-
Notifications
You must be signed in to change notification settings - Fork 806
/
bucket_stores.go
419 lines (356 loc) · 13.8 KB
/
bucket_stores.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
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
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
package storegateway
import (
"context"
"fmt"
"path/filepath"
"strings"
"sync"
"time"
"github.com/go-kit/kit/log"
"github.com/go-kit/kit/log/level"
"github.com/oklog/ulid"
"github.com/pkg/errors"
"github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/client_golang/prometheus/promauto"
tsdb_errors "github.com/prometheus/prometheus/tsdb/errors"
"github.com/thanos-io/thanos/pkg/block"
thanos_metadata "github.com/thanos-io/thanos/pkg/block/metadata"
"github.com/thanos-io/thanos/pkg/extprom"
"github.com/thanos-io/thanos/pkg/gate"
"github.com/thanos-io/thanos/pkg/objstore"
"github.com/thanos-io/thanos/pkg/store"
storecache "github.com/thanos-io/thanos/pkg/store/cache"
"github.com/thanos-io/thanos/pkg/store/storepb"
"github.com/weaveworks/common/logging"
"google.golang.org/grpc/metadata"
"github.com/cortexproject/cortex/pkg/storage/tsdb"
"github.com/cortexproject/cortex/pkg/util"
"github.com/cortexproject/cortex/pkg/util/spanlogger"
"github.com/cortexproject/cortex/pkg/util/validation"
)
// BucketStores is a multi-tenant wrapper of Thanos BucketStore.
type BucketStores struct {
logger log.Logger
cfg tsdb.BlocksStorageConfig
limits *validation.Overrides
bucket objstore.Bucket
logLevel logging.Level
bucketStoreMetrics *BucketStoreMetrics
metaFetcherMetrics *MetadataFetcherMetrics
shardingStrategy ShardingStrategy
// Index cache shared across all tenants.
indexCache storecache.IndexCache
// Gate used to limit query concurrency across all tenants.
queryGate gate.Gate
// Keeps a bucket store for each tenant.
storesMu sync.RWMutex
stores map[string]*store.BucketStore
// Metrics.
syncTimes prometheus.Histogram
syncLastSuccess prometheus.Gauge
tenantsDiscovered prometheus.Gauge
tenantsSynced prometheus.Gauge
}
// NewBucketStores makes a new BucketStores.
func NewBucketStores(cfg tsdb.BlocksStorageConfig, shardingStrategy ShardingStrategy, bucketClient objstore.Bucket, limits *validation.Overrides, logLevel logging.Level, logger log.Logger, reg prometheus.Registerer) (*BucketStores, error) {
cachingBucket, err := tsdb.CreateCachingBucket(cfg.BucketStore.ChunksCache, cfg.BucketStore.MetadataCache, bucketClient, logger, reg)
if err != nil {
return nil, errors.Wrapf(err, "create caching bucket")
}
// The number of concurrent queries against the tenants BucketStores are limited.
queryGateReg := extprom.WrapRegistererWithPrefix("cortex_bucket_stores_", reg)
queryGate := gate.NewKeeper(queryGateReg).NewGate(cfg.BucketStore.MaxConcurrent)
promauto.With(reg).NewGauge(prometheus.GaugeOpts{
Name: "cortex_bucket_stores_gate_queries_concurrent_max",
Help: "Number of maximum concurrent queries allowed.",
}).Set(float64(cfg.BucketStore.MaxConcurrent))
u := &BucketStores{
logger: logger,
cfg: cfg,
limits: limits,
bucket: cachingBucket,
shardingStrategy: shardingStrategy,
stores: map[string]*store.BucketStore{},
logLevel: logLevel,
bucketStoreMetrics: NewBucketStoreMetrics(),
metaFetcherMetrics: NewMetadataFetcherMetrics(),
queryGate: queryGate,
syncTimes: promauto.With(reg).NewHistogram(prometheus.HistogramOpts{
Name: "cortex_bucket_stores_blocks_sync_seconds",
Help: "The total time it takes to perform a sync stores",
Buckets: []float64{0.1, 1, 10, 30, 60, 120, 300, 600, 900},
}),
syncLastSuccess: promauto.With(reg).NewGauge(prometheus.GaugeOpts{
Name: "cortex_bucket_stores_blocks_last_successful_sync_timestamp_seconds",
Help: "Unix timestamp of the last successful blocks sync.",
}),
tenantsDiscovered: promauto.With(reg).NewGauge(prometheus.GaugeOpts{
Name: "cortex_bucket_stores_tenants_discovered",
Help: "Number of tenants discovered in the bucket.",
}),
tenantsSynced: promauto.With(reg).NewGauge(prometheus.GaugeOpts{
Name: "cortex_bucket_stores_tenants_synced",
Help: "Number of tenants synced.",
}),
}
// Init the index cache.
if u.indexCache, err = tsdb.NewIndexCache(cfg.BucketStore.IndexCache, logger, reg); err != nil {
return nil, errors.Wrap(err, "create index cache")
}
if reg != nil {
reg.MustRegister(u.bucketStoreMetrics, u.metaFetcherMetrics)
}
return u, nil
}
// InitialSync does an initial synchronization of blocks for all users.
func (u *BucketStores) InitialSync(ctx context.Context) error {
level.Info(u.logger).Log("msg", "synchronizing TSDB blocks for all users")
if err := u.syncUsersBlocks(ctx, func(ctx context.Context, s *store.BucketStore) error {
return s.InitialSync(ctx)
}); err != nil {
level.Warn(u.logger).Log("msg", "failed to synchronize TSDB blocks", "err", err)
return err
}
level.Info(u.logger).Log("msg", "successfully synchronized TSDB blocks for all users")
return nil
}
// SyncBlocks synchronizes the stores state with the Bucket store for every user.
func (u *BucketStores) SyncBlocks(ctx context.Context) error {
return u.syncUsersBlocks(ctx, func(ctx context.Context, s *store.BucketStore) error {
return s.SyncBlocks(ctx)
})
}
func (u *BucketStores) syncUsersBlocks(ctx context.Context, f func(context.Context, *store.BucketStore) error) (returnErr error) {
defer func(start time.Time) {
u.syncTimes.Observe(time.Since(start).Seconds())
if returnErr == nil {
u.syncLastSuccess.SetToCurrentTime()
}
}(time.Now())
type job struct {
userID string
store *store.BucketStore
}
wg := &sync.WaitGroup{}
jobs := make(chan job)
errs := tsdb_errors.MultiError{}
errsMx := sync.Mutex{}
// Scan users in the bucket. In case of error, it may return a subset of users. If we sync a subset of users
// during a periodic sync, we may end up unloading blocks for users that still belong to this store-gateway
// so we do prefer to not run the sync at all.
userIDs, err := u.scanUsers(ctx)
if err != nil {
return err
}
includeUserIDs := make(map[string]struct{})
for _, userID := range u.shardingStrategy.FilterUsers(ctx, userIDs) {
includeUserIDs[userID] = struct{}{}
}
u.tenantsDiscovered.Set(float64(len(userIDs)))
u.tenantsSynced.Set(float64(len(includeUserIDs)))
// Create a pool of workers which will synchronize blocks. The pool size
// is limited in order to avoid to concurrently sync a lot of tenants in
// a large cluster.
for i := 0; i < u.cfg.BucketStore.TenantSyncConcurrency; i++ {
wg.Add(1)
go func() {
defer wg.Done()
for job := range jobs {
if err := f(ctx, job.store); err != nil {
errsMx.Lock()
errs.Add(errors.Wrapf(err, "failed to synchronize TSDB blocks for user %s", job.userID))
errsMx.Unlock()
}
}
}()
}
// Lazily create a bucket store for each new user found
// and submit a sync job for each user.
