block

package
v0.32.0-rc.0 Latest Latest
Warning

This package is not in the latest version of its module.

Go to latest
Published: Aug 16, 2023 License: Apache-2.0 Imports: 38 Imported by: 23

Documentation

Overview

Package block contains common functionality for interacting with TSDB blocks in the context of Thanos.

Index

Constants

View Source
const (
	// MetaFilename is the known JSON filename for meta information.
	MetaFilename = "meta.json"
	// IndexFilename is the known index file for block index.
	IndexFilename = "index"
	// IndexHeaderFilename is the canonical name for binary index header file that stores essential information.
	IndexHeaderFilename = "index-header"
	// ChunksDirname is the known dir name for chunks with compressed samples.
	ChunksDirname = "chunks"

	// DebugMetas is a directory for debug meta files that happen in the past. Useful for debugging.
	DebugMetas = "debug/metas"
)
View Source
const (
	CorruptedMeta = "corrupted-meta-json"
	NoMeta        = "no-meta-json"
	LoadedMeta    = "loaded"
	FailedMeta    = "failed"

	// Blocks that are marked for deletion can be loaded as well. This is done to make sure that we load blocks that are meant to be deleted,
	// but don't have a replacement block yet.
	MarkedForDeletionMeta = "marked-for-deletion"

	// MarkedForNoCompactionMeta is label for blocks which are loaded but also marked for no compaction. This label is also counted in `loaded` label metric.
	MarkedForNoCompactionMeta = "marked-for-no-compact"

	// MarkedForNoDownsampleMeta is label for blocks which are loaded but also marked for no downsample. This label is also counted in `loaded` label metric.
	MarkedForNoDownsampleMeta = "marked-for-no-downsample"
)
View Source
const BlockIDLabel = "__block_id"

Special label that will have an ULID of the meta.json being referenced to.

View Source
const FetcherConcurrency = 32

Variables

View Source
var (
	ErrorSyncMetaNotFound  = errors.New("meta.json not found")
	ErrorSyncMetaCorrupted = errors.New("meta.json corrupted")
)
View Source
var (
	SelectorSupportedRelabelActions = map[relabel.Action]struct{}{relabel.Keep: {}, relabel.Drop: {}, relabel.HashMod: {}}
)

Functions

func Delete

func Delete(ctx context.Context, logger log.Logger, bkt objstore.Bucket, id ulid.ULID) error

Delete removes directory that is meant to be block directory. NOTE: Always prefer this method for deleting blocks.

  • We have to delete block's files in the certain order (meta.json first and deletion-mark.json last) to ensure we don't end up with malformed partial blocks. Thanos system handles well partial blocks only if they don't have meta.json. If meta.json is present Thanos assumes valid block.
  • This avoids deleting empty dir (whole bucket) by mistake.

func Download

func Download(ctx context.Context, logger log.Logger, bucket objstore.Bucket, id ulid.ULID, dst string, options ...objstore.DownloadOption) error

Download downloads directory that is mean to be block directory. If any of the files have a hash calculated in the meta file and it matches with what is in the destination path then we do not download it. We always re-download the meta file.

func DownloadMeta

func DownloadMeta(ctx context.Context, logger log.Logger, bkt objstore.Bucket, id ulid.ULID) (metadata.Meta, error)

DownloadMeta downloads only meta file from bucket by block ID. TODO(bwplotka): Differentiate between network error & partial upload.

func GatherFileStats added in v0.27.0

func GatherFileStats(blockDir string, hf metadata.HashFunc, logger log.Logger) (res []metadata.File, _ error)

GatherFileStats returns metadata.File entry for files inside TSDB block (index, chunks, meta.json).

func GetSegmentFiles added in v0.17.0

func GetSegmentFiles(blockDir string) []string

GetSegmentFiles returns list of segment files for given block. Paths are relative to the chunks directory. In case of errors, nil is returned.

