git.giacomolli.uk - onechat/blob - journal.go

   1 package main
   2
   3 import (
   4         "bufio"
   5         "context"
   6         "errors"
   7         "fmt"
   8         "io"
   9         "os"
  10         "path/filepath"
  11         "slices"
  12         "strconv"
  13         "strings"
  14 )
  15
  16 // Position names one line of a day file: the day as YYYYMMDD, the byte offset
  17 // of the line, and its length with the newline.
  18 type Position struct {
  19         day    uint32
  20         offset uint64
  21         length uint32
  22 }
  23
  24 // Journal appends JSON lines to one file for each day in dir, named
  25 // YYYY-MM-DD.jsonl, and reads them back by their position. It keeps the file of
  26 // the last day open for writes and each file that it read open for reads.
  27 type Journal struct {
  28         dir     string
  29         file    *os.File
  30         day     uint32
  31         size    uint64
  32         readers map[uint32]*os.File
  33 }
  34
  35 // Write appends line to the file of day, YYYY-MM-DD, with one write to the
  36 // operating system, without a buffer. A crash of the process then loses
  37 // nothing. It returns the position of the line.
  38 func (j *Journal) Write(day string, line []byte) (Position, error) {
  39         number, err := journalDayNumber(day)
  40         if err != nil {
  41                 return Position{}, err
  42         }
  43
  44         if number != j.day {
  45                 err = j.closeWriter()
  46                 if err != nil {
  47                         return Position{}, err
  48                 }
  49
  50                 j.file, err = os.OpenFile(filepath.Join(j.dir, day+".jsonl"), os.O_CREATE|os.O_WRONLY|os.O_APPEND, 0o600)
  51                 if err != nil {
  52                         return Position{}, fmt.Errorf("open day file: %w", err)
  53                 }
  54
  55                 info, err := j.file.Stat()
  56                 if err != nil {
  57                         return Position{}, fmt.Errorf("read day file size: %w", err)
  58                 }
  59
  60                 j.day = number
  61                 j.size = uint64(info.Size())
  62         }
  63
  64         _, err = j.file.Write(line)
  65         if err != nil {
  66                 return Position{}, fmt.Errorf("write line: %w", err)
  67         }
  68
  69         place := Position{
  70                 day:    number,
  71                 offset: j.size,
  72                 length: uint32(len(line)),
  73         }
  74         j.size += uint64(len(line))
  75
  76         return place, nil
  77 }
  78
  79 // Read returns the line at place without its newline.
  80 func (j *Journal) Read(place Position) ([]byte, error) {
  81         reader, open := j.readers[place.day]
  82         if !open {
  83                 var err error
  84
  85                 reader, err = os.Open(filepath.Join(j.dir, journalDayName(place.day)+".jsonl"))
  86                 if err != nil {
  87                         return nil, fmt.Errorf("open day file: %w", err)
  88                 }
  89
  90                 if j.readers == nil {
  91                         j.readers = make(map[uint32]*os.File)
  92                 }
  93
  94                 j.readers[place.day] = reader
  95         }
  96
  97         line := make([]byte, place.length)
  98         _, err := reader.ReadAt(line, int64(place.offset))
  99         if err != nil {
 100                 return nil, fmt.Errorf("read line of %s: %w", journalDayName(place.day), err)
 101         }
 102
 103         if len(line) == 0 || line[len(line)-1] != '\n' {
 104                 return nil, fmt.Errorf("line of %s at %d does not end its record", journalDayName(place.day), place.offset)
 105         }
 106
 107         return line[:len(line)-1], nil
 108 }
 109
 110 // Close closes the open day files.
 111 func (j *Journal) Close() error {
 112         err := j.closeWriter()
 113         for day, reader := range j.readers {
 114                 err = errors.Join(err, reader.Close())
 115
 116                 delete(j.readers, day)
 117         }
 118
 119         return err
 120 }
 121
 122 // Scan gives each line of the day files from place on to add, with its
 123 // position, oldest day first. The zero position starts at the first line. A
 124 // crash during a write can leave a part of a line at the end of the last day
 125 // file. Its write never returned, so Scan cuts it off.