for _, userID := range userIDs {
// If we don't have a store for the tenant yet, then we should skip it if it's not
// included in the store-gateway shard. If we already have it, we need to sync it
// anyway to make sure all its blocks are unloaded and metrics updated correctly
// (but bucket API calls are skipped thanks to the objstore client adapter).
if _, included := includeUserIDs[userID]; !included && u.getStore(userID) == nil {
continue
}
bs, err := u.getOrCreateStore(userID)
if err != nil {
errsMx.Lock()
errs.Add(err)
errsMx.Unlock()
continue
}
select {
case jobs <- job{userID: userID, store: bs}:
// Nothing to do. Will loop to push more jobs.
case <-ctx.Done():
return ctx.Err()
}
}
// Wait until all workers completed.
close(jobs)
wg.Wait()
return errs.Err()
}
// Series makes a series request to the underlying user bucket store.
func (u *BucketStores) Series(req *storepb.SeriesRequest, srv storepb.Store_SeriesServer) error {
spanLog, spanCtx := spanlogger.New(srv.Context(), "BucketStores.Series")
defer spanLog.Span.Finish()
userID := getUserIDFromGRPCContext(spanCtx)
if userID == "" {
return fmt.Errorf("no userID")
}
store := u.getStore(userID)
if store == nil {
return nil
}
return store.Series(req, spanSeriesServer{
Store_SeriesServer: srv,
ctx: spanCtx,
})
}
// scanUsers in the bucket and return the list of found users. If an error occurs while
// iterating the bucket, it may return both an error and a subset of the users in the bucket.
func (u *BucketStores) scanUsers(ctx context.Context) ([]string, error) {
var users []string
// Iterate the bucket to find all users in the bucket. Due to how the bucket listing
// caching works, it's more likely to have a cache hit if there's no delay while
// iterating the bucket, so we do load all users in memory and later process them.
err := u.bucket.Iter(ctx, "", func(s string) error {
users = append(users, strings.TrimSuffix(s, "/"))
return nil
})
return users, err
}
func (u *BucketStores) getStore(userID string) *store.BucketStore {
u.storesMu.RLock()
store := u.stores[userID]
u.storesMu.RUnlock()
return store
}
func (u *BucketStores) getOrCreateStore(userID string) (*store.BucketStore, error) {
// Check if the store already exists.
bs := u.getStore(userID)
if bs != nil {
return bs, nil
}
u.storesMu.Lock()
defer u.storesMu.Unlock()
// Check again for the store in the event it was created in-between locks.
bs = u.stores[userID]
if bs != nil {
return bs, nil
}
userLogger := util.WithUserID(userID, u.logger)
level.Info(userLogger).Log("msg", "creating user bucket store")
userBkt := tsdb.NewUserBucketClient(userID, u.bucket)
// Wrap the bucket reader to skip iterating the bucket at all if the user doesn't
// belong to the store-gateway shard. We need to run the BucketStore synching anyway
// in order to unload previous tenants in case of a resharding leading to tenants
// moving out from the store-gateway shard and also make sure both MetaFetcher and
// BucketStore metrics are correctly updated.
fetcherBkt := NewShardingBucketReaderAdapter(userID, u.shardingStrategy, userBkt)
fetcherReg := prometheus.NewRegistry()
fetcher, err := block.NewMetaFetcher(
userLogger,
u.cfg.BucketStore.MetaSyncConcurrency,
fetcherBkt,
filepath.Join(u.cfg.BucketStore.SyncDir, userID), // The fetcher stores cached metas in the "meta-syncer/" sub directory
fetcherReg,
// The sharding strategy filter MUST be before the ones we create here (order matters).
append([]block.MetadataFilter{NewShardingMetadataFilterAdapter(userID, u.shardingStrategy)}, []block.MetadataFilter{
block.NewConsistencyDelayMetaFilter(userLogger, u.cfg.BucketStore.ConsistencyDelay, fetcherReg),
block.NewIgnoreDeletionMarkFilter(userLogger, userBkt, u.cfg.BucketStore.IgnoreDeletionMarksDelay),
// The duplicate filter has been intentionally omitted because it could cause troubles with
// the consistency check done on the querier. The duplicate filter removes redundant blocks
// but if the store-gateway removes redundant blocks before the querier discovers them, the
// consistency check on the querier will fail.