func IgnoreCompleteOutsideChunk

func IgnoreCompleteOutsideChunk(mint, maxt int64, _, curr *chunks.Meta) (bool, error)

func IgnoreDuplicateOutsideChunk

func IgnoreDuplicateOutsideChunk(_, _ int64, last, curr *chunks.Meta) (bool, error)

func IgnoreIssue347OutsideChunk

func IgnoreIssue347OutsideChunk(_, maxt int64, _, curr *chunks.Meta) (bool, error)

func IsBlockDir

func IsBlockDir(path string) (id ulid.ULID, ok bool)

func IsBlockMetaFile added in v0.32.0

func IsBlockMetaFile(path string) bool

func MarkForDeletion added in v0.12.0

func MarkForDeletion(ctx context.Context, logger log.Logger, bkt objstore.Bucket, id ulid.ULID, details string, markedForDeletion prometheus.Counter) error

MarkForDeletion creates a file which stores information about when the block was marked for deletion.

func MarkForNoCompact added in v0.17.0

func MarkForNoCompact(ctx context.Context, logger log.Logger, bkt objstore.Bucket, id ulid.ULID, reason metadata.NoCompactReason, details string, markedForNoCompact prometheus.Counter) error

MarkForNoCompact creates a file which marks block to be not compacted.

func MarkForNoDownsample added in v0.30.0

func MarkForNoDownsample(ctx context.Context, logger log.Logger, bkt objstore.Bucket, id ulid.ULID, reason metadata.NoDownsampleReason, details string, markedForNoDownsample prometheus.Counter) error

MarkForNoDownsample creates a file which marks block to be not downsampled.

func ParseRelabelConfig added in v0.15.0

func ParseRelabelConfig(contentYaml []byte, supportedActions map[relabel.Action]struct{}) ([]*relabel.Config, error)

ParseRelabelConfig parses relabel configuration. If supportedActions not specified, all relabel actions are valid.

func RemoveMark added in v0.30.0

func RemoveMark(ctx context.Context, logger log.Logger, bkt objstore.Bucket, id ulid.ULID, removeMark prometheus.Counter, markedFilename string) error

RemoveMark removes the file which marked the block for deletion, no-downsample or no-compact.

func Repair

func Repair(logger log.Logger, dir string, id ulid.ULID, source metadata.SourceType, ignoreChkFns ...ignoreFnType) (resid ulid.ULID, err error)

Repair open the block with given id in dir and creates a new one with fixed data. It: - removes out of order duplicates - all "complete" outsiders (they will not accessed anyway) - removes all near "complete" outside chunks introduced by https://github.com/prometheus/tsdb/issues/347. Fixable inconsistencies are resolved in the new block. TODO(bplotka): https://github.com/thanos-io/thanos/issues/378.

func Upload

func Upload(ctx context.Context, logger log.Logger, bkt objstore.Bucket, bdir string, hf metadata.HashFunc, options ...objstore.UploadOption) error

Upload uploads a TSDB block to the object storage. It verifies basic features of Thanos block.

func UploadPromBlock added in v0.20.0

func UploadPromBlock(ctx context.Context, logger log.Logger, bkt objstore.Bucket, bdir string, hf metadata.HashFunc, options ...objstore.UploadOption) error

UploadPromBlock uploads a TSDB block to the object storage. It assumes the block is used in Prometheus so it doesn't check Thanos external labels.

func VerifyIndex

func VerifyIndex(logger log.Logger, fn string, minTime, maxTime int64) error

VerifyIndex does a full run over a block index and verifies that it fulfills the order invariants.

