feat: Blooms retention

docs/sources/configure/_index.md

@@ -2726,6 +2726,15 @@ ring:
 # and compact as many tables.
 # CLI flag: -bloom-compactor.max-compaction-parallelism
 [max_compaction_parallelism: <int> | default = 1]
+
+retention:
+  # Enable bloom retention.
+  # CLI flag: -bloom-compactor.retention.enabled
+  [enabled: <boolean> | default = false]
+
+  # Max lookback days for retention.
+  # CLI flag: -bloom-compactor.retention.max-lookback-days
+  [max_lookback_days: <int> | default = 365]
 ```
 
 ### limits_config

pkg/bloomcompactor/bloomcompactor.go

@@ -48,7 +48,8 @@ type Compactor struct {
 
 	tsdbStore TSDBStore
 	// TODO(owen-d): ShardingStrategy
-	controller *SimpleBloomController
+	controller       *SimpleBloomController
+	retentionManager *RetentionManager
 
 	// temporary workaround until bloomStore has implemented read/write shipper interface
 	bloomStore bloomshipper.Store
@@ -65,7 +66,8 @@ func New(
 	storeCfg storage.Config,
 	clientMetrics storage.ClientMetrics,
 	fetcherProvider stores.ChunkFetcherProvider,
-	sharding util_ring.TenantSharding,
+	ring ring.ReadRing,
+	ringLifeCycler *ring.BasicLifecycler,
 	limits Limits,
 	store bloomshipper.Store,
 	logger log.Logger,
@@ -75,7 +77,7 @@ func New(
 		cfg:        cfg,
 		schemaCfg:  schemaCfg,
 		logger:     logger,
-		sharding:   sharding,
+		sharding:   util_ring.NewTenantShuffleSharding(ring, ringLifeCycler, limits.BloomCompactorShardSize),
 		limits:     limits,
 		bloomStore: store,
 	}
@@ -104,6 +106,15 @@ func New(
 		c.logger,
 	)
 
+	c.retentionManager = NewRetentionManager(
+		c.cfg.RetentionConfig,
+		c.limits,
+		c.bloomStore,
+		newFirstTokenRetentionSharding(ring, ringLifeCycler),
+		c.metrics,
+		c.logger,
+	)
+
 	c.Service = services.NewBasicService(c.starting, c.running, c.stopping)
 	return c, nil
 }
@@ -218,10 +229,17 @@ func (c *Compactor) runOne(ctx context.Context) error {
 	c.metrics.compactionsStarted.Inc()
 	start := time.Now()
 	level.Info(c.logger).Log("msg", "running bloom compaction", "workers", c.cfg.WorkerParallelism)
-	var workersErr error
+	var workersErr, retentionErr error
 	var wg sync.WaitGroup
 	input := make(chan *tenantTableRange)
 
+	// Launch retention (will return instantly if retention is disabled or not owned by this compactor)
+	wg.Add(1)
+	go func() {
+		retentionErr = c.retentionManager.Apply(ctx)
+		wg.Done()
+	}()
+
 	tables := c.tables(time.Now())
 	level.Debug(c.logger).Log("msg", "loaded tables", "tables", tables.TotalDays())
 
@@ -240,7 +258,7 @@ func (c *Compactor) runOne(ctx context.Context) error {
 
 	wg.Wait()
 	duration := time.Since(start)
-	err = multierror.New(workersErr, err, ctx.Err()).Err()
+	err = multierror.New(retentionErr, workersErr, err, ctx.Err()).Err()
 
 	if err != nil {
 		level.Error(c.logger).Log("msg", "compaction iteration failed", "err", err, "duration", duration)

pkg/bloomcompactor/bloomcompactor_test.go

@@ -148,6 +148,14 @@ type mockLimits struct {
 	shardSize int
 }
 
+func (m mockLimits) RetentionPeriod(_ string) time.Duration {
+	panic("implement me")
+}
+
+func (m mockLimits) StreamRetention(_ string) []validation.StreamRetention {
+	panic("implement me")
+}
+
 func (m mockLimits) AllByUserID() map[string]*validation.Limits {
 	panic("implement me")
 }

pkg/bloomcompactor/config.go

@@ -3,10 +3,10 @@ package bloomcompactor
 import (
 	"flag"
 	"fmt"
-	"github.com/pkg/errors"
 	"time"
 
-	"github.com/grafana/loki/pkg/storage/stores/shipper/indexshipper/downloads"
+	"github.com/pkg/errors"
+
 	"github.com/grafana/loki/pkg/util/ring"
 )
 
@@ -31,6 +31,8 @@ type Config struct {
 	CompactionRetries        int           `yaml:"compaction_retries"`
 
 	MaxCompactionParallelism int `yaml:"max_compaction_parallelism"`
+
+	RetentionConfig RetentionConfig `yaml:"retention"`
 }
 
 // RegisterFlags registers flags for the Bloom-Compactor configuration.
