File size: 6,324 Bytes
d6f631f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
package flushhandler

import (
	"context"
	"errors"
	"fmt"
	"log/slog"
	"sync"
	"sync/atomic"
	"time"

	"go.opentelemetry.io/otel/metric"
	"go.opentelemetry.io/otel/trace"

	"github.com/openmeterio/openmeter/openmeter/sink/models"
)

const (
	defaultFlushChanSize   = 1000
	defaultCallbackTimeout = 30 * time.Second
)

type FlushEventHandlerOptions struct {
	Name     string
	Callback FlushCallback

	Logger      *slog.Logger
	MetricMeter metric.Meter

	DrainTimeout    time.Duration
	CallbackTimeout time.Duration
}

var _ FlushEventHandler = (*flushEventHandler)(nil)

type flushEventHandler struct {
	name string

	events      chan []models.SinkMessage
	eventsClose func()

	stopChan      chan struct{}
	stopChanClose func()

	drainDone      chan struct{}
	drainDoneClose func()

	callback FlushCallback

	callbackTimeout time.Duration
	drainTimeout    time.Duration

	metrics *metrics

	logger *slog.Logger

	isShutdown atomic.Bool
	mu         sync.Mutex
}

func NewFlushEventHandler(opts FlushEventHandlerOptions) (FlushEventHandler, error) {
	// validate options
	if opts.Name == "" {
		return nil, errors.New("name is required")
	}

	if opts.Callback == nil {
		return nil, errors.New("callback is required")
	}

	if opts.Logger == nil {
		return nil, errors.New("logger is required")
	}

	if opts.MetricMeter == nil {
		return nil, errors.New("metric meter is required")
	}

	if opts.CallbackTimeout == 0 {
		opts.CallbackTimeout = defaultCallbackTimeout
	}

	if opts.DrainTimeout == 0 {
		opts.DrainTimeout = defaultCallbackTimeout
	}

	// construct underlying object
	metrics, err := newMetrics(opts.Name, opts.MetricMeter)
	if err != nil {
		return nil, err
	}

	events := make(chan []models.SinkMessage, defaultFlushChanSize)
	eventsClose := sync.OnceFunc(func() {
		close(events)
	})

	stopChan := make(chan struct{})
	stopChanClose := sync.OnceFunc(func() {
		close(stopChan)
	})

	drainDone := make(chan struct{})
	drainDoneClose := sync.OnceFunc(func() {
		close(drainDone)
	})

	return &flushEventHandler{
		callback:        opts.Callback,
		callbackTimeout: opts.CallbackTimeout,
		drainTimeout:    opts.DrainTimeout,
		name:            opts.Name,
		events:          events,
		eventsClose:     eventsClose,
		stopChan:        stopChan,
		stopChanClose:   stopChanClose,
		drainDone:       drainDone,
		drainDoneClose:  drainDoneClose,
		metrics:         metrics,
		logger:          opts.Logger,
	}, nil
}

func (f *flushEventHandler) Close() error {
	if f.isShutdown.Swap(true) {
		return nil
	}

	// Close control channel
	f.stopChanClose()

	// Acquire lock to avoid closing events channel while there is an ongoing OnFlushSuccess operation
	f.mu.Lock()
	defer f.mu.Unlock()

	// Close events channel in order to avoid readers getting blocked
	f.eventsClose()

	return nil
}

func (f *flushEventHandler) Start(ctx context.Context) error {
	go f.start(ctx)
	return nil
}

func (f *flushEventHandler) start(ctx context.Context) {
	defer f.drainDoneClose()

	if f.isShutdown.Load() {
		f.logger.ErrorContext(ctx, "failed to start flush event handler as it is already shut down")
		return
	}

	// Capture the trace span from the start context so callbacks can be linked
	// to the parent trace even though they use context.Background() for cancellation isolation.
	parentSpan := trace.SpanFromContext(ctx)

	for !f.isShutdown.Load() {
		select {
		case event := <-f.events:
			if err := f.invokeCallbackWithTimeout(parentSpan, event); err != nil {
				f.logger.ErrorContext(ctx, "failed to invoke callback", "error", err)
			}
		case <-ctx.Done():
			_ = f.Close()

		case <-f.stopChan:
			_ = f.Close()
		}
	}

	// let's drain the queue using a new context, as the parent context is already canceled
	drainContext, cancel := context.WithTimeout(context.Background(), f.drainTimeout)
	defer cancel()

	// Attach trace context to drain context so drain callbacks are also linked to the parent trace.
	drainContext = trace.ContextWithSpan(drainContext, parentSpan)

	// NOTE: this will block if the events channel is not closed
	for event := range f.events {
		if err := f.invokeCallback(drainContext, event); err != nil {
			f.logger.ErrorContext(ctx, "failed to invoke callback", "error", err)
		}
	}
}

func (f *flushEventHandler) invokeCallbackWithTimeout(parentSpan trace.Span, events []models.SinkMessage) error {
	// We are using a background context here, as if the parent context is canceled, we still want to
	// allow the callbacks to call external systems. In exchange we are limiting the work with a timeout.
	ctx, cancel := context.WithTimeout(context.Background(), f.callbackTimeout)
	defer cancel()

	// Propagate trace context so callback spans are linked to the parent trace.
	ctx = trace.ContextWithSpan(ctx, parentSpan)

	return f.invokeCallback(ctx, events)
}

func (f *flushEventHandler) invokeCallback(ctx context.Context, events []models.SinkMessage) error {
	startTime := time.Now()

	if err := f.callback(ctx, events); err != nil {
		f.metrics.eventsFailed.Add(ctx, 1)
		return err
	}

	f.metrics.eventProcessingTime.Record(ctx, time.Since(startTime).Milliseconds())
	f.metrics.eventsProcessed.Add(ctx, 1)

	return nil
}

func (f *flushEventHandler) OnFlushSuccess(ctx context.Context, event []models.SinkMessage) error {
	if f.isShutdown.Load() {
		return errors.New("handler is shutting down")
	}

	f.mu.Lock()
	defer f.mu.Unlock()

	select {
	case <-f.stopChan:
		return fmt.Errorf("handler is shutting down")
	case f.events <- event:
		f.metrics.eventsReceived.Add(ctx, 1)
	case <-ctx.Done():
		f.metrics.eventsFailed.Add(ctx, 1)
		return fmt.Errorf("context canceled handler: %s", f.name)
	default:
		f.logger.ErrorContext(ctx, "flush handler: work queue full, callback might be hanging", "event", event, "name", f.name)
		f.metrics.eventChannelFull.Add(ctx, 1)

		select {
		case <-f.stopChan:
			return fmt.Errorf("handler is shutting down")
		case f.events <- event:
			f.metrics.eventsReceived.Add(ctx, 1)
		case <-ctx.Done():
			f.metrics.eventsFailed.Add(ctx, 1)
			return fmt.Errorf("context canceled handler: %s", f.name)
		}
	}

	return nil
}

func (f *flushEventHandler) WaitForDrain(ctx context.Context) error {
	select {
	case <-f.drainDone:
		return nil
	case <-ctx.Done():
		return fmt.Errorf("context canceled while wainting for drain in handler %s", f.name)
	}
}