpstreams is returned if there are no upstreams that are healthy to proxy a request to + msgNoHealthyUpstreams = "No healthy upstreams." + + // by default perform health checks every 30 seconds + defaultHealthCheckDur = time.Second * 30 + + // used when an upstream is unhealthy, health checks can be configured to perform at a faster rate + defaultFastHealthCheckDur = time.Second * 1 +) + +var ( + // defaultTransport is the default transport to use for the reverse proxy. + defaultTransport = &http.Transport{ + Dial: (&net.Dialer{ + Timeout: 5 * time.Second, + }).Dial, + TLSHandshakeTimeout: 5 * time.Second, + } + + // defaultHTTPClient is the default http client to use for the healthchecker. + defaultHTTPClient = &http.Client{ + Timeout: time.Second * 10, + Transport: defaultTransport, + } + + // typeMap maps caddy load balance configuration to the internal representation of the loadbalance algorithm type. + typeMap = map[string]int{ + "round_robin": TypeBalanceRoundRobin, + "random": TypeBalanceRandom, + } +) + +// NewLoadBalancedReverseProxy returns a collection of Upstreams that are to be loadbalanced. +func NewLoadBalancedReverseProxy(lb *LoadBalanced, state *State) error { + // set defaults + if lb.NoHealthyUpstreamsMessage == "" { + lb.NoHealthyUpstreamsMessage = msgNoHealthyUpstreams + } + + if lb.TryInterval == "" { + lb.TryInterval = "20s" + } + + // set request retry interval + ti, err := time.ParseDuration(lb.TryInterval) + if err != nil { + return fmt.Errorf("NewLoadBalancedReverseProxy: %v", err.Error()) + } + lb.tryInterval = ti + + // set load balance algorithm + t, ok := typeMap[lb.LoadBalanceType] + if !ok { + t = TypeBalanceRandom + } + lb.loadBalanceType = t + + // setup each upstream + var us []*upstream + for _, uc := range lb.Upstreams { + // pass the upstream decr and incr methods to keep track of unhealthy nodes + nu, err := newUpstream(uc, lb.decrUnhealthy, lb.incrUnhealthy) + if err != nil { + return err + } + + // setup any configured circuit breakers + var cbModule = "http.responders.reverse_proxy.circuit_breaker" + var cb CircuitBreaker + + if uc.CircuitBreaker != nil { + if _, err := caddy2.GetModule(cbModule); err == nil { + val, err := caddy2.LoadModule(cbModule, uc.CircuitBreaker) + if err == nil { + cbv, ok := val.(CircuitBreaker) + if ok { + cb = cbv + } else { + fmt.Printf("\nerr: %v; cannot load circuit_breaker, using noop", err.Error()) + cb = noopCircuitBreaker{} + } + } else { + fmt.Printf("\nerr: %v; cannot load circuit_breaker, using noop", err.Error()) + cb = noopCircuitBreaker{} + } + } else { + fmt.Println("circuit_breaker module not loaded, using noop") + cb = noopCircuitBreaker{} + } + } else { + cb = noopCircuitBreaker{} + } + nu.CB = cb + + // start a healthcheck worker which will periodically check to see if an upstream is healthy + // to proxy requests to. + nu.healthChecker = NewHealthCheckWorker(nu, defaultHealthCheckDur, defaultHTTPClient) + + // TODO :- if path is empty why does this empty the entire Target? + // nu.Target.Path = uc.HealthCheckPath + + go nu.healthChecker.ScheduleChecks(nu.Target.String()) + state.HealthCheckers = append(state.HealthCheckers, nu.healthChecker) + + us = append(us, nu) + } + + lb.upstreams = us + + return nil +} + +// LoadBalanced represents a collection of upstream hosts that are loadbalanced. It +// contains multiple features like health checking and circuit breaking functionality +// for upstreams. +type LoadBalanced struct { + mu sync.Mutex + numUnhealthy int32 + selectedServer int // used during round robin load balancing + loadBalanceType int + tryInterval time.Duration + upstreams []*upstream + + // The following struct fields are set by caddy configuration. + // TryInterval is the max duration for which request retrys will be performed for a request. + TryInterval string `json:"try_interval"` + + // Upstreams are the configs for upstream hosts + Upstreams []*UpstreamConfig `json:"upstreams"` + + // LoadBalanceType is the string representation of what loadbalancing algorithm to use. i.e. "random" or "round_robin". + LoadBalanceType string `json:"load_balance_type"` + + // NoHealthyUpstreamsMessage is returned as a response when there are no healthy upstreams to loadbalance to. + NoHealthyUpstreamsMessage string `json:"no_healthy_upstreams_message"` +} + +// ServeHTTP implements the http.Handler interface to dispatch an http request to the proper +// server. +func (lb *LoadBalanced) ServeHTTP(w http.ResponseWriter, r *http.Request) error { + // ensure requests don't hang if an upstream does not respond or is not eventually healthy + var u *upstream + var done bool + + retryTimer := time.NewTicker(lb.tryInterval) + defer retryTimer.Stop() + + go func() { + select { + case <-retryTimer.C: + done = true + } + }() + + // keep trying to get an available upstream to process the request + for { + switch lb.loadBalanceType { + case TypeBalanceRandom: + u = lb.random() + case TypeBalanceRoundRobin: + u = lb.roundRobin() + } + + // if we can't get an upstream and our retry interval has ended return an error response + if u == nil && done { + w.WriteHeader(http.StatusBadGateway) + fmt.Fprint(w, lb.NoHealthyUpstreamsMessage) + + return