summaryrefslogtreecommitdiff
path: root/modules/caddyhttp/reverseproxy/healthchecks.go
blob: ee56f33f85299201e70bcd476e95630a7437f666 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
// Copyright 2015 Matthew Holt and The Caddy Authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

package reverseproxy

import (
	"context"
	"fmt"
	"io"
	"io/ioutil"
	"log"
	"net"
	"net/http"
	"net/url"
	"regexp"
	"runtime/debug"
	"strconv"
	"time"

	"github.com/caddyserver/caddy/v2"
	"github.com/caddyserver/caddy/v2/modules/caddyhttp"
	"go.uber.org/zap"
)

// HealthChecks configures active and passive health checks.
type HealthChecks struct {
	// Active health checks run in the background on a timer. To
	// minimally enable active health checks, set either path or
	// port (or both).
	Active *ActiveHealthChecks `json:"active,omitempty"`

	// Passive health checks monitor proxied requests for errors or timeouts.
	// To minimally enable passive health checks, specify at least an empty
	// config object.
	Passive *PassiveHealthChecks `json:"passive,omitempty"`
}

// ActiveHealthChecks holds configuration related to active
// health checks (that is, health checks which occur in a
// background goroutine independently).
type ActiveHealthChecks struct {
	// The URI path to use for health checks.
	Path string `json:"path,omitempty"`

	// The port to use (if different from the upstream's dial
	// address) for health checks.
	Port int `json:"port,omitempty"`

	// HTTP headers to set on health check requests.
	Headers http.Header `json:"headers,omitempty"`

	// How frequently to perform active health checks (default 30s).
	Interval caddy.Duration `json:"interval,omitempty"`

	// How long to wait for a response from a backend before
	// considering it unhealthy (default 5s).
	Timeout caddy.Duration `json:"timeout,omitempty"`

	// The maximum response body to download from the backend
	// during a health check.
	MaxSize int64 `json:"max_size,omitempty"`

	// The HTTP status code to expect from a healthy backend.
	ExpectStatus int `json:"expect_status,omitempty"`

	// A regular expression against which to match the response
	// body of a healthy backend.
	ExpectBody string `json:"expect_body,omitempty"`

	stopChan   chan struct{}
	httpClient *http.Client
	bodyRegexp *regexp.Regexp
	logger     *zap.Logger
}

// PassiveHealthChecks holds configuration related to passive
// health checks (that is, health checks which occur during
// the normal flow of request proxying).
type PassiveHealthChecks struct {
	// How long to remember a failed request to a backend. A duration > 0
	// enables passive health checking. Default is 0.
	FailDuration caddy.Duration `json:"fail_duration,omitempty"`

	// The number of failed requests within the FailDuration window to
	// consider a backend as "down". Must be >= 1; default is 1. Requires
	// that FailDuration be > 0.
	MaxFails int `json:"max_fails,omitempty"`

	// Limits the number of simultaneous requests to a backend by
	// marking the backend as "down" if it has this many concurrent
	// requests or more.
	UnhealthyRequestCount int `json:"unhealthy_request_count,omitempty"`

	// Count the request as failed if the response comes back with
	// one of these status codes.
	UnhealthyStatus []int `json:"unhealthy_status,omitempty"`

	// Count the request as failed if the response takes at least this
	// long to receive.
	UnhealthyLatency caddy.Duration `json:"unhealthy_latency,omitempty"`

	logger *zap.Logger
}

// CircuitBreaker is a type that can act as an early-warning
// system for the health checker when backends are getting
// overloaded. This interface is still experimental and is
// subject to change.
type CircuitBreaker interface {
	OK() bool
	RecordMetric(statusCode int, latency time.Duration)
}

