eleksir/bunny-bot
0
Fork 0
Code Pull Requests Projects Activity

Initial commit

Browse Source
This commit is contained in:
eleksir
2025-05-08 18:12:52 +03:00
commit cff1b6811f
1596 changed files with 659567 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

60
vendor/github.com/cockroachdb/pebble/internal/base/cleaner.go generated vendored Normal file
View File

@@ -0,0 +1,60 @@
// Copyright 2019 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base
import "github.com/cockroachdb/pebble/vfs"
// Cleaner cleans obsolete files.
type Cleaner interface {
Clean(fs vfs.FS, fileType FileType, path string) error
}
// NeedsFileContents is implemented by a cleaner that needs the contents of the
// files that it is being asked to clean.
type NeedsFileContents interface {
needsFileContents()
}
// DeleteCleaner deletes file.
type DeleteCleaner struct{}
// Clean removes file.
func (DeleteCleaner) Clean(fs vfs.FS, fileType FileType, path string) error {
return fs.Remove(path)
}
func (DeleteCleaner) String() string {
return "delete"
}
// ArchiveCleaner archives file instead delete.
type ArchiveCleaner struct{}
var _ NeedsFileContents = ArchiveCleaner{}
// Clean archives file.
func (ArchiveCleaner) Clean(fs vfs.FS, fileType FileType, path string) error {
switch fileType {
case FileTypeLog, FileTypeManifest, FileTypeTable:
destDir := fs.PathJoin(fs.PathDir(path), "archive")
if err := fs.MkdirAll(destDir, 0755); err != nil {
return err
}
destPath := fs.PathJoin(destDir, fs.PathBase(path))
return fs.Rename(path, destPath)
default:
return fs.Remove(path)
}
}
func (ArchiveCleaner) String() string {
return "archive"
}
func (ArchiveCleaner) needsFileContents() {
}

260
vendor/github.com/cockroachdb/pebble/internal/base/comparer.go generated vendored Normal file
View File

@@ -0,0 +1,260 @@
// Copyright 2011 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base
import (
"bytes"
"encoding/binary"
"fmt"
"strconv"
"unicode/utf8"
)
// Compare returns -1, 0, or +1 depending on whether a is 'less than', 'equal
// to' or 'greater than' b. The two arguments can only be 'equal' if their
// contents are exactly equal. Furthermore, the empty slice must be 'less than'
// any non-empty slice. Compare is used to compare user keys, such as those
// passed as arguments to the various DB methods, as well as those returned
// from Separator, Successor, and Split.
type Compare func(a, b []byte) int
// Equal returns true if a and b are equivalent. For a given Compare,
// Equal(a,b) must return true iff Compare(a,b) returns zero, that is,
// Equal is a (potentially faster) specialization of Compare.
type Equal func(a, b []byte) bool
// AbbreviatedKey returns a fixed length prefix of a user key such that AbbreviatedKey(a)
// < AbbreviatedKey(b) iff a < b and AbbreviatedKey(a) > AbbreviatedKey(b) iff a > b. If
// AbbreviatedKey(a) == AbbreviatedKey(b) an additional comparison is required to
// determine if the two keys are actually equal.
//
// This helps optimize indexed batch comparisons for cache locality. If a Split
// function is specified, AbbreviatedKey usually returns the first eight bytes
// of the user key prefix in the order that gives the correct ordering.
type AbbreviatedKey func(key []byte) uint64
// FormatKey returns a formatter for the user key.
type FormatKey func(key []byte) fmt.Formatter
// FormatValue returns a formatter for the user value. The key is also
// specified for the value formatter in order to support value formatting that
// is dependent on the key.
type FormatValue func(key, value []byte) fmt.Formatter
// Separator is used to construct SSTable index blocks. A trivial implementation
// is `return a`, but appending fewer bytes leads to smaller SSTables.
//
// Given keys a, b for which Compare(a, b) < 0, Separator returns a key k such
// that:
//
// 1. Compare(a, k) <= 0, and
// 2. Compare(k, b) < 0.
//
// As a special case, b may be nil in which case the second condition is dropped.
//
// For example, if dst, a and b are the []byte equivalents of the strings
// "aqua", "black" and "blue", then the result may be "aquablb".
// Similarly, if the arguments were "aqua", "green" and "", then the result
// may be "aquah".
type Separator func(dst, a, b []byte) []byte
// Successor returns a shortened key given a key a, such that Compare(k, a) >=
// 0. A simple implementation may return a unchanged. The dst parameter may be
// used to store the returned key, though it is valid to pass nil. The returned
// key must be valid to pass to Compare.
type Successor func(dst, a []byte) []byte
// ImmediateSuccessor is invoked with a prefix key ([Split(a) == len(a)]) and
// returns the smallest key that is larger than the given prefix a.
// ImmediateSuccessor must return a prefix key k such that:
//
// Split(k) == len(k) and Compare(k, a) > 0
//
// and there exists no representable k2 such that:
//
// Split(k2) == len(k2) and Compare(k2, a) > 0 and Compare(k2, k) < 0
//
// As an example, an implementation built on the natural byte ordering using
// bytes.Compare could append a `\0` to `a`.
//
// The dst parameter may be used to store the returned key, though it is valid
// to pass nil. The returned key must be valid to pass to Compare.
type ImmediateSuccessor func(dst, a []byte) []byte
// Split returns the length of the prefix of the user key that corresponds to
// the key portion of an MVCC encoding scheme to enable the use of prefix bloom
// filters.
//
// The method will only ever be called with valid MVCC keys, that is, keys that
// the user could potentially store in the database. Pebble does not know which
// keys are MVCC keys and which are not, and may call Split on both MVCC keys
// and non-MVCC keys.
//
// A trivial MVCC scheme is one in which Split() returns len(a). This
// corresponds to assigning a constant version to each key in the database. For
// performance reasons, it is preferable to use a `nil` split in this case.
//
// The returned prefix must have the following properties:
//
// 1. The prefix must be a byte prefix:
//
// bytes.HasPrefix(a, prefix(a))
//
// 2. A key consisting of just a prefix must sort before all other keys with
// that prefix:
//
// Compare(prefix(a), a) < 0 if len(suffix(a)) > 0
//
// 3. Prefixes must be used to order keys before suffixes:
//
// If Compare(a, b) <= 0, then Compare(prefix(a), prefix(b)) <= 0
//
// 4. Suffixes themselves must be valid keys and comparable, respecting the same
// ordering as within a key.
//
// If Compare(prefix(a), prefix(b)) == 0, then Compare(suffix(a), suffix(b)) == Compare(a, b)
type Split func(a []byte) int
// Comparer defines a total ordering over the space of []byte keys: a 'less
// than' relationship.
type Comparer struct {
Compare Compare
Equal Equal
AbbreviatedKey AbbreviatedKey
FormatKey FormatKey
FormatValue FormatValue
Separator Separator
Split Split
Successor Successor
ImmediateSuccessor ImmediateSuccessor
// Name is the name of the comparer.
//
// The Level-DB on-disk format stores the comparer name, and opening a
// database with a different comparer from the one it was created with
// will result in an error.
Name string
}
// DefaultFormatter is the default implementation of user key formatting:
// non-ASCII data is formatted as escaped hexadecimal values.
var DefaultFormatter = func(key []byte) fmt.Formatter {
return FormatBytes(key)
}
// DefaultComparer is the default implementation of the Comparer interface.
// It uses the natural ordering, consistent with bytes.Compare.
var DefaultComparer = &Comparer{
Compare: bytes.Compare,
Equal: bytes.Equal,
AbbreviatedKey: func(key []byte) uint64 {
if len(key) >= 8 {
return binary.BigEndian.Uint64(key)
}
var v uint64
for _, b := range key {
v <<= 8
v |= uint64(b)
}
return v << uint(8*(8-len(key)))
},
FormatKey: DefaultFormatter,
Separator: func(dst, a, b []byte) []byte {
i, n := SharedPrefixLen(a, b), len(dst)
dst = append(dst, a...)
min := len(a)
if min > len(b) {
min = len(b)
}
if i >= min {
// Do not shorten if one string is a prefix of the other.
return dst
}
if a[i] >= b[i] {
// b is smaller than a or a is already the shortest possible.
return dst
}
if i < len(b)-1 || a[i]+1 < b[i] {
i += n
dst[i]++
return dst[:i+1]
}
i += n + 1
for ; i < len(dst); i++ {
if dst[i] != 0xff {
dst[i]++
return dst[:i+1]
}
}
return dst
},
Successor: func(dst, a []byte) (ret []byte) {
for i := 0; i < len(a); i++ {
if a[i] != 0xff {
dst = append(dst, a[:i+1]...)
dst[len(dst)-1]++
return dst
}
}
// a is a run of 0xffs, leave it alone.
return append(dst, a...)
},
ImmediateSuccessor: func(dst, a []byte) (ret []byte) {
return append(append(dst, a...), 0x00)
},
// This name is part of the C++ Level-DB implementation's default file
// format, and should not be changed.
Name: "leveldb.BytewiseComparator",
}
// SharedPrefixLen returns the largest i such that a[:i] equals b[:i].
// This function can be useful in implementing the Comparer interface.
func SharedPrefixLen(a, b []byte) int {
i, n := 0, len(a)
if n > len(b) {
n = len(b)
}
asUint64 := func(c []byte, i int) uint64 {
return binary.LittleEndian.Uint64(c[i:])
}
for i < n-7 && asUint64(a, i) == asUint64(b, i) {
i += 8
}
for i < n && a[i] == b[i] {
i++
}
return i
}
// FormatBytes formats a byte slice using hexadecimal escapes for non-ASCII
// data.
type FormatBytes []byte
const lowerhex = "0123456789abcdef"
// Format implements the fmt.Formatter interface.
func (p FormatBytes) Format(s fmt.State, c rune) {
buf := make([]byte, 0, len(p))
for _, b := range p {
if b < utf8.RuneSelf && strconv.IsPrint(rune(b)) {
buf = append(buf, b)
continue
}
buf = append(buf, `\x`...)
buf = append(buf, lowerhex[b>>4])
buf = append(buf, lowerhex[b&0xF])
}
s.Write(buf)
}

28
vendor/github.com/cockroachdb/pebble/internal/base/error.go generated vendored Normal file
View File

@@ -0,0 +1,28 @@
// Copyright 2011 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base
import "github.com/cockroachdb/errors"
// ErrNotFound means that a get or delete call did not find the requested key.
var ErrNotFound = errors.New("pebble: not found")
// ErrCorruption is a marker to indicate that data in a file (WAL, MANIFEST,
// sstable) isn't in the expected format.
var ErrCorruption = errors.New("pebble: corruption")
// MarkCorruptionError marks given error as a corruption error.
func MarkCorruptionError(err error) error {
if errors.Is(err, ErrCorruption) {
return err
}
return errors.Mark(err, ErrCorruption)
}
// CorruptionErrorf formats according to a format specifier and returns
// the string as an error value that is marked as a corruption error.
func CorruptionErrorf(format string, args ...interface{}) error {
return errors.Mark(errors.Newf(format, args...), ErrCorruption)
}

204
vendor/github.com/cockroachdb/pebble/internal/base/filenames.go generated vendored Normal file
View File

@@ -0,0 +1,204 @@
// Copyright 2012 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base
import (
"fmt"
"strconv"
"strings"
"github.com/cockroachdb/errors/oserror"
"github.com/cockroachdb/pebble/vfs"
"github.com/cockroachdb/redact"
)
// FileNum is an internal DB identifier for a file.
type FileNum uint64
// String returns a string representation of the file number.
func (fn FileNum) String() string { return fmt.Sprintf("%06d", fn) }
// SafeFormat implements redact.SafeFormatter.
func (fn FileNum) SafeFormat(w redact.SafePrinter, _ rune) {
w.Printf("%06d", redact.SafeUint(fn))
}
// DiskFileNum converts a FileNum to a DiskFileNum. DiskFileNum should only be
// called if the caller can ensure that the FileNum belongs to a physical file
// on disk. These could be manifests, log files, physical sstables on disk, the
// options file, but not virtual sstables.
func (fn FileNum) DiskFileNum() DiskFileNum {
return DiskFileNum{fn}
}
// A DiskFileNum is just a FileNum belonging to a file which exists on disk.
// Note that a FileNum is an internal DB identifier and it could belong to files
// which don't exist on disk. An example would be virtual sstable FileNums.
// Converting a DiskFileNum to a FileNum is always valid, whereas converting a
// FileNum to DiskFileNum may not be valid and care should be taken to prove
// that the FileNum actually exists on disk.
type DiskFileNum struct {
fn FileNum
}
func (dfn DiskFileNum) String() string { return dfn.fn.String() }
// SafeFormat implements redact.SafeFormatter.
func (dfn DiskFileNum) SafeFormat(w redact.SafePrinter, verb rune) {
dfn.fn.SafeFormat(w, verb)
}
// FileNum converts a DiskFileNum to a FileNum. This conversion is always valid.
func (dfn DiskFileNum) FileNum() FileNum {
return dfn.fn
}
// FileType enumerates the types of files found in a DB.
type FileType int
// The FileType enumeration.
const (
FileTypeLog FileType = iota
FileTypeLock
FileTypeTable
FileTypeManifest
FileTypeCurrent
FileTypeOptions
FileTypeOldTemp
FileTypeTemp
)
// MakeFilename builds a filename from components.
func MakeFilename(fileType FileType, dfn DiskFileNum) string {
switch fileType {
case FileTypeLog:
return fmt.Sprintf("%s.log", dfn)
case FileTypeLock:
return "LOCK"
case FileTypeTable:
return fmt.Sprintf("%s.sst", dfn)
case FileTypeManifest:
return fmt.Sprintf("MANIFEST-%s", dfn)
case FileTypeCurrent:
return "CURRENT"
case FileTypeOptions:
return fmt.Sprintf("OPTIONS-%s", dfn)
case FileTypeOldTemp:
return fmt.Sprintf("CURRENT.%s.dbtmp", dfn)
case FileTypeTemp:
return fmt.Sprintf("temporary.%s.dbtmp", dfn)
}
panic("unreachable")
}
// MakeFilepath builds a filepath from components.
func MakeFilepath(fs vfs.FS, dirname string, fileType FileType, dfn DiskFileNum) string {
return fs.PathJoin(dirname, MakeFilename(fileType, dfn))
}
// ParseFilename parses the components from a filename.
func ParseFilename(fs vfs.FS, filename string) (fileType FileType, dfn DiskFileNum, ok bool) {
filename = fs.PathBase(filename)
switch {
case filename == "CURRENT":
return FileTypeCurrent, DiskFileNum{0}, true
case filename == "LOCK":
return FileTypeLock, DiskFileNum{0}, true
case strings.HasPrefix(filename, "MANIFEST-"):
dfn, ok = parseDiskFileNum(filename[len("MANIFEST-"):])
if !ok {
break
}
return FileTypeManifest, dfn, true
case strings.HasPrefix(filename, "OPTIONS-"):
dfn, ok = parseDiskFileNum(filename[len("OPTIONS-"):])
if !ok {
break
}
return FileTypeOptions, dfn, ok
case strings.HasPrefix(filename, "CURRENT.") && strings.HasSuffix(filename, ".dbtmp"):
s := strings.TrimSuffix(filename[len("CURRENT."):], ".dbtmp")
dfn, ok = parseDiskFileNum(s)
if !ok {
break
}
return FileTypeOldTemp, dfn, ok
case strings.HasPrefix(filename, "temporary.") && strings.HasSuffix(filename, ".dbtmp"):
s := strings.TrimSuffix(filename[len("temporary."):], ".dbtmp")
dfn, ok = parseDiskFileNum(s)
if !ok {
break
}
return FileTypeTemp, dfn, ok
default:
i := strings.IndexByte(filename, '.')
if i < 0 {
break
}
dfn, ok = parseDiskFileNum(filename[:i])
if !ok {
break
}
switch filename[i+1:] {
case "sst":
return FileTypeTable, dfn, true
case "log":
return FileTypeLog, dfn, true
}
}
return 0, dfn, false
}
func parseDiskFileNum(s string) (dfn DiskFileNum, ok bool) {
u, err := strconv.ParseUint(s, 10, 64)
if err != nil {
return dfn, false
}
return DiskFileNum{FileNum(u)}, true
}
// A Fataler fatals a process with a message when called.
type Fataler interface {
Fatalf(format string, args ...interface{})
}
// MustExist checks if err is an error indicating a file does not exist.
// If it is, it lists the containing directory's files to annotate the error
// with counts of the various types of files and invokes the provided fataler.
// See cockroachdb/cockroach#56490.
func MustExist(fs vfs.FS, filename string, fataler Fataler, err error) {
if err == nil || !oserror.IsNotExist(err) {
return
}
ls, lsErr := fs.List(fs.PathDir(filename))
if lsErr != nil {
// TODO(jackson): if oserror.IsNotExist(lsErr), the the data directory
// doesn't exist anymore. Another process likely deleted it before
// killing the process. We want to fatal the process, but without
// triggering error reporting like Sentry.
fataler.Fatalf("%s:\norig err: %s\nlist err: %s", redact.Safe(fs.PathBase(filename)), err, lsErr)
}
var total, unknown, tables, logs, manifests int
total = len(ls)
for _, f := range ls {
typ, _, ok := ParseFilename(fs, f)
if !ok {
unknown++
continue
}
switch typ {
case FileTypeTable:
tables++
case FileTypeLog:
logs++
case FileTypeManifest:
manifests++
}
}
fataler.Fatalf("%s:\n%s\ndirectory contains %d files, %d unknown, %d tables, %d logs, %d manifests",
fs.PathBase(filename), err, total, unknown, tables, logs, manifests)
}

506
vendor/github.com/cockroachdb/pebble/internal/base/internal.go generated vendored Normal file
View File

@@ -0,0 +1,506 @@
// Copyright 2011 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base // import "github.com/cockroachdb/pebble/internal/base"
import (
"encoding/binary"
"fmt"
"strconv"
"strings"
"github.com/cockroachdb/redact"
)
const (
// SeqNumZero is the zero sequence number, set by compactions if they can
// guarantee there are no keys underneath an internal key.
SeqNumZero = uint64(0)
// SeqNumStart is the first sequence number assigned to a key. Sequence
// numbers 1-9 are reserved for potential future use.
SeqNumStart = uint64(10)
)
// InternalKeyKind enumerates the kind of key: a deletion tombstone, a set
// value, a merged value, etc.
type InternalKeyKind uint8
// These constants are part of the file format, and should not be changed.
const (
InternalKeyKindDelete InternalKeyKind = 0
InternalKeyKindSet InternalKeyKind = 1
InternalKeyKindMerge InternalKeyKind = 2
InternalKeyKindLogData InternalKeyKind = 3
//InternalKeyKindColumnFamilyDeletion InternalKeyKind = 4
//InternalKeyKindColumnFamilyValue InternalKeyKind = 5
//InternalKeyKindColumnFamilyMerge InternalKeyKind = 6
// InternalKeyKindSingleDelete (SINGLEDEL) is a performance optimization
// solely for compactions (to reduce write amp and space amp). Readers other
// than compactions should treat SINGLEDEL as equivalent to a DEL.
// Historically, it was simpler for readers other than compactions to treat
// SINGLEDEL as equivalent to DEL, but as of the introduction of
// InternalKeyKindSSTableInternalObsoleteBit, this is also necessary for
// correctness.
InternalKeyKindSingleDelete InternalKeyKind = 7
//InternalKeyKindColumnFamilySingleDelete InternalKeyKind = 8
//InternalKeyKindBeginPrepareXID InternalKeyKind = 9
//InternalKeyKindEndPrepareXID InternalKeyKind = 10
//InternalKeyKindCommitXID InternalKeyKind = 11
//InternalKeyKindRollbackXID InternalKeyKind = 12
//InternalKeyKindNoop InternalKeyKind = 13
//InternalKeyKindColumnFamilyRangeDelete InternalKeyKind = 14
InternalKeyKindRangeDelete InternalKeyKind = 15
//InternalKeyKindColumnFamilyBlobIndex InternalKeyKind = 16
//InternalKeyKindBlobIndex InternalKeyKind = 17
// InternalKeyKindSeparator is a key used for separator / successor keys
// written to sstable block indexes.
//
// NOTE: the RocksDB value has been repurposed. This was done to ensure that
// keys written to block indexes with value "17" (when 17 happened to be the
// max value, and InternalKeyKindMax was therefore set to 17), remain stable
// when new key kinds are supported in Pebble.
InternalKeyKindSeparator InternalKeyKind = 17
// InternalKeyKindSetWithDelete keys are SET keys that have met with a
// DELETE or SINGLEDEL key in a prior compaction. This key kind is
// specific to Pebble. See
// https://github.com/cockroachdb/pebble/issues/1255.
InternalKeyKindSetWithDelete InternalKeyKind = 18
// InternalKeyKindRangeKeyDelete removes all range keys within a key range.
// See the internal/rangekey package for more details.
InternalKeyKindRangeKeyDelete InternalKeyKind = 19
// InternalKeyKindRangeKeySet and InternalKeyKindRangeUnset represent
// keys that set and unset values associated with ranges of key
// space. See the internal/rangekey package for more details.
InternalKeyKindRangeKeyUnset InternalKeyKind = 20
InternalKeyKindRangeKeySet InternalKeyKind = 21
// InternalKeyKindIngestSST is used to distinguish a batch that corresponds to
// the WAL entry for ingested sstables that are added to the flushable
// queue. This InternalKeyKind cannot appear, amongst other key kinds in a
// batch, or in an sstable.
InternalKeyKindIngestSST InternalKeyKind = 22
// InternalKeyKindDeleteSized keys behave identically to
// InternalKeyKindDelete keys, except that they hold an associated uint64
// value indicating the (len(key)+len(value)) of the shadowed entry the
// tombstone is expected to delete. This value is used to inform compaction
// heuristics, but is not required to be accurate for correctness.
InternalKeyKindDeleteSized InternalKeyKind = 23
// This maximum value isn't part of the file format. Future extensions may
// increase this value.
//
// When constructing an internal key to pass to DB.Seek{GE,LE},
// internalKeyComparer sorts decreasing by kind (after sorting increasing by
// user key and decreasing by sequence number). Thus, use InternalKeyKindMax,
// which sorts 'less than or equal to' any other valid internalKeyKind, when
// searching for any kind of internal key formed by a certain user key and
// seqNum.
InternalKeyKindMax InternalKeyKind = 23
// Internal to the sstable format. Not exposed by any sstable iterator.
// Declared here to prevent definition of valid key kinds that set this bit.
InternalKeyKindSSTableInternalObsoleteBit InternalKeyKind = 64
InternalKeyKindSSTableInternalObsoleteMask InternalKeyKind = 191
// InternalKeyZeroSeqnumMaxTrailer is the largest trailer with a
// zero sequence number.
InternalKeyZeroSeqnumMaxTrailer = uint64(255)
// A marker for an invalid key.
InternalKeyKindInvalid InternalKeyKind = InternalKeyKindSSTableInternalObsoleteMask
// InternalKeySeqNumBatch is a bit that is set on batch sequence numbers
// which prevents those entries from being excluded from iteration.
InternalKeySeqNumBatch = uint64(1 << 55)
// InternalKeySeqNumMax is the largest valid sequence number.
InternalKeySeqNumMax = uint64(1<<56 - 1)
// InternalKeyRangeDeleteSentinel is the marker for a range delete sentinel
// key. This sequence number and kind are used for the upper stable boundary
// when a range deletion tombstone is the largest key in an sstable. This is
// necessary because sstable boundaries are inclusive, while the end key of a
// range deletion tombstone is exclusive.
InternalKeyRangeDeleteSentinel = (InternalKeySeqNumMax << 8) | uint64(InternalKeyKindRangeDelete)
// InternalKeyBoundaryRangeKey is the marker for a range key boundary. This
// sequence number and kind are used during interleaved range key and point
// iteration to allow an iterator to stop at range key start keys where
// there exists no point key.
InternalKeyBoundaryRangeKey = (InternalKeySeqNumMax << 8) | uint64(InternalKeyKindRangeKeySet)
)
// Assert InternalKeyKindSSTableInternalObsoleteBit > InternalKeyKindMax
const _ = uint(InternalKeyKindSSTableInternalObsoleteBit - InternalKeyKindMax - 1)
var internalKeyKindNames = []string{
InternalKeyKindDelete: "DEL",
InternalKeyKindSet: "SET",
InternalKeyKindMerge: "MERGE",
InternalKeyKindLogData: "LOGDATA",
InternalKeyKindSingleDelete: "SINGLEDEL",
InternalKeyKindRangeDelete: "RANGEDEL",
InternalKeyKindSeparator: "SEPARATOR",
InternalKeyKindSetWithDelete: "SETWITHDEL",
InternalKeyKindRangeKeySet: "RANGEKEYSET",
InternalKeyKindRangeKeyUnset: "RANGEKEYUNSET",
InternalKeyKindRangeKeyDelete: "RANGEKEYDEL",
InternalKeyKindIngestSST: "INGESTSST",
InternalKeyKindDeleteSized: "DELSIZED",
InternalKeyKindInvalid: "INVALID",
}
func (k InternalKeyKind) String() string {
if int(k) < len(internalKeyKindNames) {
return internalKeyKindNames[k]
}
return fmt.Sprintf("UNKNOWN:%d", k)
}
// SafeFormat implements redact.SafeFormatter.
func (k InternalKeyKind) SafeFormat(w redact.SafePrinter, _ rune) {
w.Print(redact.SafeString(k.String()))
}
// InternalKey is a key used for the in-memory and on-disk partial DBs that
// make up a pebble DB.
//
// It consists of the user key (as given by the code that uses package pebble)
// followed by 8-bytes of metadata:
// - 1 byte for the type of internal key: delete or set,
// - 7 bytes for a uint56 sequence number, in little-endian format.
type InternalKey struct {
UserKey []byte
Trailer uint64
}
// InvalidInternalKey is an invalid internal key for which Valid() will return
// false.
var InvalidInternalKey = MakeInternalKey(nil, 0, InternalKeyKindInvalid)
// MakeInternalKey constructs an internal key from a specified user key,
// sequence number and kind.
func MakeInternalKey(userKey []byte, seqNum uint64, kind InternalKeyKind) InternalKey {
return InternalKey{
UserKey: userKey,
Trailer: (seqNum << 8) | uint64(kind),
}
}
// MakeTrailer constructs an internal key trailer from the specified sequence
// number and kind.
func MakeTrailer(seqNum uint64, kind InternalKeyKind) uint64 {
return (seqNum << 8) | uint64(kind)
}
// MakeSearchKey constructs an internal key that is appropriate for searching
// for a the specified user key. The search key contain the maximal sequence
// number and kind ensuring that it sorts before any other internal keys for
// the same user key.
func MakeSearchKey(userKey []byte) InternalKey {
return InternalKey{
UserKey: userKey,
Trailer: (InternalKeySeqNumMax << 8) | uint64(InternalKeyKindMax),
}
}
// MakeRangeDeleteSentinelKey constructs an internal key that is a range
// deletion sentinel key, used as the upper boundary for an sstable when a
// range deletion is the largest key in an sstable.
func MakeRangeDeleteSentinelKey(userKey []byte) InternalKey {
return InternalKey{
UserKey: userKey,
Trailer: InternalKeyRangeDeleteSentinel,
}
}
// MakeExclusiveSentinelKey constructs an internal key that is an
// exclusive sentinel key, used as the upper boundary for an sstable
// when a ranged key is the largest key in an sstable.
func MakeExclusiveSentinelKey(kind InternalKeyKind, userKey []byte) InternalKey {
return InternalKey{
UserKey: userKey,
Trailer: (InternalKeySeqNumMax << 8) | uint64(kind),
}
}
var kindsMap = map[string]InternalKeyKind{
"DEL": InternalKeyKindDelete,
"SINGLEDEL": InternalKeyKindSingleDelete,
"RANGEDEL": InternalKeyKindRangeDelete,
"LOGDATA": InternalKeyKindLogData,
"SET": InternalKeyKindSet,
"MERGE": InternalKeyKindMerge,
"INVALID": InternalKeyKindInvalid,
"SEPARATOR": InternalKeyKindSeparator,
"SETWITHDEL": InternalKeyKindSetWithDelete,
"RANGEKEYSET": InternalKeyKindRangeKeySet,
"RANGEKEYUNSET": InternalKeyKindRangeKeyUnset,
"RANGEKEYDEL": InternalKeyKindRangeKeyDelete,
"INGESTSST": InternalKeyKindIngestSST,
"DELSIZED": InternalKeyKindDeleteSized,
}
// ParseInternalKey parses the string representation of an internal key. The
// format is <user-key>.<kind>.<seq-num>. If the seq-num starts with a "b" it
// is marked as a batch-seq-num (i.e. the InternalKeySeqNumBatch bit is set).
func ParseInternalKey(s string) InternalKey {
x := strings.Split(s, ".")
ukey := x[0]
kind, ok := kindsMap[x[1]]
if !ok {
panic(fmt.Sprintf("unknown kind: %q", x[1]))
}
j := 0
if x[2][0] == 'b' {
j = 1
}
seqNum, _ := strconv.ParseUint(x[2][j:], 10, 64)
if x[2][0] == 'b' {
seqNum |= InternalKeySeqNumBatch
}
return MakeInternalKey([]byte(ukey), seqNum, kind)
}
// ParseKind parses the string representation of an internal key kind.
func ParseKind(s string) InternalKeyKind {
kind, ok := kindsMap[s]
if !ok {
panic(fmt.Sprintf("unknown kind: %q", s))
}
return kind
}
// InternalTrailerLen is the number of bytes used to encode InternalKey.Trailer.
const InternalTrailerLen = 8
// DecodeInternalKey decodes an encoded internal key. See InternalKey.Encode().
func DecodeInternalKey(encodedKey []byte) InternalKey {
n := len(encodedKey) - InternalTrailerLen
var trailer uint64
if n >= 0 {
trailer = binary.LittleEndian.Uint64(encodedKey[n:])
encodedKey = encodedKey[:n:n]
} else {
trailer = uint64(InternalKeyKindInvalid)
encodedKey = nil
}
return InternalKey{
UserKey: encodedKey,
Trailer: trailer,
}
}
// InternalCompare compares two internal keys using the specified comparison
// function. For equal user keys, internal keys compare in descending sequence
// number order. For equal user keys and sequence numbers, internal keys
// compare in descending kind order (this may happen in practice among range
// keys).
func InternalCompare(userCmp Compare, a, b InternalKey) int {
if x := userCmp(a.UserKey, b.UserKey); x != 0 {
return x
}
if a.Trailer > b.Trailer {
return -1
}
if a.Trailer < b.Trailer {
return 1
}
return 0
}
// Encode encodes the receiver into the buffer. The buffer must be large enough
// to hold the encoded data. See InternalKey.Size().
func (k InternalKey) Encode(buf []byte) {
i := copy(buf, k.UserKey)
binary.LittleEndian.PutUint64(buf[i:], k.Trailer)
}
// EncodeTrailer returns the trailer encoded to an 8-byte array.
func (k InternalKey) EncodeTrailer() [8]byte {
var buf [8]byte
binary.LittleEndian.PutUint64(buf[:], k.Trailer)
return buf
}
// Separator returns a separator key such that k <= x && x < other, where less
// than is consistent with the Compare function. The buf parameter may be used
// to store the returned InternalKey.UserKey, though it is valid to pass a
// nil. See the Separator type for details on separator keys.
func (k InternalKey) Separator(
cmp Compare, sep Separator, buf []byte, other InternalKey,
) InternalKey {
buf = sep(buf, k.UserKey, other.UserKey)
if len(buf) <= len(k.UserKey) && cmp(k.UserKey, buf) < 0 {
// The separator user key is physically shorter than k.UserKey (if it is
// longer, we'll continue to use "k"), but logically after. Tack on the max
// sequence number to the shortened user key. Note that we could tack on
// any sequence number and kind here to create a valid separator key. We
// use the max sequence number to match the behavior of LevelDB and
// RocksDB.
return MakeInternalKey(buf, InternalKeySeqNumMax, InternalKeyKindSeparator)
}
return k
}
// Successor returns a successor key such that k <= x. A simple implementation
// may return k unchanged. The buf parameter may be used to store the returned
// InternalKey.UserKey, though it is valid to pass a nil.
func (k InternalKey) Successor(cmp Compare, succ Successor, buf []byte) InternalKey {
buf = succ(buf, k.UserKey)
if len(buf) <= len(k.UserKey) && cmp(k.UserKey, buf) < 0 {
// The successor user key is physically shorter that k.UserKey (if it is
// longer, we'll continue to use "k"), but logically after. Tack on the max
// sequence number to the shortened user key. Note that we could tack on
// any sequence number and kind here to create a valid separator key. We
// use the max sequence number to match the behavior of LevelDB and
// RocksDB.
return MakeInternalKey(buf, InternalKeySeqNumMax, InternalKeyKindSeparator)
}
return k
}
// Size returns the encoded size of the key.
func (k InternalKey) Size() int {
return len(k.UserKey) + 8
}
// SetSeqNum sets the sequence number component of the key.
func (k *InternalKey) SetSeqNum(seqNum uint64) {
k.Trailer = (seqNum << 8) | (k.Trailer & 0xff)
}
// SeqNum returns the sequence number component of the key.
func (k InternalKey) SeqNum() uint64 {
return k.Trailer >> 8
}
// SeqNumFromTrailer returns the sequence number component of a trailer.
func SeqNumFromTrailer(t uint64) uint64 {
return t >> 8
}
// Visible returns true if the key is visible at the specified snapshot
// sequence number.
func (k InternalKey) Visible(snapshot, batchSnapshot uint64) bool {
return Visible(k.SeqNum(), snapshot, batchSnapshot)
}
// Visible returns true if a key with the provided sequence number is visible at
// the specified snapshot sequence numbers.
func Visible(seqNum uint64, snapshot, batchSnapshot uint64) bool {
// There are two snapshot sequence numbers, one for committed keys and one
// for batch keys. If a seqNum is less than `snapshot`, then seqNum
// corresponds to a committed key that is visible. If seqNum has its batch
// bit set, then seqNum corresponds to an uncommitted batch key. Its
// visible if its snapshot is less than batchSnapshot.
//
// There's one complication. The maximal sequence number
// (`InternalKeySeqNumMax`) is used across Pebble for exclusive sentinel
// keys and other purposes. The maximal sequence number has its batch bit
// set, but it can never be < `batchSnapshot`, since there is no expressible
// larger snapshot. We dictate that the maximal sequence number is always
// visible.
return seqNum < snapshot ||
((seqNum&InternalKeySeqNumBatch) != 0 && seqNum < batchSnapshot) ||
seqNum == InternalKeySeqNumMax
}
// SetKind sets the kind component of the key.
func (k *InternalKey) SetKind(kind InternalKeyKind) {
k.Trailer = (k.Trailer &^ 0xff) | uint64(kind)
}
// Kind returns the kind component of the key.
func (k InternalKey) Kind() InternalKeyKind {
return TrailerKind(k.Trailer)
}
// TrailerKind returns the key kind of the key trailer.
func TrailerKind(trailer uint64) InternalKeyKind {
return InternalKeyKind(trailer & 0xff)
}
// Valid returns true if the key has a valid kind.
func (k InternalKey) Valid() bool {
return k.Kind() <= InternalKeyKindMax
}
// Clone clones the storage for the UserKey component of the key.
func (k InternalKey) Clone() InternalKey {
if len(k.UserKey) == 0 {
return k
}
return InternalKey{
UserKey: append([]byte(nil), k.UserKey...),
Trailer: k.Trailer,
}
}
// CopyFrom converts this InternalKey into a clone of the passed-in InternalKey,
// reusing any space already used for the current UserKey.
func (k *InternalKey) CopyFrom(k2 InternalKey) {
k.UserKey = append(k.UserKey[:0], k2.UserKey...)
k.Trailer = k2.Trailer
}
// String returns a string representation of the key.
func (k InternalKey) String() string {
return fmt.Sprintf("%s#%d,%d", FormatBytes(k.UserKey), k.SeqNum(), k.Kind())
}
// Pretty returns a formatter for the key.
func (k InternalKey) Pretty(f FormatKey) fmt.Formatter {
return prettyInternalKey{k, f}
}
// IsExclusiveSentinel returns whether this internal key excludes point keys
// with the same user key if used as an end boundary. See the comment on
// InternalKeyRangeDeletionSentinel.
func (k InternalKey) IsExclusiveSentinel() bool {
switch kind := k.Kind(); kind {
case InternalKeyKindRangeDelete:
return k.Trailer == InternalKeyRangeDeleteSentinel
case InternalKeyKindRangeKeyDelete, InternalKeyKindRangeKeyUnset, InternalKeyKindRangeKeySet:
return (k.Trailer >> 8) == InternalKeySeqNumMax
default:
return false
}
}
type prettyInternalKey struct {
InternalKey
formatKey FormatKey
}
func (k prettyInternalKey) Format(s fmt.State, c rune) {
if seqNum := k.SeqNum(); seqNum == InternalKeySeqNumMax {
fmt.Fprintf(s, "%s#inf,%s", k.formatKey(k.UserKey), k.Kind())
} else {
fmt.Fprintf(s, "%s#%d,%s", k.formatKey(k.UserKey), k.SeqNum(), k.Kind())
}
}
// ParsePrettyInternalKey parses the pretty string representation of an
// internal key. The format is <user-key>#<seq-num>,<kind>.
func ParsePrettyInternalKey(s string) InternalKey {
x := strings.FieldsFunc(s, func(c rune) bool { return c == '#' || c == ',' })
ukey := x[0]
kind, ok := kindsMap[x[2]]
if !ok {
panic(fmt.Sprintf("unknown kind: %q", x[2]))
}
var seqNum uint64
if x[1] == "max" || x[1] == "inf" {
seqNum = InternalKeySeqNumMax
} else {
seqNum, _ = strconv.ParseUint(x[1], 10, 64)
}
return MakeInternalKey([]byte(ukey), seqNum, kind)
}

444
vendor/github.com/cockroachdb/pebble/internal/base/iterator.go generated vendored Normal file
View File

@@ -0,0 +1,444 @@
// Copyright 2019 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base
import (
"fmt"
"time"
"github.com/cockroachdb/pebble/internal/humanize"
"github.com/cockroachdb/redact"
)
// InternalIterator iterates over a DB's key/value pairs in key order. Unlike
// the Iterator interface, the returned keys are InternalKeys composed of the
// user-key, a sequence number and a key kind. In forward iteration, key/value
// pairs for identical user-keys are returned in descending sequence order. In
// reverse iteration, key/value pairs for identical user-keys are returned in
// ascending sequence order.
//
// InternalIterators provide 5 absolute positioning methods and 2 relative
// positioning methods. The absolute positioning methods are:
//
// - SeekGE
// - SeekPrefixGE
// - SeekLT
// - First
// - Last
//
// The relative positioning methods are:
//
// - Next
// - Prev
//
// The relative positioning methods can be used in conjunction with any of the
// absolute positioning methods with one exception: SeekPrefixGE does not
// support reverse iteration via Prev. It is undefined to call relative
// positioning methods without ever calling an absolute positioning method.
//
// InternalIterators can optionally implement a prefix iteration mode. This
// mode is entered by calling SeekPrefixGE and exited by any other absolute
// positioning method (SeekGE, SeekLT, First, Last). When in prefix iteration
// mode, a call to Next will advance to the next key which has the same
// "prefix" as the one supplied to SeekPrefixGE. Note that "prefix" in this
// context is not a strict byte prefix, but defined by byte equality for the
// result of the Comparer.Split method. An InternalIterator is not required to
// support prefix iteration mode, and can implement SeekPrefixGE by forwarding
// to SeekGE. When the iteration prefix is exhausted, it is not valid to call
// Next on an internal iterator that's already returned (nil,nilv) or a key
// beyond the prefix.
//
// Bounds, [lower, upper), can be set on iterators, either using the SetBounds()
// function in the interface, or in implementation specific ways during iterator
// creation. The forward positioning routines (SeekGE, First, and Next) only
// check the upper bound. The reverse positioning routines (SeekLT, Last, and
// Prev) only check the lower bound. It is up to the caller to ensure that the
// forward positioning routines respect the lower bound and the reverse
// positioning routines respect the upper bound (i.e. calling SeekGE instead of
// First if there is a lower bound, and SeekLT instead of Last if there is an
// upper bound). This imposition is done in order to elevate that enforcement to
// the caller (generally pebble.Iterator or pebble.mergingIter) rather than
// having it duplicated in every InternalIterator implementation.
//
// Additionally, the caller needs to ensure that SeekGE/SeekPrefixGE are not
// called with a key > the upper bound, and SeekLT is not called with a key <
// the lower bound. InternalIterator implementations are required to respect
// the iterator bounds, never returning records outside of the bounds with one
// exception: an iterator may generate synthetic RANGEDEL marker records. See
// levelIter.syntheticBoundary for the sole existing example of this behavior.
// Specifically, levelIter can return synthetic keys whose user key is equal to
// the lower/upper bound.
//
// The bounds provided to an internal iterator must remain valid until a
// subsequent call to SetBounds has returned. This requirement exists so that
// iterator implementations may compare old and new bounds to apply low-level
// optimizations. The pebble.Iterator satisfies this requirement by maintaining
// two bound buffers and switching between them.
//
// An iterator must be closed after use, but it is not necessary to read an
// iterator until exhaustion.
//
// An iterator is not goroutine-safe, but it is safe to use multiple iterators
// concurrently, either in separate goroutines or switching between the
// iterators in a single goroutine.
//
// It is also safe to use an iterator concurrently with modifying its
// underlying DB, if that DB permits modification. However, the resultant
// key/value pairs are not guaranteed to be a consistent snapshot of that DB
// at a particular point in time.
//
// InternalIterators accumulate errors encountered during operation, exposing
// them through the Error method. All of the absolute positioning methods
// reset any accumulated error before positioning. Relative positioning
// methods return without advancing if the iterator has accumulated an error.
//
// nilv == shorthand for LazyValue{}, which represents a nil value.
type InternalIterator interface {
// SeekGE moves the iterator to the first key/value pair whose key is greater
// than or equal to the given key. Returns the key and value if the iterator
// is pointing at a valid entry, and (nil, nilv) otherwise. Note that SeekGE
// only checks the upper bound. It is up to the caller to ensure that key
// is greater than or equal to the lower bound.
SeekGE(key []byte, flags SeekGEFlags) (*InternalKey, LazyValue)
// SeekPrefixGE moves the iterator to the first key/value pair whose key is
// greater than or equal to the given key. Returns the key and value if the
// iterator is pointing at a valid entry, and (nil, nilv) otherwise. Note that
// SeekPrefixGE only checks the upper bound. It is up to the caller to ensure
// that key is greater than or equal to the lower bound.
//
// The prefix argument is used by some InternalIterator implementations (e.g.
// sstable.Reader) to avoid expensive operations. A user-defined Split
// function must be supplied to the Comparer for the DB. The supplied prefix
// will be the prefix of the given key returned by that Split function. If
// the iterator is able to determine that no key with the prefix exists, it
// can return (nil,nilv). Unlike SeekGE, this is not an indication that
// iteration is exhausted.
//
// Note that the iterator may return keys not matching the prefix. It is up
// to the caller to check if the prefix matches.
//
// Calling SeekPrefixGE places the receiver into prefix iteration mode. Once
// in this mode, reverse iteration may not be supported and will return an
// error. Note that pebble/Iterator.SeekPrefixGE has this same restriction on
// not supporting reverse iteration in prefix iteration mode until a
// different positioning routine (SeekGE, SeekLT, First or Last) switches the
// iterator out of prefix iteration.
SeekPrefixGE(prefix, key []byte, flags SeekGEFlags) (*InternalKey, LazyValue)
// SeekLT moves the iterator to the last key/value pair whose key is less
// than the given key. Returns the key and value if the iterator is pointing
// at a valid entry, and (nil, nilv) otherwise. Note that SeekLT only checks
// the lower bound. It is up to the caller to ensure that key is less than
// the upper bound.
SeekLT(key []byte, flags SeekLTFlags) (*InternalKey, LazyValue)
// First moves the iterator the the first key/value pair. Returns the key and
// value if the iterator is pointing at a valid entry, and (nil, nilv)
// otherwise. Note that First only checks the upper bound. It is up to the
// caller to ensure that First() is not called when there is a lower bound,
// and instead call SeekGE(lower).
First() (*InternalKey, LazyValue)
// Last moves the iterator the the last key/value pair. Returns the key and
// value if the iterator is pointing at a valid entry, and (nil, nilv)
// otherwise. Note that Last only checks the lower bound. It is up to the
// caller to ensure that Last() is not called when there is an upper bound,
// and instead call SeekLT(upper).
Last() (*InternalKey, LazyValue)
// Next moves the iterator to the next key/value pair. Returns the key and
// value if the iterator is pointing at a valid entry, and (nil, nilv)
// otherwise. Note that Next only checks the upper bound. It is up to the
// caller to ensure that key is greater than or equal to the lower bound.
//
// It is valid to call Next when the iterator is positioned before the first
// key/value pair due to either a prior call to SeekLT or Prev which returned
// (nil, nilv). It is not allowed to call Next when the previous call to SeekGE,
// SeekPrefixGE or Next returned (nil, nilv).
Next() (*InternalKey, LazyValue)
// NextPrefix moves the iterator to the next key/value pair with a different
// prefix than the key at the current iterator position. Returns the key and
// value if the iterator is pointing at a valid entry, and (nil, nil)
// otherwise. Note that NextPrefix only checks the upper bound. It is up to
// the caller to ensure that key is greater than or equal to the lower
// bound.
//
// NextPrefix is passed the immediate successor to the current prefix key. A
// valid implementation of NextPrefix is to call SeekGE with succKey.
//
// It is not allowed to call NextPrefix when the previous call was a reverse
// positioning operation or a call to a forward positioning method that
// returned (nil, nilv). It is also not allowed to call NextPrefix when the
// iterator is in prefix iteration mode.
NextPrefix(succKey []byte) (*InternalKey, LazyValue)
// Prev moves the iterator to the previous key/value pair. Returns the key
// and value if the iterator is pointing at a valid entry, and (nil, nilv)
// otherwise. Note that Prev only checks the lower bound. It is up to the
// caller to ensure that key is less than the upper bound.
//
// It is valid to call Prev when the iterator is positioned after the last
// key/value pair due to either a prior call to SeekGE or Next which returned
// (nil, nilv). It is not allowed to call Prev when the previous call to SeekLT
// or Prev returned (nil, nilv).
Prev() (*InternalKey, LazyValue)
// Error returns any accumulated error. It may not include errors returned
// to the client when calling LazyValue.Value().
Error() error
// Close closes the iterator and returns any accumulated error. Exhausting
// all the key/value pairs in a table is not considered to be an error.
// It is valid to call Close multiple times. Other methods should not be
// called after the iterator has been closed.
Close() error
// SetBounds sets the lower and upper bounds for the iterator. Note that the
// result of Next and Prev will be undefined until the iterator has been
// repositioned with SeekGE, SeekPrefixGE, SeekLT, First, or Last.
//
// The bounds provided must remain valid until a subsequent call to
// SetBounds has returned. This requirement exists so that iterator
// implementations may compare old and new bounds to apply low-level
// optimizations.
SetBounds(lower, upper []byte)
fmt.Stringer
}
// SeekGEFlags holds flags that may configure the behavior of a forward seek.
// Not all flags are relevant to all iterators.
type SeekGEFlags uint8
const (
seekGEFlagTrySeekUsingNext uint8 = iota
seekGEFlagRelativeSeek
seekGEFlagBatchJustRefreshed
)
// SeekGEFlagsNone is the default value of SeekGEFlags, with all flags disabled.
const SeekGEFlagsNone = SeekGEFlags(0)
// TrySeekUsingNext indicates whether a performance optimization was enabled
// by a caller, indicating the caller has not done any action to move this
// iterator beyond the first key that would be found if this iterator were to
// honestly do the intended seek. For example, say the caller did a
// SeekGE(k1...), followed by SeekGE(k2...) where k1 <= k2, without any
// intermediate positioning calls. The caller can safely specify true for this
// parameter in the second call. As another example, say the caller did do one
// call to Next between the two Seek calls, and k1 < k2. Again, the caller can
// safely specify a true value for this parameter. Note that a false value is
// always safe. The callee is free to ignore the true value if its
// implementation does not permit this optimization.
//
// We make the caller do this determination since a string comparison of k1, k2
// is not necessarily cheap, and there may be many iterators in the iterator
// stack. Doing it once at the root of the iterator stack is cheaper.
//
// This optimization could also be applied to SeekLT (where it would be
// trySeekUsingPrev). We currently only do it for SeekPrefixGE and SeekGE
// because this is where this optimization helps the performance of CockroachDB.
// The SeekLT cases in CockroachDB are typically accompanied with bounds that
// change between seek calls, and is optimized inside certain iterator
// implementations, like singleLevelIterator, without any extra parameter
// passing (though the same amortization of string comparisons could be done to
// improve that optimization, by making the root of the iterator stack do it).
func (s SeekGEFlags) TrySeekUsingNext() bool { return (s & (1 << seekGEFlagTrySeekUsingNext)) != 0 }
// RelativeSeek is set when in the course of a forward positioning operation, a
// higher-level iterator seeks a lower-level iterator to a larger key than the
// one at the current iterator position.
//
// Concretely, this occurs when the merging iterator observes a range deletion
// covering the key at a level's current position, and the merging iterator
// seeks the level to the range deletion's end key. During lazy-combined
// iteration, this flag signals to the level iterator that the seek is NOT an
// absolute-positioning operation from the perspective of the pebble.Iterator,
// and the level iterator must look for range keys in tables between the current
// iterator position and the new seeked position.
func (s SeekGEFlags) RelativeSeek() bool { return (s & (1 << seekGEFlagRelativeSeek)) != 0 }
// BatchJustRefreshed is set by Seek[Prefix]GE when an iterator's view of an
// indexed batch was just refreshed. It serves as a signal to the batch iterator
// to ignore the TrySeekUsingNext optimization, because the external knowledge
// imparted by the TrySeekUsingNext flag does not apply to the batch iterator's
// position. See (pebble.Iterator).batchJustRefreshed.
func (s SeekGEFlags) BatchJustRefreshed() bool { return (s & (1 << seekGEFlagBatchJustRefreshed)) != 0 }
// EnableTrySeekUsingNext returns the provided flags with the
// try-seek-using-next optimization enabled. See TrySeekUsingNext for an
// explanation of this optimization.
func (s SeekGEFlags) EnableTrySeekUsingNext() SeekGEFlags {
return s | (1 << seekGEFlagTrySeekUsingNext)
}
// DisableTrySeekUsingNext returns the provided flags with the
// try-seek-using-next optimization disabled.
func (s SeekGEFlags) DisableTrySeekUsingNext() SeekGEFlags {
return s &^ (1 << seekGEFlagTrySeekUsingNext)
}
// EnableRelativeSeek returns the provided flags with the relative-seek flag
// enabled. See RelativeSeek for an explanation of this flag's use.
func (s SeekGEFlags) EnableRelativeSeek() SeekGEFlags {
return s | (1 << seekGEFlagRelativeSeek)
}
// DisableRelativeSeek returns the provided flags with the relative-seek flag
// disabled.
func (s SeekGEFlags) DisableRelativeSeek() SeekGEFlags {
return s &^ (1 << seekGEFlagRelativeSeek)
}
// EnableBatchJustRefreshed returns the provided flags with the
// batch-just-refreshed bit set. See BatchJustRefreshed for an explanation of
// this flag.
func (s SeekGEFlags) EnableBatchJustRefreshed() SeekGEFlags {
return s | (1 << seekGEFlagBatchJustRefreshed)
}
// DisableBatchJustRefreshed returns the provided flags with the
// batch-just-refreshed bit unset.
func (s SeekGEFlags) DisableBatchJustRefreshed() SeekGEFlags {
return s &^ (1 << seekGEFlagBatchJustRefreshed)
}
// SeekLTFlags holds flags that may configure the behavior of a reverse seek.
// Not all flags are relevant to all iterators.
type SeekLTFlags uint8
const (
seekLTFlagRelativeSeek uint8 = iota
)
// SeekLTFlagsNone is the default value of SeekLTFlags, with all flags disabled.
const SeekLTFlagsNone = SeekLTFlags(0)
// RelativeSeek is set when in the course of a reverse positioning operation, a
// higher-level iterator seeks a lower-level iterator to a smaller key than the
// one at the current iterator position.
//
// Concretely, this occurs when the merging iterator observes a range deletion
// covering the key at a level's current position, and the merging iterator
// seeks the level to the range deletion's start key. During lazy-combined
// iteration, this flag signals to the level iterator that the seek is NOT an
// absolute-positioning operation from the perspective of the pebble.Iterator,
// and the level iterator must look for range keys in tables between the current
// iterator position and the new seeked position.
func (s SeekLTFlags) RelativeSeek() bool { return s&(1<<seekLTFlagRelativeSeek) != 0 }
// EnableRelativeSeek returns the provided flags with the relative-seek flag
// enabled. See RelativeSeek for an explanation of this flag's use.
func (s SeekLTFlags) EnableRelativeSeek() SeekLTFlags {
return s | (1 << seekLTFlagRelativeSeek)
}
// DisableRelativeSeek returns the provided flags with the relative-seek flag
// disabled.
func (s SeekLTFlags) DisableRelativeSeek() SeekLTFlags {
return s &^ (1 << seekLTFlagRelativeSeek)
}
// InternalIteratorStats contains miscellaneous stats produced by
// InternalIterators that are part of the InternalIterator tree. Not every
// field is relevant for an InternalIterator implementation. The field values
// are aggregated as one goes up the InternalIterator tree.
type InternalIteratorStats struct {
// Bytes in the loaded blocks. If the block was compressed, this is the
// compressed bytes. Currently, only the index blocks, data blocks
// containing points, and filter blocks are included.
BlockBytes uint64
// Subset of BlockBytes that were in the block cache.
BlockBytesInCache uint64
// BlockReadDuration accumulates the duration spent fetching blocks
// due to block cache misses.
// TODO(sumeer): this currently excludes the time spent in Reader creation,
// and in reading the rangedel and rangekey blocks. Fix that.
BlockReadDuration time.Duration
// The following can repeatedly count the same points if they are iterated
// over multiple times. Additionally, they may count a point twice when
// switching directions. The latter could be improved if needed.
// Bytes in keys that were iterated over. Currently, only point keys are
// included.
KeyBytes uint64
// Bytes in values that were iterated over. Currently, only point values are
// included. For separated values, this is the size of the handle.
ValueBytes uint64
// The count of points iterated over.
PointCount uint64
// Points that were iterated over that were covered by range tombstones. It
// can be useful for discovering instances of
// https://github.com/cockroachdb/pebble/issues/1070.
PointsCoveredByRangeTombstones uint64
// Stats related to points in value blocks encountered during iteration.
// These are useful to understand outliers, since typical user facing
// iteration should tend to only look at the latest point, and hence have
// the following stats close to 0.
SeparatedPointValue struct {
// Count is a count of points that were in value blocks. This is not a
// subset of PointCount: PointCount is produced by mergingIter and if
// positioned once, and successful in returning a point, will have a
// PointCount of 1, regardless of how many sstables (and memtables etc.)
// in the heap got positioned. The count here includes every sstable
// iterator that got positioned in the heap.
Count uint64
// ValueBytes represent the total byte length of the values (in value
// blocks) of the points corresponding to Count.
ValueBytes uint64
// ValueBytesFetched is the total byte length of the values (in value
// blocks) that were retrieved.
ValueBytesFetched uint64
}
}
// Merge merges the stats in from into the given stats.
func (s *InternalIteratorStats) Merge(from InternalIteratorStats) {
s.BlockBytes += from.BlockBytes
s.BlockBytesInCache += from.BlockBytesInCache
s.BlockReadDuration += from.BlockReadDuration
s.KeyBytes += from.KeyBytes
s.ValueBytes += from.ValueBytes
s.PointCount += from.PointCount
s.PointsCoveredByRangeTombstones += from.PointsCoveredByRangeTombstones
s.SeparatedPointValue.Count += from.SeparatedPointValue.Count
s.SeparatedPointValue.ValueBytes += from.SeparatedPointValue.ValueBytes
s.SeparatedPointValue.ValueBytesFetched += from.SeparatedPointValue.ValueBytesFetched
}
func (s *InternalIteratorStats) String() string {
return redact.StringWithoutMarkers(s)
}
// SafeFormat implements the redact.SafeFormatter interface.
func (s *InternalIteratorStats) SafeFormat(p redact.SafePrinter, verb rune) {
p.Printf("blocks: %s cached",
humanize.Bytes.Uint64(s.BlockBytesInCache),
)
if s.BlockBytes != s.BlockBytesInCache || s.BlockReadDuration != 0 {
p.Printf(", %s not cached (read time: %s)",
humanize.Bytes.Uint64(s.BlockBytes-s.BlockBytesInCache),
humanize.FormattedString(s.BlockReadDuration.String()),
)
}
p.Printf("; points: %s", humanize.Count.Uint64(s.PointCount))
if s.PointsCoveredByRangeTombstones != 0 {
p.Printf("(%s tombstoned)", humanize.Count.Uint64(s.PointsCoveredByRangeTombstones))
}
p.Printf(" (%s keys, %s values)",
humanize.Bytes.Uint64(s.KeyBytes),
humanize.Bytes.Uint64(s.ValueBytes),
)
if s.SeparatedPointValue.Count != 0 {
p.Printf("; separated: %s (%s, %s fetched)",
humanize.Count.Uint64(s.SeparatedPointValue.Count),
humanize.Bytes.Uint64(s.SeparatedPointValue.ValueBytes),
humanize.Bytes.Uint64(s.SeparatedPointValue.ValueBytesFetched))
}
}

287
vendor/github.com/cockroachdb/pebble/internal/base/lazy_value.go generated vendored Normal file
View File

@@ -0,0 +1,287 @@
// Copyright 2022 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base
import "github.com/cockroachdb/pebble/internal/invariants"
// A value can have user-defined attributes that are a function of the value
// byte slice. For now, we only support "short attributes", which can be
// encoded in 3 bits. We will likely extend this to "long attributes" later
// for values that are even more expensive to access than those in value
// blocks in the same sstable.
//
// When a sstable writer chooses not to store a value together with the key,
// it can call the ShortAttributeExtractor to extract the attribute and store
// it together with the key. This allows for cheap retrieval of
// AttributeAndLen on the read-path, without doing a more expensive retrieval
// of the value. In general, the extraction code may want to also look at the
// key to decide how to treat the value, hence the key* parameters.
//
// Write path performance: The ShortAttributeExtractor func cannot be inlined,
// so we will pay the cost of this function call. However, we will only pay
// this when (a) the value is not being stored together with the key, and (b)
// the key-value pair is being initially written to the DB, or a compaction is
// transitioning the key-value pair from being stored together to being stored
// separately.
// ShortAttribute encodes a user-specified attribute of the value.
type ShortAttribute uint8
// MaxShortAttribute is the maximum value of the short attribute (3 bits).
const MaxShortAttribute = 7
// ShortAttributeExtractor is an extractor that given the value, will return
// the ShortAttribute.
type ShortAttributeExtractor func(
key []byte, keyPrefixLen int, value []byte) (ShortAttribute, error)
// AttributeAndLen represents the pair of value length and the short
// attribute.
type AttributeAndLen struct {
ValueLen int32
ShortAttribute ShortAttribute
}
// LazyValue represents a value that may not already have been extracted.
// Currently, it can represent either an in-place value (stored with the key)
// or a value stored in the value section. However, the interface is general
// enough to support values that are stored in separate files.
//
// LazyValue is used in the InternalIterator interface, such that all
// positioning calls return (*InternalKey, LazyValue). It is also exposed via
// the public Iterator for callers that need to remember a recent but not
// necessarily latest LazyValue, in case they need the actual value in the
// future. An example is a caller that is iterating in reverse and looking for
// the latest MVCC version for a key -- it cannot identify the latest MVCC
// version without stepping to the previous key-value pair e.g.
// storage.pebbleMVCCScanner in CockroachDB.
//
// Performance note: It is important for this struct to not exceed a sizeof 32
// bytes, for optimizing the common case of the in-place value. Prior to
// introducing LazyValue, we were passing around a []byte which is 24 bytes.
// Passing a 40 byte or larger struct causes performance to drop by 75% on
// some benchmarks that do tight iteration loops.
//
// Memory management:
// This is subtle, but important for performance.
//
// A LazyValue returned by an InternalIterator or Iterator is unstable in that
// repositioning the iterator will invalidate the memory inside it. A caller
// wishing to maintain that LazyValue needs to call LazyValue.Clone(). Note
// that this does not fetch the value if it is not in-place. Clone() should
// ideally not be called if LazyValue.Value() has been called, since the
// cloned LazyValue will forget the extracted/fetched value, and calling
// Value() on this clone will cause the value to be extracted again. That is,
// Clone() does not make any promise about the memory stability of the
// underlying value.
//
// A user of an iterator that calls LazyValue.Value() wants as much as
// possible for the returned value []byte to point to iterator owned memory.
//
// 1. [P1] The underlying iterator that owns that memory also needs a promise
// from that user that at any time there is at most one value []byte slice
// that the caller is expecting it to maintain. Otherwise, the underlying
// iterator has to maintain multiple such []byte slices which results in
// more complicated and inefficient code.
//
// 2. [P2] The underlying iterator, in order to make the promise that it is
// maintaining the one value []byte slice, also needs a way to know when
// it is relieved of that promise. One way it is relieved of that promise
// is by being told that it is being repositioned. Typically, the owner of
// the value []byte slice is a sstable iterator, and it will know that it
// is relieved of the promise when it is repositioned. However, consider
// the case where the caller has used LazyValue.Clone() and repositioned
// the iterator (which is actually a tree of iterators). In this case the
// underlying sstable iterator may not even be open. LazyValue.Value()
// will still work (at a higher cost), but since the sstable iterator is
// not open, it does not have a mechanism to know when the retrieved value
// is no longer in use. We refer to this situation as "not satisfying P2".
// To handle this situation, the LazyValue.Value() method accepts a caller
// owned buffer, that the callee will use if needed. The callee explicitly
// tells the caller whether the []byte slice for the value is now owned by
// the caller. This will be true if the callee attempted to use buf and
// either successfully used it or allocated a new []byte slice.
//
// To ground the above in reality, we consider three examples of callers of
// LazyValue.Value():
//
// - Iterator: it calls LazyValue.Value for its own use when merging values.
// When merging during reverse iteration, it may have cloned the LazyValue.
// In this case it calls LazyValue.Value() on the cloned value, merges it,
// and then calls LazyValue.Value() on the current iterator position and
// merges it. So it is honoring P1.
//
// - Iterator on behalf of Iterator clients: The Iterator.Value() method
// needs to call LazyValue.Value(). The client of Iterator is satisfying P1
// because of the inherent Iterator interface constraint, i.e., it is calling
// Iterator.Value() on the current Iterator position. It is possible that
// the Iterator has cloned this LazyValue (for the reverse iteration case),
// which the client is unaware of, so the underlying sstable iterator may
// not be able to satisfy P2. This is ok because Iterator will call
// LazyValue.Value with its (reusable) owned buffer.
//
// - CockroachDB's pebbleMVCCScanner: This will use LazyValues from Iterator
// since during reverse iteration in order to find the highest version that
// satisfies a read it needs to clone the LazyValue, step back the iterator
// and then decide whether it needs the value from the previously cloned
// LazyValue. The pebbleMVCCScanner will satisfy P1. The P2 story is
// similar to the previous case in that it will call LazyValue.Value with
// its (reusable) owned buffer.
//
// Corollary: callers that directly use InternalIterator can know that they
// have done nothing to interfere with promise P2 can pass in a nil buf and be
// sure that it will not trigger an allocation.
//
// Repeated calling of LazyValue.Value:
// This is ok as long as the caller continues to satisfy P1. The previously
// fetched value will be remembered inside LazyValue to avoid fetching again.
// So if the caller's buffer is used the first time the value was fetched, it
// is still in use.
//
// LazyValue fields are visible outside the package for use in
// InternalIterator implementations and in Iterator, but not meant for direct
// use by users of Pebble.
type LazyValue struct {
// ValueOrHandle represents a value, or a handle to be passed to ValueFetcher.
// - Fetcher == nil: ValueOrHandle is a value.
// - Fetcher != nil: ValueOrHandle is a handle and Fetcher.Attribute is
// initialized.
// The ValueOrHandle exposed by InternalIterator or Iterator may not be stable
// if the iterator is stepped. To make it stable, make a copy using Clone.
ValueOrHandle []byte
// Fetcher provides support for fetching an actually lazy value.
Fetcher *LazyFetcher
}
// LazyFetcher supports fetching a lazy value.
//
// Fetcher and Attribute are to be initialized at creation time. The fields
// are arranged to reduce the sizeof this struct.
type LazyFetcher struct {
// Fetcher, given a handle, returns the value.
Fetcher ValueFetcher
err error
value []byte
// Attribute includes the short attribute and value length.
Attribute AttributeAndLen
fetched bool
callerOwned bool
}
// ValueFetcher is an interface for fetching a value.
type ValueFetcher interface {
// Fetch returns the value, given the handle. It is acceptable to call the
// ValueFetcher.Fetch as long as the DB is open. However, one should assume
// there is a fast-path when the iterator tree has not moved off the sstable
// iterator that initially provided this LazyValue. Hence, to utilize this
// fast-path the caller should try to decide whether it needs the value or
// not as soon as possible, with minimal possible stepping of the iterator.
//
// buf will be used if the fetcher cannot satisfy P2 (see earlier comment).
// If the fetcher attempted to use buf *and* len(buf) was insufficient, it
// will allocate a new slice for the value. In either case it will set
// callerOwned to true.
Fetch(
handle []byte, valLen int32, buf []byte) (val []byte, callerOwned bool, err error)
}
// Value returns the underlying value.
func (lv *LazyValue) Value(buf []byte) (val []byte, callerOwned bool, err error) {
if lv.Fetcher == nil {
return lv.ValueOrHandle, false, nil
}
// Do the rest of the work in a separate method to attempt mid-stack
// inlining of Value(). Unfortunately, this still does not inline since the
// cost of 85 exceeds the budget of 80.
//
// TODO(sumeer): Packing the return values into a struct{[]byte error bool}
// causes it to be below the budget. Consider this if we need to recover
// more performance. I suspect that inlining this only matters in
// micro-benchmarks, and in actual use cases in CockroachDB it will not
// matter because there is substantial work done with a fetched value.
return lv.fetchValue(buf)
}
// INVARIANT: lv.Fetcher != nil
func (lv *LazyValue) fetchValue(buf []byte) (val []byte, callerOwned bool, err error) {
f := lv.Fetcher
if !f.fetched {
f.fetched = true
f.value, f.callerOwned, f.err = f.Fetcher.Fetch(
lv.ValueOrHandle, lv.Fetcher.Attribute.ValueLen, buf)
}
return f.value, f.callerOwned, f.err
}
// InPlaceValue returns the value under the assumption that it is in-place.
// This is for Pebble-internal code.
func (lv *LazyValue) InPlaceValue() []byte {
if invariants.Enabled && lv.Fetcher != nil {
panic("value must be in-place")
}
return lv.ValueOrHandle
}
// Len returns the length of the value.
func (lv *LazyValue) Len() int {
if lv.Fetcher == nil {
return len(lv.ValueOrHandle)
}
return int(lv.Fetcher.Attribute.ValueLen)
}
// TryGetShortAttribute returns the ShortAttribute and a bool indicating
// whether the ShortAttribute was populated.
func (lv *LazyValue) TryGetShortAttribute() (ShortAttribute, bool) {
if lv.Fetcher == nil {
return 0, false
}
return lv.Fetcher.Attribute.ShortAttribute, true
}
// Clone creates a stable copy of the LazyValue, by appending bytes to buf.
// The fetcher parameter must be non-nil and may be over-written and used
// inside the returned LazyValue -- this is needed to avoid an allocation.
// Most callers have at most K cloned LazyValues, where K is hard-coded, so
// they can have a pool of exactly K LazyFetcher structs they can reuse in
// these calls. The alternative of allocating LazyFetchers from a sync.Pool is
// not viable since we have no code trigger for returning to the pool
// (LazyValues are simply GC'd).
//
// NB: It is highly preferable that LazyValue.Value() has not been called,
// since the Clone will forget any previously extracted value, and a future
// call to Value will cause it to be fetched again. We do this since we don't
// want to reason about whether or not to clone an already extracted value
// inside the Fetcher (we don't). Property P1 applies here too: if lv1.Value()
// has been called, and then lv2 is created as a clone of lv1, then calling
// lv2.Value() can invalidate any backing memory maintained inside the fetcher
// for lv1 (even though these are the same values). We initially prohibited
// calling LazyValue.Clone() if LazyValue.Value() has been called, but there
// is at least one complex caller (pebbleMVCCScanner inside CockroachDB) where
// it is not easy to prove this invariant.
func (lv *LazyValue) Clone(buf []byte, fetcher *LazyFetcher) (LazyValue, []byte) {
var lvCopy LazyValue
if lv.Fetcher != nil {
*fetcher = LazyFetcher{
Fetcher: lv.Fetcher.Fetcher,
Attribute: lv.Fetcher.Attribute,
// Not copying anything that has been extracted.
}
lvCopy.Fetcher = fetcher
}
vLen := len(lv.ValueOrHandle)
if vLen == 0 {
return lvCopy, buf
}
bufLen := len(buf)
buf = append(buf, lv.ValueOrHandle...)
lvCopy.ValueOrHandle = buf[bufLen : bufLen+vLen]
return lvCopy, buf
}
// MakeInPlaceValue constructs an in-place value.
func MakeInPlaceValue(val []byte) LazyValue {
return LazyValue{ValueOrHandle: val}
}

144
vendor/github.com/cockroachdb/pebble/internal/base/logger.go generated vendored Normal file
View File

@@ -0,0 +1,144 @@
// Copyright 2023 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base
import (
"bytes"
"context"
"fmt"
"log"
"os"
"runtime"
"sync"
"github.com/cockroachdb/pebble/internal/invariants"
)
// Logger defines an interface for writing log messages.
type Logger interface {
Infof(format string, args ...interface{})
Fatalf(format string, args ...interface{})
}
type defaultLogger struct{}
// DefaultLogger logs to the Go stdlib logs.
var DefaultLogger defaultLogger
var _ Logger = DefaultLogger
// Infof implements the Logger.Infof interface.
func (defaultLogger) Infof(format string, args ...interface{}) {
_ = log.Output(2, fmt.Sprintf(format, args...))
}
// Fatalf implements the Logger.Fatalf interface.
func (defaultLogger) Fatalf(format string, args ...interface{}) {
_ = log.Output(2, fmt.Sprintf(format, args...))
os.Exit(1)
}
// InMemLogger implements Logger using an in-memory buffer (used for testing).
// The buffer can be read via String() and cleared via Reset().
type InMemLogger struct {
mu struct {
sync.Mutex
buf bytes.Buffer
}
}
var _ Logger = (*InMemLogger)(nil)
// Reset clears the internal buffer.
func (b *InMemLogger) Reset() {
b.mu.Lock()
defer b.mu.Unlock()
b.mu.buf.Reset()
}
// String returns the current internal buffer.
func (b *InMemLogger) String() string {
b.mu.Lock()
defer b.mu.Unlock()
return b.mu.buf.String()
}
// Infof is part of the Logger interface.
func (b *InMemLogger) Infof(format string, args ...interface{}) {
s := fmt.Sprintf(format, args...)
b.mu.Lock()
defer b.mu.Unlock()
b.mu.buf.Write([]byte(s))
if n := len(s); n == 0 || s[n-1] != '\n' {
b.mu.buf.Write([]byte("\n"))
}
}
// Fatalf is part of the Logger interface.
func (b *InMemLogger) Fatalf(format string, args ...interface{}) {
b.Infof(format, args...)
runtime.Goexit()
}
// LoggerAndTracer defines an interface for logging and tracing.
type LoggerAndTracer interface {
Logger
// Eventf formats and emits a tracing log, if tracing is enabled in the
// current context.
Eventf(ctx context.Context, format string, args ...interface{})
// IsTracingEnabled returns true if tracing is enabled. It can be used as an
// optimization to avoid calling Eventf (which will be a noop when tracing
// is not enabled) to avoid the overhead of boxing the args.
IsTracingEnabled(ctx context.Context) bool
}
// LoggerWithNoopTracer wraps a logger and does no tracing.
type LoggerWithNoopTracer struct {
Logger
}
var _ LoggerAndTracer = &LoggerWithNoopTracer{}
// Eventf implements LoggerAndTracer.
func (*LoggerWithNoopTracer) Eventf(ctx context.Context, format string, args ...interface{}) {
if invariants.Enabled && ctx == nil {
panic("Eventf context is nil")
}
}
// IsTracingEnabled implements LoggerAndTracer.
func (*LoggerWithNoopTracer) IsTracingEnabled(ctx context.Context) bool {
if invariants.Enabled && ctx == nil {
panic("IsTracingEnabled ctx is nil")
}
return false
}
// NoopLoggerAndTracer does no logging and tracing. Remember that struct{} is
// special cased in Go and does not incur an allocation when it backs the
// interface LoggerAndTracer.
type NoopLoggerAndTracer struct{}
var _ LoggerAndTracer = NoopLoggerAndTracer{}
// Infof implements LoggerAndTracer.
func (l NoopLoggerAndTracer) Infof(format string, args ...interface{}) {}
// Fatalf implements LoggerAndTracer.
func (l NoopLoggerAndTracer) Fatalf(format string, args ...interface{}) {}
// Eventf implements LoggerAndTracer.
func (l NoopLoggerAndTracer) Eventf(ctx context.Context, format string, args ...interface{}) {
if invariants.Enabled && ctx == nil {
panic("Eventf context is nil")
}
}
// IsTracingEnabled implements LoggerAndTracer.
func (l NoopLoggerAndTracer) IsTracingEnabled(ctx context.Context) bool {
if invariants.Enabled && ctx == nil {
panic("IsTracingEnabled ctx is nil")
}
return false
}

133
vendor/github.com/cockroachdb/pebble/internal/base/merger.go generated vendored Normal file
View File

@@ -0,0 +1,133 @@
// Copyright 2018 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base
import "io"
// Merge creates a ValueMerger for the specified key initialized with the value
// of one merge operand.
type Merge func(key, value []byte) (ValueMerger, error)
// ValueMerger receives merge operands one by one. The operand received is either
// newer or older than all operands received so far as indicated by the function
// names, `MergeNewer()` and `MergeOlder()`. Once all operands have been received,
// the client will invoke `Finish()` to obtain the final result. The order of
// a merge is not changed after the first call to `MergeNewer()` or
// `MergeOlder()`, i.e. the same method is used to submit all operands.
//
// The implementation may choose to merge values into the result immediately upon
// receiving each operand, or buffer operands until Finish() is called. For example,
// buffering may be useful to avoid (de)serializing partial merge results.
//
// The merge operation must be associative. That is, for the values A, B, C:
//
// Merge(A).MergeOlder(B).MergeOlder(C) == Merge(C).MergeNewer(B).MergeNewer(A)
//
// Examples of merge operators are integer addition, list append, and string
// concatenation.
type ValueMerger interface {
// MergeNewer adds an operand that is newer than all existing operands.
// The caller retains ownership of value.
//
// If an error is returned the merge is aborted and no other methods must
// be called.
MergeNewer(value []byte) error
// MergeOlder adds an operand that is older than all existing operands.
// The caller retains ownership of value.
//
// If an error is returned the merge is aborted and no other methods must
// be called.
MergeOlder(value []byte) error
// Finish does any final processing of the added operands and returns a
// result. The caller can assume the returned byte slice will not be mutated.
//
// Finish must be the last function called on the ValueMerger. The caller
// must not call any other ValueMerger functions after calling Finish.
//
// If `includesBase` is true, the oldest merge operand was part of the
// merge. This will always be the true during normal iteration, but may be
// false during compaction when only a subset of operands may be
// available. Note that `includesBase` is set to true conservatively: a false
// value means that we could not definitely determine that the base merge
// operand was included.
//
// If a Closer is returned, the returned slice will remain valid until it is
// closed. The caller must arrange for the closer to be eventually closed.
Finish(includesBase bool) ([]byte, io.Closer, error)
}
// DeletableValueMerger is an extension to ValueMerger which allows indicating that the
// result of a merge operation is non-existent. Such non-existent entries will eventually
// be deleted during compaction. Note that during compaction, non-existence of the result
// of a merge means that the merge operands will not result in any record being output.
// This is not the same as transforming the merge operands into a deletion tombstone, as
// older merge operands will still be visible during iteration. Deletion of the merge operands
// in this way is akin to the way a SingleDelete+Set combine into non-existence while leaving
// older records for the same key unaffected.
type DeletableValueMerger interface {
ValueMerger
// DeletableFinish enables a value merger to indicate that the result of a merge operation
// is non-existent. See Finish for a description of includesBase.
DeletableFinish(includesBase bool) (value []byte, delete bool, closer io.Closer, err error)
}
// Merger defines an associative merge operation. The merge operation merges
// two or more values for a single key. A merge operation is requested by
// writing a value using {Batch,DB}.Merge(). The value at that key is merged
// with any existing value. It is valid to Set a value at a key and then Merge
// a new value. Similar to non-merged values, a merged value can be deleted by
// either Delete or DeleteRange.
//
// The merge operation is invoked when a merge value is encountered during a
// read, either during a compaction or during iteration.
type Merger struct {
Merge Merge
// Name is the name of the merger.
//
// Pebble stores the merger name on disk, and opening a database with a
// different merger from the one it was created with will result in an error.
Name string
}
// AppendValueMerger concatenates merge operands in order from oldest to newest.
type AppendValueMerger struct {
buf []byte
}
// MergeNewer appends value to the result.
func (a *AppendValueMerger) MergeNewer(value []byte) error {
a.buf = append(a.buf, value...)
return nil
}
// MergeOlder prepends value to the result, which involves allocating a new buffer.
func (a *AppendValueMerger) MergeOlder(value []byte) error {
buf := make([]byte, len(a.buf)+len(value))
copy(buf, value)
copy(buf[len(value):], a.buf)
a.buf = buf
return nil
}
// Finish returns the buffer that was constructed on-demand in `Merge{OlderNewer}()` calls.
func (a *AppendValueMerger) Finish(includesBase bool) ([]byte, io.Closer, error) {
return a.buf, nil, nil
}
// DefaultMerger is the default implementation of the Merger interface. It
// concatenates the two values to merge.
var DefaultMerger = &Merger{
Merge: func(key, value []byte) (ValueMerger, error) {
res := &AppendValueMerger{}
res.buf = append(res.buf, value...)
return res, nil
},
Name: "pebble.concatenate",
}

98
vendor/github.com/cockroachdb/pebble/internal/base/metrics.go generated vendored Normal file
View File

@@ -0,0 +1,98 @@
// Copyright 2022 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base
import "time"
// ThroughputMetric is used to measure the byte throughput of some component
// that performs work in a single-threaded manner. The throughput can be
// approximated by Bytes/(WorkDuration+IdleTime). The idle time is represented
// separately, so that the user of this metric could approximate the peak
// throughput as Bytes/WorkTime. The metric is designed to be cumulative (see
// Merge).
type ThroughputMetric struct {
// Bytes is the processes bytes by the component.
Bytes int64
// WorkDuration is the duration that the component spent doing work.
WorkDuration time.Duration
// IdleDuration is the duration that the component was idling, waiting for
// work.
IdleDuration time.Duration
}
// Merge accumulates the information from another throughput metric.
func (tm *ThroughputMetric) Merge(x ThroughputMetric) {
tm.Bytes += x.Bytes
tm.WorkDuration += x.WorkDuration
tm.IdleDuration += x.IdleDuration
}
// Subtract subtracts the information from another ThroughputMetric
func (tm *ThroughputMetric) Subtract(x ThroughputMetric) {
tm.Bytes -= x.Bytes
tm.WorkDuration -= x.WorkDuration
tm.IdleDuration -= x.IdleDuration
}
// PeakRate returns the approximate peak rate if there was no idling.
func (tm *ThroughputMetric) PeakRate() int64 {
if tm.Bytes == 0 {
return 0
}
return int64((float64(tm.Bytes) / float64(tm.WorkDuration)) * float64(time.Second))
}
// Rate returns the observed rate.
func (tm *ThroughputMetric) Rate() int64 {
if tm.Bytes == 0 {
return 0
}
return int64((float64(tm.Bytes) / float64(tm.WorkDuration+tm.IdleDuration)) *
float64(time.Second))
}
// Utilization returns a percent [0, 1.0] indicating the percent of time
// work was performed.
func (tm *ThroughputMetric) Utilization() float64 {
if tm.WorkDuration == 0 {
return 0
}
return float64(tm.WorkDuration) / float64(tm.WorkDuration+tm.IdleDuration)
}
// GaugeSampleMetric is used to measure a gauge value (e.g. queue length) by
// accumulating samples of that gauge.
type GaugeSampleMetric struct {
// The sum of all the samples.
sampleSum int64
// The number of samples.
count int64
}
// AddSample adds the given sample.
func (gsm *GaugeSampleMetric) AddSample(sample int64) {
gsm.sampleSum += sample
gsm.count++
}
// Merge accumulates the information from another gauge metric.
func (gsm *GaugeSampleMetric) Merge(x GaugeSampleMetric) {
gsm.sampleSum += x.sampleSum
gsm.count += x.count
}
// Subtract subtracts the information from another gauge metric.
func (gsm *GaugeSampleMetric) Subtract(x GaugeSampleMetric) {
gsm.sampleSum -= x.sampleSum
gsm.count -= x.count
}
// Mean returns the mean value.
func (gsm *GaugeSampleMetric) Mean() float64 {
if gsm.count == 0 {
return 0
}
return float64(gsm.sampleSum) / float64(gsm.count)
}

76
vendor/github.com/cockroachdb/pebble/internal/base/options.go generated vendored Normal file
View File

@@ -0,0 +1,76 @@
// Copyright 2011 The LevelDB-Go and Pebble Authors. All rights reserved. Use
// of this source code is governed by a BSD-style license that can be found in
// the LICENSE file.
package base
// SSTable block defaults.
const (
DefaultBlockRestartInterval = 16
DefaultBlockSize = 4096
DefaultBlockSizeThreshold = 90
)
// FilterType is the level at which to apply a filter: block or table.
type FilterType int
// The available filter types.
const (
TableFilter FilterType = iota
)
func (t FilterType) String() string {
switch t {
case TableFilter:
return "table"
}
return "unknown"
}
// FilterWriter provides an interface for creating filter blocks. See
// FilterPolicy for more details about filters.
type FilterWriter interface {
// AddKey adds a key to the current filter block.
AddKey(key []byte)
// Finish appends to dst an encoded filter tha holds the current set of
// keys. The writer state is reset after the call to Finish allowing the
// writer to be reused for the creation of additional filters.
Finish(dst []byte) []byte
}
// FilterPolicy is an algorithm for probabilistically encoding a set of keys.
// The canonical implementation is a Bloom filter.
//
// Every FilterPolicy has a name. This names the algorithm itself, not any one
// particular instance. Aspects specific to a particular instance, such as the
// set of keys or any other parameters, will be encoded in the []byte filter
// returned by NewWriter.
//
// The name may be written to files on disk, along with the filter data. To use
// these filters, the FilterPolicy name at the time of writing must equal the
// name at the time of reading. If they do not match, the filters will be
// ignored, which will not affect correctness but may affect performance.
type FilterPolicy interface {
// Name names the filter policy.
Name() string
// MayContain returns whether the encoded filter may contain given key.
// False positives are possible, where it returns true for keys not in the
// original set.
MayContain(ftype FilterType, filter, key []byte) bool
// NewWriter creates a new FilterWriter.
NewWriter(ftype FilterType) FilterWriter
}
// BlockPropertyFilter is used in an Iterator to filter sstables and blocks
// within the sstable. It should not maintain any per-sstable state, and must
// be thread-safe.
type BlockPropertyFilter interface {
// Name returns the name of the block property collector.
Name() string
// Intersects returns true if the set represented by prop intersects with
// the set in the filter.
Intersects(prop []byte) (bool, error)
}