bootstrapper

README

bootstrapper

The collection of bootstrappers comprise the task executed when bootstrapping a node.

Bootstrappers

• fs: The filesystem bootstrapper, used to bootstrap as much data as possible from the local filesystem.
• peers: The peers bootstrapper, used to bootstrap any remaining data from peers. This is used for a full node join too.
• commitlog: The commit log bootstrapper, currently only used in the case that peers bootstrapping fails. Once the current block is being snapshotted frequently to disk it might be faster and make more sense to not actively use the peers bootstrapper and just use a combination of the filesystem bootstrapper and the minimal time range required from the commit log bootstrapper.

Cache policies

The tasks carried out by each bootstrapper vary a lot on the series cache policy being used.

CacheAll series cache policy

For the cache all policy the filesystem bootstrapper will load all series and all the data for each block and return the entire set of data. This will keep every series and series block on heap.

The peers bootstrapper similarly bootstraps all the data from peers that the filesystem does not have and returns the entire set of data fetched.

RecentlyRead series cache policy

For the recently read policy the filesystem bootstrapper will simply fulfill the time ranges requested matching without actually loading the series and blocks from the files it discovers. This relies on data been fetched lazily from the filesystem when data is required for a series that does not live on heap.

The peers bootstrapper will bootstrap all time ranges requested, and if performing a bootstrap with persistence enabled for a time range, will write the data to disk and then remove the results from memory. A bootstrap with persistence enabled is used for any data that is immutable at the time that bootstrapping commences. For time ranges that are mutable the peer bootstrapper will still write the data out to disk in a durable manner, but in the form of a snapshot, and the series and blocks will still be returned directly as a result from the bootstrapper. This enables the commit log bootstrapper to recover the data in case the node shuts down before the in-memory data can be flushed.

Topology Changes

When nodes are added to a replication group, shards are given away to the joining node. Those shards are closed and we re-bootstrap with the shards that we own. When nodes are removed from a replication group, shards from the removed node are given to remaining nodes in a replication group. The remaining nodes in the replication group will bootstrap the "new" shards that were assigned to it. Note that we also take writes for shards that we own while bootstrapping. However, we do not allow warm/cold flushes to happen while bootstrapping.

For example, see the following sequences: (Node add)

• Node 1:
  • Initial bootstrap (256 shards)
  • Node add
  • Bootstrap (128 shards) // These are the remaining shards it owns.
• Node 2:
  • Node add
  • Inital bootstrap (128 shards) // These are received from Node 1

(Node remove)

• Node 1:
  • Node remove
  • Bootstrap (128 shards) // These are received form Node 2, it owns 256 now.
• Node 2:
  • Node remove

Documentation

Index

• Constants
• func BuildBootstrapIndexSegment(ns namespace.Metadata, requestedRanges result.ShardTimeRanges, ...) (result.IndexBlock, error)
• func EnqueueReaders(ns namespace.Metadata, runOpts bootstrap.RunOptions, ...)
• func GetDefaultIndexBlockForBlockStart(results result.IndexResults, blockStart time.Time) (result.IndexBlock, bool)
• func NewBaseBootstrapper(name string, src bootstrap.Source, opts result.Options, ...) (bootstrap.Bootstrapper, error)
• func NewNoOpAllBootstrapperProvider() bootstrap.BootstrapperProvider
• func NewNoOpNoneBootstrapperProvider() bootstrap.BootstrapperProvider
• func PersistBootstrapIndexSegment(ns namespace.Metadata, requestedRanges result.ShardTimeRanges, ...) (result.IndexBlock, error)
• type NewReaderPoolOptions
• type ReaderPool
  • func NewReaderPool(opts NewReaderPoolOptions) *ReaderPool
  • func (p *ReaderPool) Get() (fs.DataFileSetReader, error)
  • func (p *ReaderPool) Put(r fs.DataFileSetReader)
• type ReaderPoolAllocFn
• type Segment
  • func NewSegment(segment segment.Segment, persisted bool) *Segment
  • func (s *Segment) IsPersisted() bool
• type ShardID
• type ShardReaders
• type ShardTimeRangesTimeWindowGroup
  • func NewShardTimeRangesTimeWindowGroups(shardTimeRanges result.ShardTimeRanges, windowSize time.Duration) []ShardTimeRangesTimeWindowGroup
• type SharedCompactor
• type SharedPersistManager
• type TimeWindowReaders

Constants

const (
	// NoOpNoneBootstrapperName is the name of the noOpNoneBootstrapper
	NoOpNoneBootstrapperName = "noop-none"

	// NoOpAllBootstrapperName is the name of the noOpAllBootstrapperProvider
	NoOpAllBootstrapperName = "noop-all"
)

Functions

func BuildBootstrapIndexSegment

func BuildBootstrapIndexSegment(
	ns namespace.Metadata,
	requestedRanges result.ShardTimeRanges,
	builder segment.DocumentsBuilder,
	compactor *SharedCompactor,
	resultOpts result.Options,
	mmapReporter mmap.Reporter,
	blockStart time.Time,
	blockEnd time.Time,
) (result.IndexBlock, error)

