Implement partition compaction planner (cortexproject#6469)

* Implement partition compaction grouper Signed-off-by: Alex Le <leqiyue@amazon.com> * fix comment Signed-off-by: Alex Le <leqiyue@amazon.com> * replace level 1 compaction limits with ingestion replication factor Signed-off-by: Alex Le <leqiyue@amazon.com> * fix doc Signed-off-by: Alex Le <leqiyue@amazon.com> * update compaction_visit_marker_timeout default value Signed-off-by: Alex Le <leqiyue@amazon.com> * update default value for compactor_partition_index_size_limit_in_bytes Signed-off-by: Alex Le <leqiyue@amazon.com> * refactor code Signed-off-by: Alex Le <leqiyue@amazon.com> * address comments and refactor Signed-off-by: Alex Le <leqiyue@amazon.com> * address comment Signed-off-by: Alex Le <leqiyue@amazon.com> * address comment Signed-off-by: Alex Le <leqiyue@amazon.com> * update config name Signed-off-by: Alex Le <leqiyue@amazon.com> * Implement partition compaction planner Signed-off-by: Alex Le <leqiyue@amazon.com> * fix after rebase Signed-off-by: Alex Le <leqiyue@amazon.com> * addressed comments Signed-off-by: Alex Le <leqiyue@amazon.com> * updated doc and refactored metric Signed-off-by: Alex Le <leqiyue@amazon.com> * fix test Signed-off-by: Alex Le <leqiyue@amazon.com> --------- Signed-off-by: Alex Le <leqiyue@amazon.com>
alexqyle · Jan 14, 2025 · 68a2993 · 68a2993
1 parent d5e9b5f
commit 68a2993
Show file tree

Hide file tree

Showing 7 changed files with 497 additions and 11 deletions.
diff --git a/docs/blocks-storage/compactor.md b/docs/blocks-storage/compactor.md
@@ -285,6 +285,12 @@ compactor:
     # CLI flag: -compactor.ring.wait-active-instance-timeout
     [wait_active_instance_timeout: <duration> | default = 10m]
 
+  # How long shuffle sharding planner would wait before running planning code.
+  # This delay would prevent double compaction when two compactors claimed same
+  # partition in grouper at same time.
+  # CLI flag: -compactor.sharding-planner-delay
+  [sharding_planner_delay: <duration> | default = 10s]
+
   # The compaction strategy to use. Supported values are: default, partitioning.
   # CLI flag: -compactor.compaction-strategy
   [compaction_strategy: <string> | default = "default"]

diff --git a/docs/configuration/config-file-reference.md b/docs/configuration/config-file-reference.md
@@ -2334,6 +2334,12 @@ sharding_ring:
   # CLI flag: -compactor.ring.wait-active-instance-timeout
   [wait_active_instance_timeout: <duration> | default = 10m]
 
+# How long shuffle sharding planner would wait before running planning code.
+# This delay would prevent double compaction when two compactors claimed same
+# partition in grouper at same time.
+# CLI flag: -compactor.sharding-planner-delay
+[sharding_planner_delay: <duration> | default = 10s]
+
 # The compaction strategy to use. Supported values are: default, partitioning.
 # CLI flag: -compactor.compaction-strategy
 [compaction_strategy: <string> | default = "default"]

diff --git a/pkg/compactor/compactor.go b/pkg/compactor/compactor.go
@@ -153,7 +153,7 @@ var (
 		plannerFactory := func(ctx context.Context, bkt objstore.InstrumentedBucket, logger log.Logger, cfg Config, noCompactionMarkFilter *compact.GatherNoCompactionMarkFilter, ringLifecycle *ring.Lifecycler, userID string, blockVisitMarkerReadFailed prometheus.Counter, blockVisitMarkerWriteFailed prometheus.Counter, compactorMetrics *compactorMetrics) compact.Planner {
 
 			if cfg.CompactionStrategy == util.CompactionStrategyPartitioning {
-				return NewPartitionCompactionPlanner(ctx, bkt, logger)
+				return NewPartitionCompactionPlanner(ctx, bkt, logger, cfg.BlockRanges.ToMilliseconds(), noCompactionMarkFilter.NoCompactMarkedBlocks, ringLifecycle.ID, userID, cfg.ShardingPlannerDelay, cfg.CompactionVisitMarkerTimeout, cfg.CompactionVisitMarkerFileUpdateInterval, compactorMetrics)
 			} else {
 				return NewShuffleShardingPlanner(ctx, bkt, logger, cfg.BlockRanges.ToMilliseconds(), noCompactionMarkFilter.NoCompactMarkedBlocks, ringLifecycle.ID, cfg.CompactionVisitMarkerTimeout, cfg.CompactionVisitMarkerFileUpdateInterval, blockVisitMarkerReadFailed, blockVisitMarkerWriteFailed)
 			}
@@ -234,9 +234,10 @@ type Config struct {
 	DisabledTenants flagext.StringSliceCSV `yaml:"disabled_tenants"`
 