 126 func (j *Journal) Scan(ctx context.Context, place Position, add func([]byte, Position) error) error {
 127         paths, err := filepath.Glob(filepath.Join(j.dir, "*.jsonl"))
 128         if err != nil {
 129                 return fmt.Errorf("list day files: %w", err)
 130         }
 131
 132         slices.Sort(paths)
 133
 134         for n, path := range paths {
 135                 day, err := journalDayNumber(strings.TrimSuffix(filepath.Base(path), ".jsonl"))
 136                 if err != nil {
 137                         return err
 138                 }
 139
 140                 if day < place.day {
 141                         continue
 142                 }
 143
 144                 offset := uint64(0)
 145                 if day == place.day {
 146                         offset = place.offset
 147                 }
 148
 149                 err = journalScanFile(ctx, path, day, offset, n == len(paths)-1, add)
 150                 if err != nil {
 151                         return err
 152                 }
 153         }
 154
 155         return nil
 156 }
 157
 158 func (j *Journal) closeWriter() error {
 159         if j.file == nil {
 160                 return nil
 161         }
 162
 163         err := j.file.Close()
 164         j.file = nil
 165         j.day = 0
 166         if err != nil {
 167                 return fmt.Errorf("close day file: %w", err)
 168         }
 169
 170         return nil
 171 }
 172
 173 // journalDayNumber returns the day YYYY-MM-DD as the number YYYYMMDD.
 174 func journalDayNumber(day string) (uint32, error) {
 175         number, err := strconv.ParseUint(strings.ReplaceAll(day, "-", ""), 10, 32)
 176         if err != nil || len(day) != len("2006-01-02") {
 177                 return 0, fmt.Errorf("invalid day %q", day)
 178         }
 179
 180         return uint32(number), nil
 181 }
 182
 183 // journalDayName returns the day number YYYYMMDD as YYYY-MM-DD.
 184 func journalDayName(day uint32) string {
 185         return fmt.Sprintf("%04d-%02d-%02d", day/10000, day/100%100, day%100)
 186 }
 187
 188 // journalScanFile gives each line of the day file at path from offset on to
 189 // add. In the last day file, it cuts a part of a line at the end.
 190 func journalScanFile(ctx context.Context, path string, day uint32, offset uint64, last bool, add func([]byte, Position) error) (err error) {
 191         file, err := os.Open(path)
 192         if err != nil {
 193                 return fmt.Errorf("open day file: %w", err)
 194         }
 195
 196         defer func() {
 197                 err = errors.Join(err, file.Close())
 198         }()
 199
 200         _, err = file.Seek(int64(offset), io.SeekStart)
 201         if err != nil {
 202                 return fmt.Errorf("seek day file: %w", err)
 203         }
 204
 205         reader := bufio.NewReader(file)
 206         for {
 207                 if ctx.Err() != nil {
 208                         return ctx.Err()
 209                 }
 210
 211                 line, err := reader.ReadBytes('\n')
 212                 if errors.Is(err, io.EOF) && len(line) == 0 {
 213                         return nil
 214                 }
 215
 216                 if errors.Is(err, io.EOF) && last {
 217                         return journalCutFile(path, offset)
 218                 }
 219
 220                 if errors.Is(err, io.EOF) {
 221                         return fmt.Errorf("%s ends in an incomplete line at %d", filepath.Base(path), offset)
 222                 }
 223
 224                 if err != nil {
 225                         return fmt.Errorf("read %s: %w", filepath.Base(path), err)
 226                 }
 227
 228                 place := Position{
 229                         day:    day,
 230                         offset: offset,
 231                         length: uint32(len(line)),
 232                 }
 233
 234                 err = add(line[:len(line)-1], place)
 235                 if err != nil {
 236                         return fmt.Errorf("%s at %d: %w", filepath.Base(path), offset, err)
 237                 }
 238
 239                 offset += uint64(len(line))
 240         }
 241 }
 242
 243 // journalCutFile cuts the file at path to size bytes.
 244 func journalCutFile(path string, size uint64) error {
 245         err := os.Truncate(path, int64(size))
 246         if err != nil {
 247                 return fmt.Errorf("cut incomplete line: %w", err)
 248         }
 249
 250         return nil
 251 }