File: listoffsets.go

package info (click to toggle)
golang-github-segmentio-kafka-go 0.4.49%2Bds1-1
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid
  • size: 2,292 kB
  • sloc: sh: 17; makefile: 10
file content (230 lines) | stat: -rw-r--r-- 6,387 bytes parent folder | download
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
package listoffsets

import (
	"sort"

	"github.com/segmentio/kafka-go/protocol"
)

func init() {
	protocol.Register(&Request{}, &Response{})
}

type Request struct {
	ReplicaID      int32          `kafka:"min=v1,max=v5"`
	IsolationLevel int8           `kafka:"min=v2,max=v5"`
	Topics         []RequestTopic `kafka:"min=v1,max=v5"`
}

type RequestTopic struct {
	Topic      string             `kafka:"min=v1,max=v5"`
	Partitions []RequestPartition `kafka:"min=v1,max=v5"`
}

type RequestPartition struct {
	Partition          int32 `kafka:"min=v1,max=v5"`
	CurrentLeaderEpoch int32 `kafka:"min=v4,max=v5"`
	Timestamp          int64 `kafka:"min=v1,max=v5"`
	// v0 of the API predates kafka 0.10, and doesn't make much sense to
	// use so we chose not to support it. It had this extra field to limit
	// the number of offsets returned, which has been removed in v1.
	//
	// MaxNumOffsets int32 `kafka:"min=v0,max=v0"`
}

func (r *Request) ApiKey() protocol.ApiKey { return protocol.ListOffsets }

func (r *Request) Broker(cluster protocol.Cluster) (protocol.Broker, error) {
	// Expects r to be a request that was returned by Map, will likely panic
	// or produce the wrong result if that's not the case.
	partition := r.Topics[0].Partitions[0].Partition
	topic := r.Topics[0].Topic

	for _, p := range cluster.Topics[topic].Partitions {
		if p.ID == partition {
			return cluster.Brokers[p.Leader], nil
		}
	}

	return protocol.Broker{ID: -1}, nil
}

func (r *Request) Split(cluster protocol.Cluster) ([]protocol.Message, protocol.Merger, error) {
	// Because kafka refuses to answer ListOffsets requests containing multiple
	// entries of unique topic/partition pairs, we submit multiple requests on
	// the wire and merge their results back.
	//
	// ListOffsets requests also need to be sent to partition leaders, to keep
	// the logic simple we simply split each offset request into a single
	// message. This may cause a bit more requests to be sent on the wire but
	// it keeps the code sane, we can still optimize the aggregation mechanism
	// later if it becomes a problem.
	//
	// Really the idea here is to shield applications from having to deal with
	// the limitation of the kafka server, so they can request any combinations
	// of topic/partition/offsets.
	requests := make([]Request, 0, 2*len(r.Topics))

	for _, t := range r.Topics {
		for _, p := range t.Partitions {
			requests = append(requests, Request{
				ReplicaID:      r.ReplicaID,
				IsolationLevel: r.IsolationLevel,
				Topics: []RequestTopic{{
					Topic: t.Topic,
					Partitions: []RequestPartition{{
						Partition:          p.Partition,
						CurrentLeaderEpoch: p.CurrentLeaderEpoch,
						Timestamp:          p.Timestamp,
					}},
				}},
			})
		}
	}

	messages := make([]protocol.Message, len(requests))

	for i := range requests {
		messages[i] = &requests[i]
	}

	return messages, new(Response), nil
}

type Response struct {
	ThrottleTimeMs int32           `kafka:"min=v2,max=v5"`
	Topics         []ResponseTopic `kafka:"min=v1,max=v5"`
}

type ResponseTopic struct {
	Topic      string              `kafka:"min=v1,max=v5"`
	Partitions []ResponsePartition `kafka:"min=v1,max=v5"`
}

type ResponsePartition struct {
	Partition   int32 `kafka:"min=v1,max=v5"`
	ErrorCode   int16 `kafka:"min=v1,max=v5"`
	Timestamp   int64 `kafka:"min=v1,max=v5"`
	Offset      int64 `kafka:"min=v1,max=v5"`
	LeaderEpoch int32 `kafka:"min=v4,max=v5"`
}

func (r *Response) ApiKey() protocol.ApiKey { return protocol.ListOffsets }

func (r *Response) Merge(requests []protocol.Message, results []interface{}) (protocol.Message, error) {
	type topicPartition struct {
		topic     string
		partition int32
	}

	// Kafka doesn't always return the timestamp in the response, for example
	// when the request sends -2 (for the first offset) it always returns -1,
	// probably to indicate that the timestamp is unknown. This means that we
	// can't correlate the requests and responses based on their timestamps,
	// the primary key is the topic/partition pair.
	//
	// To make the API a bit friendly, we reconstructing an index of topic
	// partitions to the timestamps that were requested, and override the
	// timestamp value in the response.
	timestamps := make([]map[topicPartition]int64, len(requests))

	for i, m := range requests {
		req := m.(*Request)
		ts := make(map[topicPartition]int64, len(req.Topics))

		for _, t := range req.Topics {
			for _, p := range t.Partitions {
				ts[topicPartition{
					topic:     t.Topic,
					partition: p.Partition,
				}] = p.Timestamp
			}
		}

		timestamps[i] = ts
	}

	topics := make(map[string][]ResponsePartition)
	errors := 0

	for i, res := range results {
		m, err := protocol.Result(res)
		if err != nil {
			for _, t := range requests[i].(*Request).Topics {
				partitions := topics[t.Topic]

				for _, p := range t.Partitions {
					partitions = append(partitions, ResponsePartition{
						Partition:   p.Partition,
						ErrorCode:   -1, // UNKNOWN, can we do better?
						Timestamp:   -1,
						Offset:      -1,
						LeaderEpoch: -1,
					})
				}

				topics[t.Topic] = partitions
			}
			errors++
			continue
		}

		response := m.(*Response)

		if r.ThrottleTimeMs < response.ThrottleTimeMs {
			r.ThrottleTimeMs = response.ThrottleTimeMs
		}

		for _, t := range response.Topics {
			for _, p := range t.Partitions {
				if timestamp, ok := timestamps[i][topicPartition{
					topic:     t.Topic,
					partition: p.Partition,
				}]; ok {
					p.Timestamp = timestamp
				}
				topics[t.Topic] = append(topics[t.Topic], p)
			}
		}

	}

	if errors > 0 && errors == len(results) {
		_, err := protocol.Result(results[0])
		return nil, err
	}

	r.Topics = make([]ResponseTopic, 0, len(topics))

	for topicName, partitions := range topics {
		r.Topics = append(r.Topics, ResponseTopic{
			Topic:      topicName,
			Partitions: partitions,
		})
	}

	sort.Slice(r.Topics, func(i, j int) bool {
		return r.Topics[i].Topic < r.Topics[j].Topic
	})

	for _, t := range r.Topics {
		sort.Slice(t.Partitions, func(i, j int) bool {
			p1 := &t.Partitions[i]
			p2 := &t.Partitions[j]

			if p1.Partition != p2.Partition {
				return p1.Partition < p2.Partition
			}

			return p1.Offset < p2.Offset
		})
	}

	return r, nil
}

var (
	_ protocol.BrokerMessage = (*Request)(nil)
	_ protocol.Splitter      = (*Request)(nil)
	_ protocol.Merger        = (*Response)(nil)
)