Types

type BaseFetcher added in v0.12.0

type BaseFetcher struct {
	// contains filtered or unexported fields
}

BaseFetcher is a struct that synchronizes filtered metadata of all block in the object storage with the local state. Go-routine safe.

func NewBaseFetcher added in v0.12.0

func NewBaseFetcher(logger log.Logger, concurrency int, bkt objstore.InstrumentedBucketReader, dir string, reg prometheus.Registerer) (*BaseFetcher, error)

NewBaseFetcher constructs BaseFetcher.

func (*BaseFetcher) NewMetaFetcher added in v0.12.0

func (f *BaseFetcher) NewMetaFetcher(reg prometheus.Registerer, filters []MetadataFilter, logTags ...interface{}) *MetaFetcher

NewMetaFetcher transforms BaseFetcher into actually usable *MetaFetcher.

type ConsistencyDelayMetaFilter added in v0.11.0

type ConsistencyDelayMetaFilter struct {
	// contains filtered or unexported fields
}

ConsistencyDelayMetaFilter is a BaseFetcher filter that filters out blocks that are created before a specified consistency delay. Not go-routine safe.

func NewConsistencyDelayMetaFilter added in v0.11.0

func NewConsistencyDelayMetaFilter(logger log.Logger, consistencyDelay time.Duration, reg prometheus.Registerer) *ConsistencyDelayMetaFilter

NewConsistencyDelayMetaFilter creates ConsistencyDelayMetaFilter.

func (*ConsistencyDelayMetaFilter) Filter added in v0.11.0

func (f *ConsistencyDelayMetaFilter) Filter(_ context.Context, metas map[ulid.ULID]*metadata.Meta, synced GaugeVec, modified GaugeVec) error

Filter filters out blocks that filters blocks that have are created before a specified consistency delay.

type DeduplicateFilter added in v0.11.0

type DeduplicateFilter interface {
	DuplicateIDs() []ulid.ULID
}

type DefaultDeduplicateFilter added in v0.32.0

type DefaultDeduplicateFilter struct {
	// contains filtered or unexported fields
}

DefaultDeduplicateFilter is a BaseFetcher filter that filters out older blocks that have exactly the same data. Not go-routine safe.

func NewDeduplicateFilter added in v0.11.0

func NewDeduplicateFilter(concurrency int) *DefaultDeduplicateFilter

NewDeduplicateFilter creates DefaultDeduplicateFilter.

func (*DefaultDeduplicateFilter) DuplicateIDs added in v0.32.0

func (f *DefaultDeduplicateFilter) DuplicateIDs() []ulid.ULID

DuplicateIDs returns slice of block ids that are filtered out by DefaultDeduplicateFilter.

func (*DefaultDeduplicateFilter) Filter added in v0.32.0

func (f *DefaultDeduplicateFilter) Filter(_ context.Context, metas map[ulid.ULID]*metadata.Meta, synced GaugeVec, modified GaugeVec) error

Filter filters out duplicate blocks that can be formed from two or more overlapping blocks that fully submatches the source blocks of the older blocks.

type DiskWriter added in v0.18.0

type DiskWriter struct {
	// contains filtered or unexported fields
}

func NewDiskWriter added in v0.18.0

func NewDiskWriter(ctx context.Context, logger log.Logger, bDir string) (_ *DiskWriter, err error)

NewDiskWriter allows to write single TSDB block to disk and returns statistics. Destination block directory has to exists.

func (*DiskWriter) AddSeries added in v0.18.0

func (s *DiskWriter) AddSeries(ref storage.SeriesRef, l labels.Labels, chks ...chunks.Meta) error

func (*DiskWriter) AddSymbol added in v0.18.0

func (s *DiskWriter) AddSymbol(sym string) error

func (DiskWriter) Close added in v0.18.0

func (s DiskWriter) Close() error

func (*DiskWriter) Flush added in v0.18.0

func (d *DiskWriter) Flush() (_ tsdb.BlockStats, err error)

func (*DiskWriter) WriteChunks added in v0.18.0

func (s *DiskWriter) WriteChunks(chks ...chunks.Meta) error

type FetcherMetrics added in v0.19.0

type FetcherMetrics struct {
	Syncs        prometheus.Counter
	SyncFailures prometheus.Counter
	SyncDuration prometheus.Histogram