// activeHealthChecker runs active health checks on a
// regular basis and blocks until
// h.HealthChecks.Active.stopChan is closed.
func (h *Handler) activeHealthChecker() {
	defer func() {
		if err := recover(); err != nil {
			log.Printf("[PANIC] active health checks: %v\n%s", err, debug.Stack())
		}
	}()
	ticker := time.NewTicker(time.Duration(h.HealthChecks.Active.Interval))
	h.doActiveHealthCheckForAllHosts()
	for {
		select {
		case <-ticker.C:
			h.doActiveHealthCheckForAllHosts()
		case <-h.HealthChecks.Active.stopChan:
			// TODO: consider using a Context for cancellation instead
			ticker.Stop()
			return
		}
	}
}

// doActiveHealthCheckForAllHosts immediately performs a
// health checks for all upstream hosts configured by h.
func (h *Handler) doActiveHealthCheckForAllHosts() {
	for _, upstream := range h.Upstreams {
		go func(upstream *Upstream) {
			defer func() {
				if err := recover(); err != nil {
					log.Printf("[PANIC] active health check: %v\n%s", err, debug.Stack())
				}
			}()
			networkAddr := upstream.Dial
			addr, err := caddy.ParseNetworkAddress(networkAddr)
			if err != nil {
				h.HealthChecks.Active.logger.Error("bad network address",
					zap.String("address", networkAddr),
					zap.Error(err),
				)
				return
			}
			if addr.PortRangeSize() != 1 {
				h.HealthChecks.Active.logger.Error("multiple addresses (upstream must map to only one address)",
					zap.String("address", networkAddr),
				)
				return
			}
			hostAddr := addr.JoinHostPort(0)
			if addr.IsUnixNetwork() {
				// this will be used as the Host portion of a http.Request URL, and
				// paths to socket files would produce an error when creating URL,
				// so use a fake Host value instead; unix sockets are usually local
				hostAddr = "localhost"
			}
			err = h.doActiveHealthCheck(DialInfo{Network: addr.Network, Address: hostAddr}, hostAddr, upstream.Host)
			if err != nil {
				h.HealthChecks.Active.logger.Error("active health check failed",
					zap.String("address", networkAddr),
					zap.Error(err),
				)
			}
		}(upstream)
	}
}

// doActiveHealthCheck performs a health check to host which
// can be reached at address hostAddr. The actual address for
// the request will be built according to active health checker
// config. The health status of the host will be updated
// according to whether it passes the health check. An error is
// returned only if the health check fails to occur or if marking
// the host's health status fails.
func (h *Handler) doActiveHealthCheck(dialInfo DialInfo, hostAddr string, host Host) error {
	// create the URL for the request that acts as a health check
	scheme := "http"
	if ht, ok := h.Transport.(TLSTransport); ok && ht.TLSEnabled() {
		// this is kind of a hacky way to know if we should use HTTPS, but whatever
		scheme = "https"
	}
	u := &url.URL{
		Scheme: scheme,
		Host:   hostAddr,
		Path:   h.HealthChecks.Active.Path,
	}

	// adjust the port, if configured to be different
	if h.HealthChecks.Active.Port != 0 {
		portStr := strconv.Itoa(h.HealthChecks.Active.Port)
		host, _, err := net.SplitHostPort(hostAddr)
		if err != nil {
			host = hostAddr
		}
		u.Host = net.JoinHostPort(host, portStr)
	}

	// attach dialing information to this request - TODO: use caddy.Context's context
	// so it can be canceled on config reload
	ctx := context.Background()
	ctx = context.WithValue(ctx, caddy.ReplacerCtxKey, caddy.NewReplacer())
	ctx = context.WithValue(ctx, caddyhttp.VarsCtxKey, map[string]interface{}{
		dialInfoVarKey: dialInfo,
	})
	req, err := http.NewRequestWithContext(ctx, http.MethodGet, u.String(), nil)
	if err != nil {
		return fmt.Errorf("making request: %v", err)
	}
	for key, hdrs := range h.HealthChecks.Active.Headers {
		req.Header[key] = hdrs
	}

	// do the request, being careful to tame the response body
	resp, err := h.HealthChecks.Active.httpClient.Do(req)
	if err != nil {
		h.HealthChecks.Active.logger.Info("HTTP request failed",
			zap.String("host", hostAddr),
			zap.Error(err),
		)
		_, err2 := host.SetHealthy(false)
		if err2 != nil {
			return fmt.Errorf("marking unhealthy: %v", err2)
		}
		return nil
	}
	var body io.Reader = resp.Body
	if h.HealthChecks.Active.MaxSize > 0 {
		body = io.LimitReader(body, h.HealthChecks.Active.MaxSize)
	}
	defer func() {
		// drain any remaining body so connection could be re-used
		io.Copy(ioutil.Discard, body)
		resp.Body.Close()
	}()

	// if status code is outside criteria, mark down
	if h.HealthChecks.Active.ExpectStatus > 0 {
		if !caddyhttp.StatusCodeMatches(resp.StatusCode, h.HealthChecks.Active.ExpectStatus) {
			h.HealthChecks.Active.logger.Info("unexpected status code",
				zap.Int("status_code", resp.StatusCode),
				zap.String("host", hostAddr),
			)
			_, err := host.SetHealthy(false)
			if err != nil {
				return fmt.Errorf("marking unhealthy: %v", err)
			}
			return nil
		}
	} else if resp.StatusCode < 200 || resp.StatusCode >= 400 {
		h.HealthChecks.Active.logger.Info("status code out of tolerances",
			zap.Int("status_code", resp.StatusCode),
			zap.String("host", hostAddr),
		)
		_, err := host.SetHealthy(false)
		if err != nil {
			return fmt.Errorf("marking unhealthy: %v", err)
		}
		return nil
	}

	// if body does not match regex, mark down
	if h.HealthChecks.Active.bodyRegexp != nil {
		bodyBytes, err := ioutil.ReadAll(body)
		if err != nil {
			h.HealthChecks.Active.logger.Info("failed to read response body",
				zap.String("host", hostAddr),
				zap.Error(err),
			)
			_, err := host.SetHealthy(false)
			if err != nil {
				return fmt.Errorf("marking unhealthy: %v", err)
			}
			return nil
		}
		if !h.HealthChecks.Active.bodyRegexp.Match(bodyBytes) {
			h.HealthChecks.Active.logger.Info("response body failed expectations",
				zap.String("host", hostAddr),
			)
			_, err := host.SetHealthy(false)
			if err != nil {
				return fmt.Errorf("marking unhealthy: %v", err)
			}
			return nil
		}
	}

	// passed health check parameters, so mark as healthy
	swapped, err := host.SetHealthy(true)
	if swapped {
		h.HealthChecks.Active.logger.Info("host is up",
			zap.String("host", hostAddr),
		)
	}
	if err != nil {
		return fmt.Errorf("marking healthy: %v", err)
	}

	return nil
}

// countFailure is used with passive health checks. It
// remembers 1 failure for upstream for the configured
// duration. If passive health checks are disabled or
// failure expiry is 0, this is a no-op.
func (h *Handler) countFailure(upstream *Upstream) {
	// only count failures if passive health checking is enabled
	// and if failures are configured have a non-zero expiry
	if h.HealthChecks == nil || h.HealthChecks.Passive == nil {
		return
	}
	failDuration := time.Duration(h.HealthChecks.Passive.FailDuration)
	if failDuration == 0 {
		return
	}

	// count failure immediately
	err := upstream.Host.CountFail(1)
	if err != nil {
		h.HealthChecks.Passive.logger.Error("could not count failure",
			zap.String("host", upstream.Dial),
			zap.Error(err),
		)
	}

	// forget it later
	go func(host Host, failDuration time.Duration) {
		defer func() {
			if err := recover(); err != nil {
				log.Printf("[PANIC] health check failure forgetter: %v\n%s", err, debug.Stack())
			}
		}()
		time.Sleep(failDuration)
		err := host.CountFail(-1)
		if err != nil {
			h.HealthChecks.Passive.logger.Error("could not forget failure",
				zap.String("host", upstream.Dial),
				zap.Error(err),
			)
		}
	}(upstream.Host, failDuration)
}