@@ -52,6 +54,7 @@ func (cfg *Config) RegisterFlags(f *flag.FlagSet) {
 	f.DurationVar(&cfg.RetryMaxBackoff, "bloom-compactor.compaction-retries-max-backoff", time.Minute, "Maximum backoff time between retries.")
 	f.IntVar(&cfg.CompactionRetries, "bloom-compactor.compaction-retries", 3, "Number of retries to perform when compaction fails.")
 	f.IntVar(&cfg.MaxCompactionParallelism, "bloom-compactor.max-compaction-parallelism", 1, "Maximum number of tables to compact in parallel. While increasing this value, please make sure compactor has enough disk space allocated to be able to store and compact as many tables.")
+	cfg.RetentionConfig.RegisterFlags(f)
 
 	// Ring
 	skipFlags := []string{
@@ -76,7 +79,7 @@ func (cfg *Config) Validate() error {
 }
 
 type Limits interface {
-	downloads.Limits
+	RetentionLimits
 	BloomCompactorShardSize(tenantID string) int
 	BloomCompactorEnabled(tenantID string) bool
 	BloomNGramLength(tenantID string) int

pkg/bloomcompactor/metrics.go

@@ -43,6 +43,12 @@ type Metrics struct {
 
 	progress      prometheus.Gauge
 	timePerTenant *prometheus.CounterVec
+
+	// Retention metrics
+	retentionRunning             prometheus.Gauge
+	retentionTime                *prometheus.HistogramVec
+	retentionDaysPerIteration    *prometheus.HistogramVec
+	retentionTenantsPerIteration *prometheus.HistogramVec
 }
 
 func NewMetrics(r prometheus.Registerer, bloomMetrics *v1.Metrics) *Metrics {
@@ -175,6 +181,41 @@ func NewMetrics(r prometheus.Registerer, bloomMetrics *v1.Metrics) *Metrics {
 			Name:      "tenant_compaction_seconds_total",
 			Help:      "Time spent processing a tenant.",
 		}, []string{tenantLabel}),
+
+		// Retention
+		retentionRunning: promauto.With(r).NewGauge(prometheus.GaugeOpts{
+			Namespace: metricsNamespace,
+			Subsystem: metricsSubsystem,
+			Name:      "retention_running",
+			Help:      "1 if retention is running in this compactor.",
+		}),
+
+		retentionTime: promauto.With(r).NewHistogramVec(prometheus.HistogramOpts{
+			Namespace: metricsNamespace,
+			Subsystem: metricsSubsystem,
+			Name:      "retention_time_seconds",
+			Help:      "Time this retention process took to complete.",
+			// 1second -> 5 years, 10 buckets
+			Buckets: prometheus.DefBuckets,
+		}, []string{"status"}),
+
+		retentionDaysPerIteration: promauto.With(r).NewHistogramVec(prometheus.HistogramOpts{
+			Namespace: metricsNamespace,
+			Subsystem: metricsSubsystem,
+			Name:      "retention_days_processed",
+			Help:      "Number of days iterated over during the retention process.",
+			// 1day -> 5 years, 10 buckets
+			Buckets: prometheus.ExponentialBucketsRange(1, 365*5, 10),
+		}, []string{"status"}),
+
+		retentionTenantsPerIteration: promauto.With(r).NewHistogramVec(prometheus.HistogramOpts{
+			Namespace: metricsNamespace,
+			Subsystem: metricsSubsystem,
+			Name:      "retention_tenants_processed",
+			Help:      "Number of tenants on which retention was applied during the retention process.",
+			// 1 tenant -> 10k tenants, 10 buckets
+			Buckets: prometheus.ExponentialBucketsRange(1, 10000, 10),
+		}, []string{"status"}),
 	}
 
 	return &m