 	// Compactors sharding.
-	ShardingEnabled  bool       `yaml:"sharding_enabled"`
-	ShardingStrategy string     `yaml:"sharding_strategy"`
-	ShardingRing     RingConfig `yaml:"sharding_ring"`
+	ShardingEnabled      bool          `yaml:"sharding_enabled"`
+	ShardingStrategy     string        `yaml:"sharding_strategy"`
+	ShardingRing         RingConfig    `yaml:"sharding_ring"`
+	ShardingPlannerDelay time.Duration `yaml:"sharding_planner_delay"`
 
 	// Compaction strategy.
 	CompactionStrategy string `yaml:"compaction_strategy"`
@@ -304,6 +305,8 @@ func (cfg *Config) RegisterFlags(f *flag.FlagSet) {
 
 	f.BoolVar(&cfg.AcceptMalformedIndex, "compactor.accept-malformed-index", false, "When enabled, index verification will ignore out of order label names.")
 	f.BoolVar(&cfg.CachingBucketEnabled, "compactor.caching-bucket-enabled", false, "When enabled, caching bucket will be used for compactor, except cleaner service, which serves as the source of truth for block status")
+
+	f.DurationVar(&cfg.ShardingPlannerDelay, "compactor.sharding-planner-delay", 10*time.Second, "How long shuffle sharding planner would wait before running planning code. This delay would prevent double compaction when two compactors claimed same partition in grouper at same time.")
 }
 
 func (cfg *Config) Validate(limits validation.Limits) error {

diff --git a/pkg/compactor/compactor_metrics.go b/pkg/compactor/compactor_metrics.go
@@ -39,6 +39,7 @@ type compactorMetrics struct {
 	remainingPlannedCompactions *prometheus.GaugeVec
 	compactionErrorsCount       *prometheus.CounterVec
 	partitionCount              *prometheus.GaugeVec
+	compactionsNotPlanned       *prometheus.CounterVec
 }
 
 const (
@@ -174,6 +175,10 @@ func newCompactorMetricsWithLabels(reg prometheus.Registerer, commonLabels []str
 		Name: "cortex_compactor_group_partition_count",
 		Help: "Number of partitions for each compaction group.",
 	}, compactionLabels)
+	m.compactionsNotPlanned = promauto.With(reg).NewCounterVec(prometheus.CounterOpts{
+		Name: "cortex_compactor_group_compactions_not_planned_total",
+		Help: "Total number of group compaction not planned due to error.",
+	}, compactionLabels)
 
 	return &m
 }
@@ -225,6 +230,7 @@ func (m *compactorMetrics) initMetricWithCompactionLabelValues(labelValue ...str
 	m.compactionFailures.WithLabelValues(labelValue...)
 	m.verticalCompactions.WithLabelValues(labelValue...)
 	m.partitionCount.WithLabelValues(labelValue...)
+	m.compactionsNotPlanned.WithLabelValues(labelValue...)
 }
 
 func (m *compactorMetrics) deleteMetricsForDeletedTenant(userID string) {
@@ -236,4 +242,5 @@ func (m *compactorMetrics) deleteMetricsForDeletedTenant(userID string) {
 	m.compactionFailures.DeleteLabelValues(userID)
 	m.verticalCompactions.DeleteLabelValues(userID)
 	m.partitionCount.DeleteLabelValues(userID)
+	m.compactionsNotPlanned.DeleteLabelValues(userID)
 }
diff --git a/pkg/compactor/compactor_metrics_test.go b/pkg/compactor/compactor_metrics_test.go
@@ -135,6 +135,11 @@ func TestSyncerMetrics(t *testing.T) {
 			cortex_compactor_group_partition_count{user="aaa"} 511060
 			cortex_compactor_group_partition_count{user="bbb"} 522170
 			cortex_compactor_group_partition_count{user="ccc"} 533280
+			# HELP cortex_compactor_group_compactions_not_planned_total Total number of group compaction not planned due to error.
+			# TYPE cortex_compactor_group_compactions_not_planned_total counter
+			cortex_compactor_group_compactions_not_planned_total{user="aaa"} 544390
+			cortex_compactor_group_compactions_not_planned_total{user="bbb"} 555500
+			cortex_compactor_group_compactions_not_planned_total{user="ccc"} 566610
 	`))
 	require.NoError(t, err)
 
@@ -191,4 +196,7 @@ func generateTestData(cm *compactorMetrics, base float64) {
 	cm.partitionCount.WithLabelValues("aaa").Add(46 * base)
 	cm.partitionCount.WithLabelValues("bbb").Add(47 * base)
 	cm.partitionCount.WithLabelValues("ccc").Add(48 * base)
+	cm.compactionsNotPlanned.WithLabelValues("aaa").Add(49 * base)
+	cm.compactionsNotPlanned.WithLabelValues("bbb").Add(50 * base)
+	cm.compactionsNotPlanned.WithLabelValues("ccc").Add(51 * base)
 }
diff --git a/pkg/compactor/partition_compaction_planner.go b/pkg/compactor/partition_compaction_planner.go
@@ -2,30 +2,148 @@ package compactor
 
 import (
 	"context"
+	"fmt"
+	"time"
 
 	"github.com/go-kit/log"
+	"github.com/go-kit/log/level"
+	"github.com/oklog/ulid"
+	"github.com/pkg/errors"
 	"github.com/thanos-io/objstore"
 	"github.com/thanos-io/thanos/pkg/block/metadata"
+
+	"github.com/cortexproject/cortex/pkg/storage/tsdb"
+)
+
+var (
+	plannerCompletedPartitionError = errors.New("got completed partition")
+	plannerVisitedPartitionError   = errors.New("got partition visited by other compactor")
 )
 
 type PartitionCompactionPlanner struct {
-	ctx    context.Context
-	bkt    objstore.InstrumentedBucket
-	logger log.Logger
+	ctx                                    context.Context
+	bkt                                    objstore.InstrumentedBucket
+	logger                                 log.Logger
+	ranges                                 []int64
+	noCompBlocksFunc                       func() map[ulid.ULID]*metadata.NoCompactMark
+	ringLifecyclerID                       string
+	userID                                 string
+	plannerDelay                           time.Duration
+	partitionVisitMarkerTimeout            time.Duration
+	partitionVisitMarkerFileUpdateInterval time.Duration
+	compactorMetrics                       *compactorMetrics
 }
 
 func NewPartitionCompactionPlanner(
 	ctx context.Context,
 	bkt objstore.InstrumentedBucket,
 	logger log.Logger,
+	ranges []int64,
+	noCompBlocksFunc func() map[ulid.ULID]*metadata.NoCompactMark,
+	ringLifecyclerID string,
+	userID string,
+	plannerDelay time.Duration,
+	partitionVisitMarkerTimeout time.Duration,
+	partitionVisitMarkerFileUpdateInterval time.Duration,
+	compactorMetrics *compactorMetrics,
 ) *PartitionCompactionPlanner {
 	return &PartitionCompactionPlanner{
-		ctx:    ctx,
-		bkt:    bkt,
-		logger: logger,
+		ctx:                                    ctx,
+		bkt:                                    bkt,
+		logger:                                 logger,
+		ranges:                                 ranges,
+		noCompBlocksFunc:                       noCompBlocksFunc,
+		ringLifecyclerID:                       ringLifecyclerID,
+		userID:                                 userID,
+		plannerDelay:                           plannerDelay,
+		partitionVisitMarkerTimeout:            partitionVisitMarkerTimeout,
+		partitionVisitMarkerFileUpdateInterval: partitionVisitMarkerFileUpdateInterval,
+		compactorMetrics:                       compactorMetrics,
 	}
 }
 