	Synced   *extprom.TxGaugeVec
	Modified *extprom.TxGaugeVec
}

FetcherMetrics holds metrics tracked by the metadata fetcher. This struct and its fields are exported to allow depending projects (eg. Cortex) to implement their own custom metadata fetcher while tracking compatible metrics.

func NewFetcherMetrics added in v0.19.0

func NewFetcherMetrics(reg prometheus.Registerer, syncedExtraLabels, modifiedExtraLabels [][]string) *FetcherMetrics

func (*FetcherMetrics) ResetTx added in v0.19.0

func (s *FetcherMetrics) ResetTx()

ResetTx starts new transaction for metrics tracked by transaction GaugeVec.

func (*FetcherMetrics) Submit added in v0.19.0

func (s *FetcherMetrics) Submit()

Submit applies new values for metrics tracked by transaction GaugeVec.

type GaugeVec added in v0.29.0

type GaugeVec interface {
	WithLabelValues(lvs ...string) prometheus.Gauge
}

GaugeVec hides something like a Prometheus GaugeVec or an extprom.TxGaugeVec.

type HealthStats added in v0.18.0

type HealthStats struct {
	// TotalSeries represents total number of series in block.
	TotalSeries int64
	// OutOfOrderSeries represents number of series that have out of order chunks.
	OutOfOrderSeries int

	// OutOfOrderChunks represents number of chunks that are out of order (older time range is after younger one).
	OutOfOrderChunks int
	// DuplicatedChunks represents number of chunks with same time ranges within same series, potential duplicates.
	DuplicatedChunks int
	// OutsideChunks represents number of all chunks that are before or after time range specified in block meta.
	OutsideChunks int
	// CompleteOutsideChunks is subset of OutsideChunks that will be never accessed. They are completely out of time range specified in block meta.
	CompleteOutsideChunks int
	// Issue347OutsideChunks represents subset of OutsideChunks that are outsiders caused by https://github.com/prometheus/tsdb/issues/347
	// and is something that Thanos handle.
	//
	// Specifically we mean here chunks with minTime == block.maxTime and maxTime > block.MaxTime. These are
	// are segregated into separate counters. These chunks are safe to be deleted, since they are duplicated across 2 blocks.
	Issue347OutsideChunks int
	// OutOfOrderLabels represents the number of postings that contained out
	// of order labels, a bug present in Prometheus 2.8.0 and below.
	OutOfOrderLabels int

	// Debug Statistics.
	SeriesMinLifeDuration time.Duration
	SeriesAvgLifeDuration time.Duration
	SeriesMaxLifeDuration time.Duration

	SeriesMinLifeDurationWithoutSingleSampleSeries time.Duration
	SeriesAvgLifeDurationWithoutSingleSampleSeries time.Duration
	SeriesMaxLifeDurationWithoutSingleSampleSeries time.Duration

	SeriesMinChunks int64
	SeriesAvgChunks int64
	SeriesMaxChunks int64

	TotalChunks int64

	ChunkMinDuration time.Duration
	ChunkAvgDuration time.Duration
	ChunkMaxDuration time.Duration

	ChunkMinSize int64
	ChunkAvgSize int64
	ChunkMaxSize int64

	SeriesMinSize int64
	SeriesAvgSize int64
	SeriesMaxSize int64

	SingleSampleSeries int64
	SingleSampleChunks int64

	LabelNamesCount        int64
	MetricLabelValuesCount int64
}

func GatherIndexHealthStats added in v0.18.0

func GatherIndexHealthStats(logger log.Logger, fn string, minTime, maxTime int64) (stats HealthStats, err error)

GatherIndexHealthStats returns useful counters as well as outsider chunks (chunks outside of block time range) that helps to assess index health. It considers https://github.com/prometheus/tsdb/issues/347 as something that Thanos can handle. See HealthStats.Issue347OutsideChunks for details.