fmt.Errorf(msgNoHealthyUpstreams) + } + + // attempt to get an available upstream + if u == nil { + continue + } + + start := time.Now() + + // if we get an error retry until we get a healthy upstream + res, err := u.ReverseProxy.ServeHTTP(w, r) + if err != nil { + if err == context.Canceled { + return nil + } + + continue + } + + // record circuit breaker metrics + go u.CB.RecordMetric(res.StatusCode, time.Now().Sub(start)) + + return nil + } +} + +// incrUnhealthy increments the amount of unhealthy nodes in a loadbalancer. +func (lb *LoadBalanced) incrUnhealthy() { + atomic.AddInt32(&lb.numUnhealthy, 1) +} + +// decrUnhealthy decrements the amount of unhealthy nodes in a loadbalancer. +func (lb *LoadBalanced) decrUnhealthy() { + atomic.AddInt32(&lb.numUnhealthy, -1) +} + +// roundRobin implements a round robin load balancing algorithm to select +// which server to forward requests to. +func (lb *LoadBalanced) roundRobin() *upstream { + if atomic.LoadInt32(&lb.numUnhealthy) == int32(len(lb.upstreams)) { + return nil + } + + selected := lb.upstreams[lb.selectedServer] + + lb.mu.Lock() + lb.selectedServer++ + if lb.selectedServer >= len(lb.upstreams) { + lb.selectedServer = 0 + } + lb.mu.Unlock() + + if selected.IsHealthy() && selected.CB.Ok() { + return selected + } + + return nil +} + +// random implements a random server selector for load balancing. +func (lb *LoadBalanced) random() *upstream { + if atomic.LoadInt32(&lb.numUnhealthy) == int32(len(lb.upstreams)) { + return nil + } + + n := rand.Int() % len(lb.upstreams) + selected := lb.upstreams[n] + + if selected.IsHealthy() && selected.CB.Ok() { + return selected + } + + return nil +} + +// UpstreamConfig represents the config of an upstream. +type UpstreamConfig struct { + // Host is the host name of the upstream server. + Host string `json:"host"` + + // FastHealthCheckDuration is the duration for which a health check is performed when a node is considered unhealthy. + FastHealthCheckDuration string `json:"fast_health_check_duration"` + + CircuitBreaker json.RawMessage `json:"circuit_breaker"` + + // // CircuitBreakerConfig is the config passed to setup a circuit breaker. + // CircuitBreakerConfig *circuitbreaker.Config `json:"circuit_breaker"` + circuitbreaker CircuitBreaker + + // HealthCheckDuration is the default duration for which a health check is performed. + HealthCheckDuration string `json:"health_check_duration"` + + // HealthCheckPath is the path at the upstream host to use for healthchecks. + HealthCheckPath string `json:"health_check_path"` +} + +// upstream represents an upstream host. +type upstream struct { + Healthy int32 // 0 = false, 1 = true + Target *url.URL + ReverseProxy *ReverseProxy + Incr func() + Decr func() + CB CircuitBreaker + healthChecker *HealthChecker + healthCheckDur time.Duration + fastHealthCheckDur time.Duration +} + +// newUpstream returns a new upstream. +func newUpstream(uc *UpstreamConfig, d func(), i func()) (*upstream, error) { + host := strings.TrimSpace(uc.Host) + protoIdx := strings.Index(host, "://") + if protoIdx == -1 || len(host[:protoIdx]) == 0 { + return nil, fmt.Errorf("protocol is required for host") + } + + hostURL, err := url.Parse(host) + if err != nil { + return nil, err + } + + // parse healthcheck durations + hcd, err := time.ParseDuration(uc.HealthCheckDuration) + if err != nil { + hcd = defaultHealthCheckDur + } + + fhcd, err := time.ParseDuration(uc.FastHealthCheckDuration) + if err != nil { + fhcd = defaultFastHealthCheckDur + } + + u := upstream{ + healthCheckDur: hcd, + fastHealthCheckDur: fhcd, + Target: hostURL, + Decr: d, + Incr: i, + Healthy: int32(0), // assume is unhealthy on start + } + + u.ReverseProxy = newReverseProxy(hostURL, u.SetHealthiness) + return &u, nil +} + +// SetHealthiness sets whether an upstream is healthy or not. The health check worker is updated to +// perform checks faster if a node is unhealthy. +func (u *upstream) SetHealthiness(ok bool) { + h := atomic.LoadInt32(&u.Healthy) + var wasHealthy bool + if h == 1 { + wasHealthy = true + } else { + wasHealthy = false + } + + if ok { + u.healthChecker.Ticker = time.NewTicker(u.healthCheckDur) + + if !wasHealthy { + atomic.AddInt32(&u.Healthy, 1) + u.Decr() + } + } else { + u.healthChecker.Ticker = time.NewTicker(u.fastHealthCheckDur) + + if wasHealthy { + atomic.AddInt32(&u.Healthy, -1) + u.Incr() + } + } +} + +// IsHealthy returns whether an Upstream is healthy or not. +func (u *upstream) IsHealthy() bool { + i := atomic.LoadInt32(&u.Healthy) + if i == 1 { + return true + } + + return false +} + +// newReverseProxy returns a new reverse proxy handler. +func newReverseProxy(target *url.URL, setHealthiness func(bool)) *ReverseProxy { + errorHandler := func(w http.ResponseWriter, r *http.Request, err error) { + // we don't need to worry about cancelled contexts since this doesn't necessarilly mean that + // the upstream is unhealthy. + if err != context.Canceled { + setHealthiness(false) + } + } + + rp := NewSingleHostReverseProxy(target) + rp.ErrorHandler = errorHandler + rp.Transport = defaultTransport // use default transport that times out in 5 seconds + return rp +} -- cgit v1.2.3