BuildBootstrapIndexSegment is a helper function that builds (in memory) bootstrapped index segments for a ns -> block of time.

func EnqueueReaders

func EnqueueReaders(
	ns namespace.Metadata,
	runOpts bootstrap.RunOptions,
	runtimeOpts runtime.Options,
	fsOpts fs.Options,
	shardsTimeRanges result.ShardTimeRanges,
	readerPool *ReaderPool,
	readersCh chan<- TimeWindowReaders,
	blockSize time.Duration,
	logger *zap.Logger,
)

EnqueueReaders into a readers channel grouped by data block.

func GetDefaultIndexBlockForBlockStart

func GetDefaultIndexBlockForBlockStart(
	results result.IndexResults,
	blockStart time.Time,
) (result.IndexBlock, bool)

GetDefaultIndexBlockForBlockStart gets the index block for the default volume type from the index results.

func NewBaseBootstrapper

func NewBaseBootstrapper(
	name string,
	src bootstrap.Source,
	opts result.Options,
	next bootstrap.Bootstrapper,
) (bootstrap.Bootstrapper, error)

NewBaseBootstrapper creates a new base bootstrapper.

func NewNoOpAllBootstrapperProvider

func NewNoOpAllBootstrapperProvider() bootstrap.BootstrapperProvider

NewNoOpAllBootstrapperProvider creates a new noOpAllBootstrapperProvider.

func NewNoOpNoneBootstrapperProvider

func NewNoOpNoneBootstrapperProvider() bootstrap.BootstrapperProvider

NewNoOpNoneBootstrapperProvider creates a new noOpNoneBootstrapper.

func PersistBootstrapIndexSegment

func PersistBootstrapIndexSegment(
	ns namespace.Metadata,
	requestedRanges result.ShardTimeRanges,
	builder segment.DocumentsBuilder,
	persistManager *SharedPersistManager,
	resultOpts result.Options,
	fulfilled result.ShardTimeRanges,
	blockStart time.Time,
	blockEnd time.Time,
) (result.IndexBlock, error)

PersistBootstrapIndexSegment is a helper function that persists bootstrapped index segments for a ns -> block of time.

Types

type NewReaderPoolOptions

type NewReaderPoolOptions struct {
	Alloc        ReaderPoolAllocFn
	DisableReuse bool
}

NewReaderPoolOptions contains reader pool options.

type ReaderPool

type ReaderPool struct {
	sync.Mutex
	// contains filtered or unexported fields
}

ReaderPool is a lean pool that does not allocate instances up front and is used per bootstrap call.

func NewReaderPool

func NewReaderPool(
	opts NewReaderPoolOptions,
) *ReaderPool

NewReaderPool creates a new share-able fileset reader pool

func (*ReaderPool) Get

func (p *ReaderPool) Get() (fs.DataFileSetReader, error)

Get gets a fileset reader from the pool in synchronized fashion.

func (*ReaderPool) Put

func (p *ReaderPool) Put(r fs.DataFileSetReader)

Put returns a fileset reader back the the pool in synchronized fashion.

type ReaderPoolAllocFn

type ReaderPoolAllocFn func() (fs.DataFileSetReader, error)

ReaderPoolAllocFn allocates a new fileset reader.

type Segment

type Segment struct {
	segment.Segment
	// contains filtered or unexported fields
}

Segment wraps an index segment so we can easily determine whether or not the segment is persisted to disk.

func NewSegment

func NewSegment(segment segment.Segment, persisted bool) *Segment

NewSegment returns an index segment w/ persistence metadata.

func (*Segment) IsPersisted

func (s *Segment) IsPersisted() bool

IsPersisted returns whether or not the underlying segment was persisted to disk.

type ShardID

type ShardID uint32

ShardID is the shard #.

type ShardReaders

type ShardReaders struct {
	Readers []fs.DataFileSetReader
}

ShardReaders are the fileset readers for a shard.

type ShardTimeRangesTimeWindowGroup

type ShardTimeRangesTimeWindowGroup struct {
	Ranges result.ShardTimeRanges
	// contains filtered or unexported fields
}

ShardTimeRangesTimeWindowGroup represents all time ranges for every shard w/in a block of time.

func NewShardTimeRangesTimeWindowGroups

func NewShardTimeRangesTimeWindowGroups(
	shardTimeRanges result.ShardTimeRanges,
	windowSize time.Duration,
) []ShardTimeRangesTimeWindowGroup

NewShardTimeRangesTimeWindowGroups divides shard time ranges into grouped blocks.

type SharedCompactor

type SharedCompactor struct {
	sync.Mutex
	Compactor *compaction.Compactor
}

SharedCompactor is a lockable compactor that's safe to be shared across threads.

type SharedPersistManager

type SharedPersistManager struct {
	sync.Mutex
	Mgr persist.Manager
}

SharedPersistManager is a lockable persist manager that's safe to be shared across threads.

type TimeWindowReaders

type TimeWindowReaders struct {
	Ranges  result.ShardTimeRanges
	Readers map[ShardID]ShardReaders
}

TimeWindowReaders are grouped by data block.