func (HealthStats) AnyErr added in v0.18.0

func (i HealthStats) AnyErr() error

AnyErr returns error if stats indicates any block issue.

func (HealthStats) CriticalErr added in v0.18.0

func (i HealthStats) CriticalErr() error

CriticalErr returns error if stats indicates critical block issue, that might solved only by manual repair procedure.

func (HealthStats) Issue347OutsideChunksErr added in v0.18.0

func (i HealthStats) Issue347OutsideChunksErr() error

Issue347OutsideChunksErr returns error if stats indicates issue347 block issue, that is repaired explicitly before compaction (on plan block).

func (HealthStats) OutOfOrderChunksErr added in v0.23.0

func (i HealthStats) OutOfOrderChunksErr() error

func (HealthStats) OutOfOrderLabelsErr added in v0.28.0

func (i HealthStats) OutOfOrderLabelsErr() error

OutOfOrderLabelsErr returns an error if the HealthStats object indicates postings with out of order labels. This is corrected by Prometheus Issue #5372 and affects Prometheus versions 2.8.0 and below.

type IgnoreDeletionMarkFilter added in v0.12.0

type IgnoreDeletionMarkFilter struct {
	// contains filtered or unexported fields
}

IgnoreDeletionMarkFilter is a filter that filters out the blocks that are marked for deletion after a given delay. The delay duration is to make sure that the replacement block can be fetched before we filter out the old block. Delay is not considered when computing DeletionMarkBlocks map. Not go-routine safe.

func NewIgnoreDeletionMarkFilter added in v0.12.0

func NewIgnoreDeletionMarkFilter(logger log.Logger, bkt objstore.InstrumentedBucketReader, delay time.Duration, concurrency int) *IgnoreDeletionMarkFilter

NewIgnoreDeletionMarkFilter creates IgnoreDeletionMarkFilter.

func (*IgnoreDeletionMarkFilter) DeletionMarkBlocks added in v0.12.0

func (f *IgnoreDeletionMarkFilter) DeletionMarkBlocks() map[ulid.ULID]*metadata.DeletionMark

DeletionMarkBlocks returns block ids that were marked for deletion.

func (*IgnoreDeletionMarkFilter) Filter added in v0.12.0

func (f *IgnoreDeletionMarkFilter) Filter(ctx context.Context, metas map[ulid.ULID]*metadata.Meta, synced GaugeVec, modified GaugeVec) error

Filter filters out blocks that are marked for deletion after a given delay. It also returns the blocks that can be deleted since they were uploaded delay duration before current time.

type LabelShardedMetaFilter added in v0.10.0

type LabelShardedMetaFilter struct {
	// contains filtered or unexported fields
}

LabelShardedMetaFilter represents struct that allows sharding. Not go-routine safe.

func NewLabelShardedMetaFilter added in v0.10.0

func NewLabelShardedMetaFilter(relabelConfig []*relabel.Config) *LabelShardedMetaFilter

NewLabelShardedMetaFilter creates LabelShardedMetaFilter.

func (*LabelShardedMetaFilter) Filter added in v0.10.0

func (f *LabelShardedMetaFilter) Filter(_ context.Context, metas map[ulid.ULID]*metadata.Meta, synced GaugeVec, modified GaugeVec) error

Filter filters out blocks that have no labels after relabelling of each block external (Thanos) labels.

type MetaFetcher added in v0.10.0

type MetaFetcher struct {
	// contains filtered or unexported fields
}

func NewMetaFetcher added in v0.10.0

func NewMetaFetcher(logger log.Logger, concurrency int, bkt objstore.InstrumentedBucketReader, dir string, reg prometheus.Registerer, filters []MetadataFilter) (*MetaFetcher, error)

NewMetaFetcher returns meta fetcher.

func NewRawMetaFetcher added in v0.19.0

func NewRawMetaFetcher(logger log.Logger, bkt objstore.InstrumentedBucketReader) (*MetaFetcher, error)

NewRawMetaFetcher returns basic meta fetcher without proper handling for eventual consistent backends or partial uploads. NOTE: Not suitable to use in production.

func (*MetaFetcher) Fetch added in v0.10.0

func (f *MetaFetcher) Fetch(ctx context.Context) (metas map[ulid.ULID]*metadata.Meta, partial map[ulid.ULID]error, err error)

Fetch returns all block metas as well as partial blocks (blocks without or with corrupted meta file) from the bucket. It's caller responsibility to not change the returned metadata files. Maps can be modified.

