Line data    Source code

       1             : // Copyright 2019 The LevelDB-Go and Pebble Authors. All rights reserved. Use
       2             : // of this source code is governed by a BSD-style license that can be found in
       3             : // the LICENSE file.
       4             : 
       5             : package sstable
       6             : 
       7             : import (
       8             :         "github.com/cockroachdb/fifo"
       9             :         "github.com/cockroachdb/pebble/internal/base"
      10             :         "github.com/cockroachdb/pebble/internal/sstableinternal"
      11             :         "github.com/cockroachdb/pebble/sstable/block"
      12             :         "github.com/cockroachdb/pebble/sstable/rowblk"
      13             : )
      14             : 
      15             : const (
      16             :         // MaximumBlockSize is the maximum permissible size of a block.
      17             :         MaximumBlockSize = rowblk.MaximumSize
      18             :         // DefaultNumDeletionsThreshold defines the minimum number of point
      19             :         // tombstones that must be present in a data block for it to be
      20             :         // considered tombstone-dense.
      21             :         DefaultNumDeletionsThreshold = 100
      22             :         // DefaultDeletionSizeRatioThreshold defines the minimum ratio of the size
      23             :         // of point tombstones to the size of the data block in order to consider the
      24             :         // block as tombstone-dense.
      25             :         DefaultDeletionSizeRatioThreshold = 0.5
      26             : )
      27             : 
      28             : var ignoredInternalProperties = map[string]struct{}{
      29             :         "rocksdb.column.family.id":             {},
      30             :         "rocksdb.fixed.key.length":             {},
      31             :         "rocksdb.index.key.is.user.key":        {},
      32             :         "rocksdb.index.value.is.delta.encoded": {},
      33             :         "rocksdb.oldest.key.time":              {},
      34             :         "rocksdb.creation.time":                {},
      35             :         "rocksdb.file.creation.time":           {},
      36             :         "rocksdb.format.version":               {},
      37             : }
      38             : 
      39             : // FilterType exports the base.FilterType type.
      40             : type FilterType = base.FilterType
      41             : 
      42             : // Exported TableFilter constants.
      43             : const (
      44             :         TableFilter = base.TableFilter
      45             : )
      46             : 
      47             : // FilterWriter exports the base.FilterWriter type.
      48             : type FilterWriter = base.FilterWriter
      49             : 
      50             : // FilterPolicy exports the base.FilterPolicy type.
      51             : type FilterPolicy = base.FilterPolicy
      52             : 
      53             : // Comparers is a map from comparer name to comparer. It is used for debugging
      54             : // tools which may be used on multiple databases configured with different
      55             : // comparers.
      56             : type Comparers map[string]*base.Comparer
      57             : 
      58             : // Mergers is a map from merger name to merger. It is used for debugging tools
      59             : // which may be used on multiple databases configured with different
      60             : // mergers.
      61             : type Mergers map[string]*base.Merger
      62             : 
      63             : // ReaderOptions holds the parameters needed for reading an sstable.
      64             : type ReaderOptions struct {
      65             :         // LoadBlockSema, if set, is used to limit the number of blocks that can be
      66             :         // loaded (i.e. read from the filesystem) in parallel. Each load acquires one
      67             :         // unit from the semaphore for the duration of the read.
      68             :         LoadBlockSema *fifo.Semaphore
      69             : 
      70             :         // User properties specified in this map will not be added to sst.Properties.UserProperties.
      71             :         DeniedUserProperties map[string]struct{}
      72             : 
      73             :         // Comparer defines a total ordering over the space of []byte keys: a 'less
      74             :         // than' relationship. The same comparison algorithm must be used for reads
      75             :         // and writes over the lifetime of the DB.
      76             :         //
      77             :         // The default value uses the same ordering as bytes.Compare.
      78             :         Comparer *Comparer
      79             : 
      80             :         // Merger defines the Merge function in use for this keyspace.
      81             :         Merger *Merger
      82             : 
      83             :         Comparers Comparers
      84             :         Mergers   Mergers
      85             : 
      86             :         // Filters is a map from filter policy name to filter policy. Filters with
      87             :         // policies that are not in this map will be ignored.
      88             :         Filters map[string]FilterPolicy
      89             : 
      90             :         // Logger is an optional logger and tracer.
      91             :         LoggerAndTracer base.LoggerAndTracer
      92             : 
      93             :         // FilterMetricsTracker is optionally used to track filter metrics.
      94             :         FilterMetricsTracker *FilterMetricsTracker
      95             : 
      96             :         // internal options can only be used from within the pebble package.
      97             :         internal sstableinternal.ReaderOptions
      98             : }
      99             : 
     100             : // SetInternal sets the internal reader options. Note that even though this
     101             : // method is public, a caller outside the pebble package can't construct a value
     102             : // to pass to it.
     103           0 : func (o *ReaderOptions) SetInternal(internalOpts sstableinternal.ReaderOptions) {
     104           0 :         o.internal = internalOpts
     105           0 : }
     106             : 
     107             : // SetInternalCacheOpts sets the internal cache options. Note that even though
     108             : // this method is public, a caller outside the pebble package can't construct a
     109             : // value to pass to it.
     110           1 : func (o *ReaderOptions) SetInternalCacheOpts(cacheOpts sstableinternal.CacheOptions) {
     111           1 :         o.internal.CacheOpts = cacheOpts
     112           1 : }
     113             : 
     114           1 : func (o ReaderOptions) ensureDefaults() ReaderOptions {
     115           1 :         if o.Comparer == nil {
     116           0 :                 o.Comparer = base.DefaultComparer
     117           0 :         }
     118           1 :         if o.Merger == nil {
     119           1 :                 o.Merger = base.DefaultMerger
     120           1 :         }
     121           1 :         if o.LoggerAndTracer == nil {
     122           1 :                 o.LoggerAndTracer = base.NoopLoggerAndTracer{}
     123           1 :         }
     124           1 :         if o.DeniedUserProperties == nil {
     125           1 :                 o.DeniedUserProperties = ignoredInternalProperties
     126           1 :         }
     127           1 :         return o
     128             : }
     129             : 
     130             : // WriterOptions holds the parameters used to control building an sstable.
     131             : type WriterOptions struct {
     132             :         // BlockRestartInterval is the number of keys between restart points
     133             :         // for delta encoding of keys.
     134             :         //
     135             :         // The default value is 16.
     136             :         BlockRestartInterval int
     137             : 
     138             :         // BlockSize is the target uncompressed size in bytes of each table block.
     139             :         //
     140             :         // The default value is 4096.
     141             :         BlockSize int
     142             : 
     143             :         // BlockSizeThreshold finishes a block if the block size is larger than the
     144             :         // specified percentage of the target block size and adding the next entry
     145             :         // would cause the block to be larger than the target block size.
     146             :         //
     147             :         // The default value is 90.
     148             :         BlockSizeThreshold int
     149             : 
     150             :         // SizeClassAwareThreshold imposes a minimum block size restriction for blocks
     151             :         // to be flushed, that is computed as the percentage of the target block size.
     152             :         // Note that this threshold takes precedence over BlockSizeThreshold when
     153             :         // valid AllocatorSizeClasses are specified.
     154             :         //
     155             :         // The default value is 60.
     156             :         SizeClassAwareThreshold int
     157             : 
     158             :         // Comparer defines a total ordering over the space of []byte keys: a 'less
     159             :         // than' relationship. The same comparison algorithm must be used for reads
     160             :         // and writes over the lifetime of the DB.
     161             :         //
     162             :         // The default value uses the same ordering as bytes.Compare.
     163             :         Comparer *Comparer
     164             : 
     165             :         // Compression defines the per-block compression to use.
     166             :         //
     167             :         // The default value (DefaultCompression) uses snappy compression.
     168             :         Compression block.Compression
     169             : 
     170             :         // FilterPolicy defines a filter algorithm (such as a Bloom filter) that can
     171             :         // reduce disk reads for Get calls.
     172             :         //
     173             :         // One such implementation is bloom.FilterPolicy(10) from the pebble/bloom
     174             :         // package.
     175             :         //
     176             :         // The default value means to use no filter.
     177             :         FilterPolicy FilterPolicy
     178             : 
     179             :         // FilterType defines whether an existing filter policy is applied at a
     180             :         // block-level or table-level. Block-level filters use less memory to create,
     181             :         // but are slower to access as a check for the key in the index must first be
     182             :         // performed to locate the filter block. A table-level filter will require
     183             :         // memory proportional to the number of keys in an sstable to create, but
     184             :         // avoids the index lookup when determining if a key is present. Table-level
     185             :         // filters should be preferred except under constrained memory situations.
     186             :         FilterType FilterType
     187             : 
     188             :         // IndexBlockSize is the target uncompressed size in bytes of each index
     189             :         // block. When the index block size is larger than this target, two-level
     190             :         // indexes are automatically enabled. Setting this option to a large value
     191             :         // (such as math.MaxInt32) disables the automatic creation of two-level
     192             :         // indexes.
     193             :         //
     194             :         // The default value is the value of BlockSize.
     195             :         IndexBlockSize int
     196             : 
     197             :         // Merger defines the associative merge operation to use for merging values
     198             :         // written with {Batch,DB}.Merge. The MergerName is checked for consistency
     199             :         // with the value stored in the sstable when it was written.
     200             :         MergerName string
     201             : 
     202             :         // TableFormat specifies the format version for writing sstables. The default
     203             :         // is TableFormatMinSupported.
     204             :         TableFormat TableFormat
     205             : 
     206             :         // IsStrictObsolete is only relevant for >= TableFormatPebblev4. See comment
     207             :         // in format.go. Must be false if format < TableFormatPebblev4.
     208             :         //
     209             :         // TODO(bilal): set this when writing shared ssts.
     210             :         IsStrictObsolete bool
     211             : 
     212             :         // WritingToLowestLevel is only relevant for >= TableFormatPebblev4. It is
     213             :         // used to set the obsolete bit on DEL/DELSIZED/SINGLEDEL if they are the
     214             :         // youngest for a userkey.
     215             :         WritingToLowestLevel bool
     216             : 
     217             :         // BlockPropertyCollectors is a list of BlockPropertyCollector creation
     218             :         // functions. A new BlockPropertyCollector is created for each sstable
     219             :         // built and lives for the lifetime of writing that table.
     220             :         BlockPropertyCollectors []func() BlockPropertyCollector
     221             : 
     222             :         // Checksum specifies which checksum to use.
     223             :         Checksum block.ChecksumType
     224             : 
     225             :         // Parallelism is used to indicate that the sstable Writer is allowed to
     226             :         // compress data blocks and write datablocks to disk in parallel with the
     227             :         // Writer client goroutine.
     228             :         Parallelism bool
     229             : 
     230             :         // ShortAttributeExtractor mirrors
     231             :         // Options.Experimental.ShortAttributeExtractor.
     232             :         ShortAttributeExtractor base.ShortAttributeExtractor
     233             : 
     234             :         // RequiredInPlaceValueBound mirrors
     235             :         // Options.Experimental.RequiredInPlaceValueBound.
     236             :         RequiredInPlaceValueBound UserKeyPrefixBound
     237             : 
     238             :         // DisableValueBlocks is only used for TableFormat >= TableFormatPebblev3,
     239             :         // and if set to true, does not write any values to value blocks. This is
     240             :         // only intended for cases where the in-memory buffering of all value blocks
     241             :         // while writing a sstable is too expensive and likely to cause an OOM. It
     242             :         // is never set to true by a Pebble DB, and can be set to true when some
     243             :         // external code is directly generating huge sstables using Pebble's
     244             :         // sstable.Writer (for example, CockroachDB backups can sometimes write
     245             :         // 750MB sstables -- see
     246             :         // https://github.com/cockroachdb/cockroach/issues/117113).
     247             :         DisableValueBlocks bool
     248             : 
     249             :         // AllocatorSizeClasses provides a sorted list containing the supported size
     250             :         // classes of the underlying memory allocator. This provides hints to the
     251             :         // writer's flushing policy to select block sizes that preemptively reduce
     252             :         // internal fragmentation when loaded into the block cache.
     253             :         AllocatorSizeClasses []int
     254             : 
     255             :         // internal options can only be used from within the pebble package.
     256             :         internal sstableinternal.WriterOptions
     257             : 
     258             :         // NumDeletionsThreshold mirrors Options.Experimental.NumDeletionsThreshold.
     259             :         NumDeletionsThreshold int
     260             : 
     261             :         // DeletionSizeRatioThreshold mirrors
     262             :         // Options.Experimental.DeletionSizeRatioThreshold.
     263             :         DeletionSizeRatioThreshold float32
     264             : }
     265             : 
     266             : // SetInternal sets the internal writer options. Note that even though this
     267             : // method is public, a caller outside the pebble package can't construct a value
     268             : // to pass to it.
     269           1 : func (o *WriterOptions) SetInternal(internalOpts sstableinternal.WriterOptions) {
     270           1 :         o.internal = internalOpts
     271           1 : }
     272             : 
     273           1 : func (o WriterOptions) ensureDefaults() WriterOptions {
     274           1 :         if o.BlockRestartInterval <= 0 {
     275           0 :                 o.BlockRestartInterval = base.DefaultBlockRestartInterval
     276           0 :         }
     277           1 :         if o.BlockSize <= 0 {
     278           0 :                 o.BlockSize = base.DefaultBlockSize
     279           0 :         }
     280           1 :         if o.BlockSizeThreshold <= 0 {
     281           0 :                 o.BlockSizeThreshold = base.DefaultBlockSizeThreshold
     282           0 :         }
     283           1 :         if o.SizeClassAwareThreshold <= 0 {
     284           1 :                 o.SizeClassAwareThreshold = base.SizeClassAwareBlockSizeThreshold
     285           1 :         }
     286           1 :         if o.Comparer == nil {
     287           0 :                 o.Comparer = base.DefaultComparer
     288           0 :         }
     289           1 :         if o.Compression <= block.DefaultCompression || o.Compression >= block.NCompression {
     290           0 :                 o.Compression = block.SnappyCompression
     291           0 :         }
     292           1 :         if o.IndexBlockSize <= 0 {
     293           0 :                 o.IndexBlockSize = o.BlockSize
     294           0 :         }
     295           1 :         if o.MergerName == "" {
     296           0 :                 o.MergerName = base.DefaultMerger.Name
     297           0 :         }
     298           1 :         if o.Checksum == block.ChecksumTypeNone {
     299           1 :                 o.Checksum = block.ChecksumTypeCRC32c
     300           1 :         }
     301             :         // By default, if the table format is not specified, fall back to using the
     302             :         // most compatible format that is supported by Pebble.
     303           1 :         if o.TableFormat == TableFormatUnspecified {
     304           0 :                 o.TableFormat = TableFormatMinSupported
     305           0 :         }
     306           1 :         if o.NumDeletionsThreshold == 0 {
     307           0 :                 o.NumDeletionsThreshold = DefaultNumDeletionsThreshold
     308           0 :         }
     309           1 :         if o.DeletionSizeRatioThreshold == 0 {
     310           0 :                 o.DeletionSizeRatioThreshold = DefaultDeletionSizeRatioThreshold
     311           0 :         }
     312           1 :         return o
     313             : }