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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387
|
package notifications
import (
"container/list"
"fmt"
"sync"
"time"
"github.com/sirupsen/logrus"
)
// NOTE(stevvooe): This file contains definitions for several utility sinks.
// Typically, the broadcaster is the only sink that should be required
// externally, but others are suitable for export if the need arises. Albeit,
// the tight integration with endpoint metrics should be removed.
// Broadcaster sends events to multiple, reliable Sinks. The goal of this
// component is to dispatch events to configured endpoints. Reliability can be
// provided by wrapping incoming sinks.
type Broadcaster struct {
sinks []Sink
events chan []Event
closed chan chan struct{}
}
// NewBroadcaster ...
// Add appends one or more sinks to the list of sinks. The broadcaster
// behavior will be affected by the properties of the sink. Generally, the
// sink should accept all messages and deal with reliability on its own. Use
// of EventQueue and RetryingSink should be used here.
func NewBroadcaster(sinks ...Sink) *Broadcaster {
b := Broadcaster{
sinks: sinks,
events: make(chan []Event),
closed: make(chan chan struct{}),
}
// Start the broadcaster
go b.run()
return &b
}
// Write accepts a block of events to be dispatched to all sinks. This method
// will never fail and should never block (hopefully!). The caller cedes the
// slice memory to the broadcaster and should not modify it after calling
// write.
func (b *Broadcaster) Write(events ...Event) error {
select {
case b.events <- events:
case <-b.closed:
return ErrSinkClosed
}
return nil
}
// Close the broadcaster, ensuring that all messages are flushed to the
// underlying sink before returning.
func (b *Broadcaster) Close() error {
logrus.Infof("broadcaster: closing")
select {
case <-b.closed:
// already closed
return fmt.Errorf("broadcaster: already closed")
default:
// do a little chan handoff dance to synchronize closing
closed := make(chan struct{})
b.closed <- closed
close(b.closed)
<-closed
return nil
}
}
// run is the main broadcast loop, started when the broadcaster is created.
// Under normal conditions, it waits for events on the event channel. After
// Close is called, this goroutine will exit.
func (b *Broadcaster) run() {
for {
select {
case block := <-b.events:
for _, sink := range b.sinks {
if err := sink.Write(block...); err != nil {
logrus.Errorf("broadcaster: error writing events to %v, these events will be lost: %v", sink, err)
}
}
case closing := <-b.closed:
// close all the underlying sinks
for _, sink := range b.sinks {
if err := sink.Close(); err != nil {
logrus.Errorf("broadcaster: error closing sink %v: %v", sink, err)
}
}
closing <- struct{}{}
logrus.Debugf("broadcaster: closed")
return
}
}
}
// eventQueue accepts all messages into a queue for asynchronous consumption
// by a sink. It is unbounded and thread safe but the sink must be reliable or
// events will be dropped.
type eventQueue struct {
sink Sink
events *list.List
listeners []eventQueueListener
cond *sync.Cond
mu sync.Mutex
closed bool
}
// eventQueueListener is called when various events happen on the queue.
type eventQueueListener interface {
ingress(events ...Event)
egress(events ...Event)
}
// newEventQueue returns a queue to the provided sink. If the updater is non-
// nil, it will be called to update pending metrics on ingress and egress.
func newEventQueue(sink Sink, listeners ...eventQueueListener) *eventQueue {
eq := eventQueue{
sink: sink,
events: list.New(),
listeners: listeners,
}
eq.cond = sync.NewCond(&eq.mu)
go eq.run()
return &eq
}
// Write accepts the events into the queue, only failing if the queue has
// beend closed.
func (eq *eventQueue) Write(events ...Event) error {
eq.mu.Lock()
defer eq.mu.Unlock()
if eq.closed {
return ErrSinkClosed
}
for _, listener := range eq.listeners {
listener.ingress(events...)
}
eq.events.PushBack(events)
eq.cond.Signal() // signal waiters
return nil
}
// Close shuts down the event queue, flushing
func (eq *eventQueue) Close() error {
eq.mu.Lock()
defer eq.mu.Unlock()
if eq.closed {
return fmt.Errorf("eventqueue: already closed")
}
// set closed flag
eq.closed = true
eq.cond.Signal() // signal flushes queue
eq.cond.Wait() // wait for signal from last flush
return eq.sink.Close()
}
// run is the main goroutine to flush events to the target sink.
func (eq *eventQueue) run() {
for {
block := eq.next()
if block == nil {
return // nil block means event queue is closed.
}
if err := eq.sink.Write(block...); err != nil {
logrus.Warnf("eventqueue: error writing events to %v, these events will be lost: %v", eq.sink, err)
}
for _, listener := range eq.listeners {
listener.egress(block...)
}
}
}
// next encompasses the critical section of the run loop. When the queue is
// empty, it will block on the condition. If new data arrives, it will wake
// and return a block. When closed, a nil slice will be returned.
func (eq *eventQueue) next() []Event {
eq.mu.Lock()
defer eq.mu.Unlock()
for eq.events.Len() < 1 {
if eq.closed {
eq.cond.Broadcast()
return nil
}
eq.cond.Wait()
}
front := eq.events.Front()
block := front.Value.([]Event)
eq.events.Remove(front)
return block
}
// ignoredSink discards events with ignored target media types and actions.
// passes the rest along.
type ignoredSink struct {
Sink
ignoreMediaTypes map[string]bool
ignoreActions map[string]bool
}
func newIgnoredSink(sink Sink, ignored []string, ignoreActions []string) Sink {
if len(ignored) == 0 {
return sink
}
ignoredMap := make(map[string]bool)
for _, mediaType := range ignored {
ignoredMap[mediaType] = true
}
ignoredActionsMap := make(map[string]bool)
for _, action := range ignoreActions {
ignoredActionsMap[action] = true
}
return &ignoredSink{
Sink: sink,
ignoreMediaTypes: ignoredMap,
ignoreActions: ignoredActionsMap,
}
}
// Write discards events with ignored target media types and passes the rest
// along.
func (imts *ignoredSink) Write(events ...Event) error {
var kept []Event
for _, e := range events {
if !imts.ignoreMediaTypes[e.Target.MediaType] {
kept = append(kept, e)
}
}
if len(kept) == 0 {
return nil
}
var results []Event
for _, e := range kept {
if !imts.ignoreActions[e.Action] {
results = append(results, e)
}
}
if len(results) == 0 {
return nil
}
return imts.Sink.Write(results...)
}
// retryingSink retries the write until success or an ErrSinkClosed is
// returned. Underlying sink must have p > 0 of succeeding or the sink will
// block. Internally, it is a circuit breaker retries to manage reset.
// Concurrent calls to a retrying sink are serialized through the sink,
// meaning that if one is in-flight, another will not proceed.
type retryingSink struct {
mu sync.Mutex
sink Sink
closed bool
// circuit breaker heuristics
failures struct {
threshold int
recent int
last time.Time
backoff time.Duration // time after which we retry after failure.
}
}
// TODO(stevvooe): We are using circuit break here, which actually doesn't
// make a whole lot of sense for this use case, since we always retry. Move
// this to use bounded exponential backoff.
// newRetryingSink returns a sink that will retry writes to a sink, backing
// off on failure. Parameters threshold and backoff adjust the behavior of the
// circuit breaker.
func newRetryingSink(sink Sink, threshold int, backoff time.Duration) *retryingSink {
rs := &retryingSink{
sink: sink,
}
rs.failures.threshold = threshold
rs.failures.backoff = backoff
return rs
}
// Write attempts to flush the events to the downstream sink until it succeeds
// or the sink is closed.
func (rs *retryingSink) Write(events ...Event) error {
rs.mu.Lock()
defer rs.mu.Unlock()
retry:
if rs.closed {
return ErrSinkClosed
}
if !rs.proceed() {
logrus.Warnf("%v encountered too many errors, backing off", rs.sink)
rs.wait(rs.failures.backoff)
goto retry
}
if err := rs.write(events...); err != nil {
if err == ErrSinkClosed {
// terminal!
return err
}
logrus.Errorf("retryingsink: error writing events: %v, retrying", err)
goto retry
}
return nil
}
// Close closes the sink and the underlying sink.
func (rs *retryingSink) Close() error {
rs.mu.Lock()
defer rs.mu.Unlock()
if rs.closed {
return fmt.Errorf("retryingsink: already closed")
}
rs.closed = true
return rs.sink.Close()
}
// write provides a helper that dispatches failure and success properly. Used
// by write as the single-flight write call.
func (rs *retryingSink) write(events ...Event) error {
if err := rs.sink.Write(events...); err != nil {
rs.failure()
return err
}
rs.reset()
return nil
}
// wait backoff time against the sink, unlocking so others can proceed. Should
// only be called by methods that currently have the mutex.
func (rs *retryingSink) wait(backoff time.Duration) {
rs.mu.Unlock()
defer rs.mu.Lock()
// backoff here
time.Sleep(backoff)
}
// reset marks a successful call.
func (rs *retryingSink) reset() {
rs.failures.recent = 0
rs.failures.last = time.Time{}
}
// failure records a failure.
func (rs *retryingSink) failure() {
rs.failures.recent++
rs.failures.last = time.Now().UTC()
}
// proceed returns true if the call should proceed based on circuit breaker
// heuristics.
func (rs *retryingSink) proceed() bool {
return rs.failures.recent < rs.failures.threshold ||
time.Now().UTC().After(rs.failures.last.Add(rs.failures.backoff))
}
|