Returned error indicates a failure in fetching metadata. Returned meta can be assumed as correct, with some blocks missing.

func (*MetaFetcher) UpdateOnChange added in v0.12.0

func (f *MetaFetcher) UpdateOnChange(listener func([]metadata.Meta, error))

UpdateOnChange allows to add listener that will be update on every change.

type MetadataFetcher added in v0.10.0

type MetadataFetcher interface {
	Fetch(ctx context.Context) (metas map[ulid.ULID]*metadata.Meta, partial map[ulid.ULID]error, err error)
	UpdateOnChange(func([]metadata.Meta, error))
}

type MetadataFilter added in v0.12.0

type MetadataFilter interface {
	Filter(ctx context.Context, metas map[ulid.ULID]*metadata.Meta, synced GaugeVec, modified GaugeVec) error
}

Filter allows filtering or modifying metas from the provided map or returns error.

type Reader added in v0.18.0

type Reader interface {
	// Index returns an IndexReader over the block's data.
	Index() (tsdb.IndexReader, error)

	// Chunks returns a ChunkReader over the block's data.
	Chunks() (tsdb.ChunkReader, error)

	// Meta returns block metadata file.
	Meta() tsdb.BlockMeta
}

Reader is like tsdb.BlockReader but without tombstones and size methods.

type ReplicaLabelRemover added in v0.12.0

type ReplicaLabelRemover struct {
	// contains filtered or unexported fields
}

ReplicaLabelRemover is a BaseFetcher filter that modifies external labels of existing blocks, it removes given replica labels from the metadata of blocks that have it.

func NewReplicaLabelRemover added in v0.12.0

func NewReplicaLabelRemover(logger log.Logger, replicaLabels []string) *ReplicaLabelRemover

NewReplicaLabelRemover creates a ReplicaLabelRemover.

func (*ReplicaLabelRemover) Filter added in v0.25.0

func (r *ReplicaLabelRemover) Filter(_ context.Context, metas map[ulid.ULID]*metadata.Meta, synced GaugeVec, modified GaugeVec) error

Filter modifies external labels of existing blocks, it removes given replica labels from the metadata of blocks that have it.

type SeriesWriter added in v0.18.0

type SeriesWriter interface {
	tsdb.IndexWriter
	tsdb.ChunkWriter
}

SeriesWriter is interface for writing series into one or multiple Blocks. Statistics has to be counted by implementation.

type TimePartitionMetaFilter added in v0.10.0

type TimePartitionMetaFilter struct {
	// contains filtered or unexported fields
}

TimePartitionMetaFilter is a BaseFetcher filter that filters out blocks that are outside of specified time range. Not go-routine safe.

func NewTimePartitionMetaFilter added in v0.10.0

func NewTimePartitionMetaFilter(MinTime, MaxTime model.TimeOrDurationValue) *TimePartitionMetaFilter

NewTimePartitionMetaFilter creates TimePartitionMetaFilter.

func (*TimePartitionMetaFilter) Filter added in v0.10.0

func (f *TimePartitionMetaFilter) Filter(_ context.Context, metas map[ulid.ULID]*metadata.Meta, synced GaugeVec, modified GaugeVec) error

Filter filters out blocks that are outside of specified time range.

type Writer added in v0.18.0

type Writer interface {
	SeriesWriter

	Flush() (tsdb.BlockStats, error)
}

Writer is interface for creating block(s).

Directories

Path Synopsis

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL