-
Notifications
You must be signed in to change notification settings - Fork 53
/
intercept.go
359 lines (318 loc) · 10.5 KB
/
intercept.go
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
package cortex
import (
"context"
"sync"
"time"
"unsafe"
"errors"
"github.com/cortexproject/cortex/pkg/cortexpb"
"github.com/weaveworks/common/user"
"google.golang.org/grpc"
"google.golang.org/grpc/metadata"
"google.golang.org/grpc/status"
)
type WriteHandlerFunc = func(context.Context, *cortexpb.WriteRequest, ...grpc.CallOption) (*cortexpb.WriteResponse, error)
type RequestInterceptor interface {
Intercept(ctx context.Context, req *cortexpb.WriteRequest, handler WriteHandlerFunc) (*cortexpb.WriteResponse, error)
}
type federatingInterceptor struct {
conf InterceptorConfig
mdTracker *metadataTracker
}
type InterceptorConfig struct {
// The label to use as the tenant ID
IdLabelName string
}
func NewFederatingInterceptor(conf InterceptorConfig) RequestInterceptor {
return &federatingInterceptor{
conf: conf,
mdTracker: newMetadataTracker(),
}
}
var emptyResponse = &cortexpb.WriteResponse{}
const (
lowerOrgIDHeaderName = "x-scope-orgid"
underscoreName = "__name__"
)
// Intercept takes a write request and splits it into multiple requests, grouped
// by tenant id found within the configured label name.
// The outgoingCtx must have an org id set, as well as outgoing metadata, to
// use as the default tenant id for timeseries without the configured label.
func (t *federatingInterceptor) Intercept(
outgoingCtx context.Context, req *cortexpb.WriteRequest,
handler WriteHandlerFunc,
) (*cortexpb.WriteResponse, error) {
if len(req.Metadata) > 0 {
// Metadata comes in periodically, separate from timeseries, and in one
// large batch. Because we are splitting timeseries samples to other
// tenants, we also need to replicate the metadata to those tenants.
// We need to keep track of names of metrics for which we have sent
// timeseries data to, so that we can send metadata for those metrics
// to the same tenants the next time metadata comes in.
var errs []error
_, err := handler(outgoingCtx, req)
if err != nil {
errs = append(errs, err)
}
errs = append(errs, t.mdTracker.replicate(req.Metadata, func(tenantId string, metricMetadata []*cortexpb.MetricMetadata) error {
req.Metadata = metricMetadata
md, _, _ := metadata.FromOutgoingContextRaw(outgoingCtx)
md[lowerOrgIDHeaderName] = []string{tenantId}
if _, err := handler(outgoingCtx, req); err != nil {
return err
}
return nil
})...)
return nil, errors.Join(errs...)
}
// for each time series in the request, check if there is a matching tenant ID label.
// if there is, start a new request for this and all subsequent entries with that tenant ID.
defaultId, err := user.ExtractOrgID(outgoingCtx)
if err != nil {
panic(err)
}
record := true
startTime := time.Now()
// save the original slice (header) over the whole timeseries array,
// so that we can restore it after modifying req.Timeseries later on.
allTimeseries := req.Timeseries
var errs []error
// partition the timeseries into buckets by tenant ID. The first partition is
// the default tenant ID, which is used for timeseries without the configured
// label. Subsequent partitions are for federated tenants.
PartitionById(reinterpretTimeseries(req.Timeseries),
// note: check gc annotations to make sure this function is inlined
func(idx int) string {
labels := req.Timeseries[idx].Labels
for i := len(labels) - 1; i >= 0; i-- {
if labels[i].Name == t.conf.IdLabelName {
req.Timeseries[idx].Labels = append(req.Timeseries[idx].Labels[:i], req.Timeseries[idx].Labels[i+1:]...)
return labels[i].Value
}
}
return ""
},
func(id string, start, end int) {
if record {
record = false
latencyPerTimeseries := time.Since(startTime) / time.Duration(len(req.Timeseries))
hRemoteWriteProcessingLatency.Record(outgoingCtx, latencyPerTimeseries.Nanoseconds())
cRemoteWriteTotalProcessedSeries.Add(outgoingCtx, int64(len(req.Timeseries)))
}
req.Timeseries = allTimeseries[start:end] // reuse the same request
md, _, _ := metadata.FromOutgoingContextRaw(outgoingCtx)
if id == "" {
id = defaultId
} else {
for _, ts := range req.Timeseries {
for _, l := range ts.Labels {
if l.Name == underscoreName {
t.mdTracker.track(l.Value, id)
break
}
}
}
}
if len(md[lowerOrgIDHeaderName]) == 1 {
md[lowerOrgIDHeaderName][0] = id
} else {
md[lowerOrgIDHeaderName] = []string{id}
}
if _, err := handler(outgoingCtx, req); err != nil {
errs = append(errs, err)
}
},
)
// restore the complete slice to be reused by the caller
req.Timeseries = allTimeseries
if len(errs) > 0 {
// if there was one error, return it
if len(errs) == 1 {
return nil, errs[0]
}
// it is likely that all errors are the same, and we can only return one
// error, so return the error with the largest code. The only actual
// scenario we want to catch here is if one of the errors is an http
// status code, in which case it will always be the chosen error.
//
toReturn := status.Convert(errs[0])
for _, err := range errs[1:] {
if statErr := status.Convert(err); statErr.Code() > toReturn.Code() {
toReturn = statErr
}
}
return nil, toReturn.Err()
}
return emptyResponse, nil
}
type passthroughInterceptor struct{}
func (t *passthroughInterceptor) Intercept(
ctx context.Context, req *cortexpb.WriteRequest,
handler WriteHandlerFunc,
) (*cortexpb.WriteResponse, error) {
return handler(ctx, req)
}
func (t *passthroughInterceptor) InterceptSlow(
ctx context.Context, req *cortexpb.WriteRequest,
handler WriteHandlerFunc,
) (*cortexpb.WriteResponse, error) {
return handler(ctx, req)
}
// tracks metric names for which corresponding metadata must be replicated,
// and to which tenants.
type metadataTracker struct {
mu sync.RWMutex
tenantTrackedMetrics map[string]map[string]struct{}
}
func newMetadataTracker() *metadataTracker {
return &metadataTracker{
tenantTrackedMetrics: make(map[string]map[string]struct{}),
}
}
func (mt *metadataTracker) replicate(metricMetadata []*cortexpb.MetricMetadata, replicateFn func(tenantId string, metricMetadata []*cortexpb.MetricMetadata) error) []error {
mt.mu.RLock()
defer mt.mu.RUnlock()
replications := make(map[string][]*cortexpb.MetricMetadata)
for _, md := range metricMetadata {
if tenantIds, ok := mt.tenantTrackedMetrics[md.MetricFamilyName]; ok {
for tenantId := range tenantIds {
replications[tenantId] = append(replications[tenantId], md)
}
}
}
var errs []error
for tenantId, metricMetadata := range replications {
if err := replicateFn(tenantId, metricMetadata); err != nil {
errs = append(errs, err)
}
}
return errs
}
func (mt *metadataTracker) track(metricName, tenantId string) {
mt.mu.Lock()
defer mt.mu.Unlock()
if tenants, ok := mt.tenantTrackedMetrics[metricName]; ok {
tenants[tenantId] = struct{}{}
} else {
mt.tenantTrackedMetrics[metricName] = map[string]struct{}{tenantId: {}}
}
}
func reinterpretTimeseries(ts []cortexpb.PreallocTimeseries) []*cortexpb.TimeSeries {
return *(*[]*cortexpb.TimeSeries)(unsafe.Pointer(&ts))
}
type Partition struct {
Id string
Start, End int
}
var partitionPool = sync.Pool{
New: func() interface{} {
return &Partition{}
},
}
var partitionSlicePool = sync.Pool{
New: func() interface{} {
p := make([]*Partition, 0, 32)
return &p
},
}
var lookupTablePool = sync.Pool{
New: func() interface{} {
return make(map[string]int, 32)
},
}
func reusePartition(part *Partition) {
part.Id = ""
part.Start = 0
part.End = 0
partitionPool.Put(part)
}
func reusePartitionSlice(parts *[]*Partition) {
*parts = (*parts)[:0]
partitionSlicePool.Put(parts)
}
func reuseLookupTable(table map[string]int) {
for k := range table {
delete(table, k)
}
lookupTablePool.Put(table)
}
// PartitionById arranges the elements of a slice into any number of partitions,
// where each partition contains elements grouped by a common ID. This function
// returns a list of partitions, where each has a start and end index and
// corresponding ID. The partition slice returned from this function should be
// reused by calling ReusePartitionSlice.
//
// The order of partitions (by ID) are as follows:
// 1. The first partition ID is the ID at index 0.
// 2. Subsequent partition IDs are in the inverse order of appearance when
// iterating backwards through the array.
//
// For example:
// IDs: A | D B A C A D A D E D D E B D E F C G D H A A H J D C C C I A
// Index: 0 | 9 8 7 6 5 4 3 2 1
// | -> partition order order of appearance <-
// Partition Order: A B E F G H J D C I
func PartitionById[T any](arr []*T, idForElement func(int) string, callback func(id string, start, end int)) {
if len(arr) == 0 {
return
}
firstId := idForElement(0)
partitions := partitionSlicePool.Get().(*[]*Partition)
first := Partition{
Id: firstId,
Start: 0,
End: len(arr),
}
lookup := lookupTablePool.Get().(map[string]int)
// loop backwards to improve locality for swaps
for i := len(arr) - 1; i >= 0; i-- {
tenantId := idForElement(i)
if tenantId == firstId {
continue
}
if i != first.End-1 {
// swap the element to the end of the first partition
ptrswap(&arr[i], &arr[first.End-1])
}
revIdx, ok := lookup[tenantId]
if !ok {
first.End--
// create a new partition
newPartition := partitionPool.Get().(*Partition)
newPartition.Id = tenantId
newPartition.Start = first.End
newPartition.End = first.End + 1
// move the last element of the first partition to the new partition
lookup[tenantId] = len(*partitions)
*partitions = append(*partitions, newPartition)
} else {
// append to existing partition by swapping the element to the end of
// each partition and shifting the boundary one element to the left
// start by swapping from the first partition into the most recently added
// partition (which is closest to the end of the first partition)
last := (*partitions)[len((*partitions))-1]
ptrswap(&arr[first.End-1], &arr[last.End-1])
first.End--
last.Start--
// then continue from the most recently added partition to the target
for p := len(*partitions) - 1; p > revIdx; p-- {
cur := (*partitions)[p]
next := (*partitions)[p-1]
ptrswap(&arr[(*partitions)[p].End-1], &arr[next.End-1])
cur.End--
next.Start--
}
}
}
callback(first.Id, first.Start, first.End)
for i := len((*partitions)) - 1; i >= 0; i-- {
part := (*partitions)[i]
callback(part.Id, part.Start, part.End)
reusePartition(part)
}
reuseLookupTable(lookup)
reusePartitionSlice(partitions)
}
func ptrswap[T any](a, b **T) {
*a, *b = *b, *a
}