}...),
[]block.MetadataModifier{
// Remove Cortex external labels so that they're not injected when querying blocks.
NewReplicaLabelRemover(userLogger, []string{
tsdb.TenantIDExternalLabel,
tsdb.IngesterIDExternalLabel,
tsdb.ShardIDExternalLabel,
}),
},
)
if err != nil {
return nil, err
}
bucketStoreReg := prometheus.NewRegistry()
bs, err = store.NewBucketStore(
userLogger,
bucketStoreReg,
userBkt,
fetcher,
filepath.Join(u.cfg.BucketStore.SyncDir, userID),
u.indexCache,
u.queryGate,
u.cfg.BucketStore.MaxChunkPoolBytes,
newChunksLimiterFactory(u.limits, userID),
u.logLevel.String() == "debug", // Turn on debug logging, if the log level is set to debug
u.cfg.BucketStore.BlockSyncConcurrency,
nil, // Do not limit timerange.
false, // No need to enable backward compatibility with Thanos pre 0.8.0 queriers
u.cfg.BucketStore.IndexCache.PostingsCompression,
u.cfg.BucketStore.PostingOffsetsInMemSampling,
true, // Enable series hints.
)
if err != nil {
return nil, err
}
u.stores[userID] = bs
u.metaFetcherMetrics.AddUserRegistry(userID, fetcherReg)
u.bucketStoreMetrics.AddUserRegistry(userID, bucketStoreReg)
return bs, nil
}
func getUserIDFromGRPCContext(ctx context.Context) string {
meta, ok := metadata.FromIncomingContext(ctx)
if !ok {
return ""
}
values := meta.Get(tsdb.TenantIDExternalLabel)
if len(values) != 1 {
return ""
}
return values[0]
}
// ReplicaLabelRemover is a BaseFetcher modifier modifies external labels of existing blocks, it removes given replica labels from the metadata of blocks that have it.
type ReplicaLabelRemover struct {
logger log.Logger
replicaLabels []string
}
// NewReplicaLabelRemover creates a ReplicaLabelRemover.
func NewReplicaLabelRemover(logger log.Logger, replicaLabels []string) *ReplicaLabelRemover {
return &ReplicaLabelRemover{logger: logger, replicaLabels: replicaLabels}
}
// Modify modifies external labels of existing blocks, it removes given replica labels from the metadata of blocks that have it.
func (r *ReplicaLabelRemover) Modify(_ context.Context, metas map[ulid.ULID]*thanos_metadata.Meta, modified *extprom.TxGaugeVec) error {
for u, meta := range metas {
l := meta.Thanos.Labels
for _, replicaLabel := range r.replicaLabels {
if _, exists := l[replicaLabel]; exists {
level.Debug(r.logger).Log("msg", "replica label removed", "label", replicaLabel)
delete(l, replicaLabel)
}
}
metas[u].Thanos.Labels = l
}
return nil
}
type spanSeriesServer struct {
storepb.Store_SeriesServer
ctx context.Context
}
func (s spanSeriesServer) Context() context.Context {
return s.ctx
}
func newChunksLimiterFactory(limits *validation.Overrides, userID string) store.ChunksLimiterFactory {
return func(failedCounter prometheus.Counter) store.ChunksLimiter {
// Since limit overrides could be live reloaded, we have to get the current user's limit
// each time a new limiter is instantiated.
return store.NewLimiter(uint64(limits.MaxChunksPerQuery(userID)), failedCounter)
}
}