 func (p *PartitionCompactionPlanner) Plan(ctx context.Context, metasByMinTime []*metadata.Meta, errChan chan error, extensions any) ([]*metadata.Meta, error) {
-	panic("PartitionCompactionPlanner not implemented")
+	cortexMetaExtensions, err := tsdb.ConvertToCortexMetaExtensions(extensions)
+	if err != nil {
+		return nil, err
+	}
+	if cortexMetaExtensions == nil {
+		return nil, fmt.Errorf("cortexMetaExtensions cannot be nil")
+	}
+	return p.PlanWithPartition(ctx, metasByMinTime, cortexMetaExtensions, errChan)
+}
+
+func (p *PartitionCompactionPlanner) PlanWithPartition(_ context.Context, metasByMinTime []*metadata.Meta, cortexMetaExtensions *tsdb.CortexMetaExtensions, errChan chan error) ([]*metadata.Meta, error) {
+	partitionInfo := cortexMetaExtensions.PartitionInfo
+	if partitionInfo == nil {
+		return nil, fmt.Errorf("partitionInfo cannot be nil")
+	}
+	partitionID := partitionInfo.PartitionID
+	partitionedGroupID := partitionInfo.PartitionedGroupID
+
+	// This delay would prevent double compaction when two compactors
+	// claimed same partition in grouper at same time.
+	time.Sleep(p.plannerDelay)
+
+	visitMarker := newPartitionVisitMarker(p.ringLifecyclerID, partitionedGroupID, partitionID)
+	visitMarkerManager := NewVisitMarkerManager(p.bkt, p.logger, p.ringLifecyclerID, visitMarker)
+	existingPartitionVisitMarker := &partitionVisitMarker{}
+	err := visitMarkerManager.ReadVisitMarker(p.ctx, existingPartitionVisitMarker)
+	visitMarkerExists := true
+	if err != nil {
+		if errors.Is(err, errorVisitMarkerNotFound) {
+			visitMarkerExists = false
+		} else {
+			p.compactorMetrics.compactionsNotPlanned.WithLabelValues(p.userID, cortexMetaExtensions.TimeRangeStr()).Inc()
+			return nil, fmt.Errorf("unable to get visit marker file for partition with partition ID %d, partitioned group ID %d: %s", partitionID, partitionedGroupID, err.Error())
+		}
+	}
+	if visitMarkerExists {
+		if existingPartitionVisitMarker.GetStatus() == Completed {
+			level.Warn(p.logger).Log("msg", "partition is in completed status", "partitioned_group_id", partitionedGroupID, "partition_id", partitionID, "compactor_id", p.ringLifecyclerID, existingPartitionVisitMarker.String())
+			return nil, plannerCompletedPartitionError
+		}
+		if !existingPartitionVisitMarker.IsPendingByCompactor(p.partitionVisitMarkerTimeout, partitionID, p.ringLifecyclerID) {
+			level.Warn(p.logger).Log("msg", "partition is not visited by current compactor", "partitioned_group_id", partitionedGroupID, "partition_id", partitionID, "compactor_id", p.ringLifecyclerID, existingPartitionVisitMarker.String())
+			return nil, plannerVisitedPartitionError
+		}
+	}
+
+	// Ensure all blocks fits within the largest range. This is a double check
+	// to ensure there's no bug in the previous blocks grouping, given this Plan()
+	// is just a pass-through.
+	// Modified from https://github.com/cortexproject/cortex/pull/2616/files#diff-e3051fc530c48bb276ba958dd8fadc684e546bd7964e6bc75cef9a86ef8df344R28-R63
+	largestRange := p.ranges[len(p.ranges)-1]
+	rangeStart := getRangeStart(metasByMinTime[0], largestRange)
+	rangeEnd := rangeStart + largestRange
+	noCompactMarked := p.noCompBlocksFunc()
+	resultMetas := make([]*metadata.Meta, 0, len(metasByMinTime))
+
+	for _, b := range metasByMinTime {
+		if b.ULID == DUMMY_BLOCK_ID {
+			continue
+		}
+		blockID := b.ULID.String()
+		if _, excluded := noCompactMarked[b.ULID]; excluded {
+			continue
+		}
+
+		if b.MinTime < rangeStart || b.MaxTime > rangeEnd {
+			p.compactorMetrics.compactionsNotPlanned.WithLabelValues(p.userID, cortexMetaExtensions.TimeRangeStr()).Inc()
+			level.Warn(p.logger).Log("msg", "block is outside the largest expected range", "partitioned_group_id", partitionedGroupID, "partition_id", partitionID, "block_id", blockID, "block_min_time", b.MinTime, "block_max_time", b.MaxTime, "range_start", rangeStart, "range_end", rangeEnd)
+			return nil, fmt.Errorf("block %s with time range %d:%d is outside the largest expected range %d:%d", blockID, b.MinTime, b.MaxTime, rangeStart, rangeEnd)
+		}
+
+		resultMetas = append(resultMetas, b)
+	}
+
+	if len(resultMetas) < 1 {
+		p.compactorMetrics.compactionsNotPlanned.WithLabelValues(p.userID, cortexMetaExtensions.TimeRangeStr()).Inc()
+		level.Warn(p.logger).Log("msg", "result meta size is empty", "partitioned_group_id", partitionedGroupID, "partition_id", partitionID, "group_size", len(metasByMinTime))
+		return nil, nil
+	}
+
+	go visitMarkerManager.HeartBeat(p.ctx, errChan, p.partitionVisitMarkerFileUpdateInterval, false)
+
+	return resultMetas, nil
 }