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 := dayNumber(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 func (j *journal) closeWriter() error {
  80         if j.file == nil {
  81                 return nil
  82         }
  83
  84         err := j.file.Close()
  85         j.file = nil
  86         j.day = 0
  87         if err != nil {
  88                 return fmt.Errorf("close day file: %w", err)
  89         }
  90
  91         return nil
  92 }
  93
  94 // Read returns the line at place without its newline.
  95 func (j *journal) Read(place position) ([]byte, error) {
  96         reader, open := j.readers[place.day]
  97         if !open {
  98                 var err error
  99                 reader, err = os.Open(filepath.Join(j.dir, dayName(place.day)+".jsonl"))
 100                 if err != nil {
 101                         return nil, fmt.Errorf("open day file: %w", err)
 102                 }
 103
 104                 if j.readers == nil {
 105                         j.readers = make(map[uint32]*os.File)
 106                 }
 107
 108                 j.readers[place.day] = reader
 109         }
 110
 111         line := make([]byte, place.length)
 112         _, err := reader.ReadAt(line, int64(place.offset))
 113         if err != nil {
 114                 return nil, fmt.Errorf("read line of %s: %w", dayName(place.day), err)
 115         }
 116
 117         if len(line) == 0 || line[len(line)-1] != '\n' {
 118                 return nil, fmt.Errorf("line of %s at %d does not end its record", dayName(place.day), place.offset)
 119         }
 120
 121         return line[:len(line)-1], nil
 122 }
 123
 124 // Close closes the open day files.
 125 func (j *journal) Close() error {
 126         err := j.closeWriter()
 127         for day, reader := range j.readers {
 128                 err = errors.Join(err, reader.Close())
 129                 delete(j.readers, day)
 130         }
 131
 132         return err
 133 }
 134
 135 // Scan gives each line of the day files from place on to add, with its
 136 // position, oldest day first. The zero position starts at the first line. A
 137 // crash during a write can leave a part of a line at the end of the last day
 138 // file. Its write never returned, so Scan cuts it off.
 139 func (j *journal) Scan(ctx context.Context, place position, add func([]byte, position) error) error {
 140         paths, err := filepath.Glob(filepath.Join(j.dir, "*.jsonl"))
 141         if err != nil {
 142                 return fmt.Errorf("list day files: %w", err)
 143         }
 144
 145         slices.Sort(paths)
 146         for n, path := range paths {
 147                 day, err := dayNumber(strings.TrimSuffix(filepath.Base(path), ".jsonl"))
 148                 if err != nil {
 149                         return err
 150                 }
 151
 152                 if day < place.day {
 153                         continue
 154                 }
 155
 156                 offset := uint64(0)
 157                 if day == place.day {
 158                         offset = place.offset
 159                 }
 160
 161                 err = scanFile(ctx, path, day, offset, n == len(paths)-1, add)
 162                 if err != nil {
 163                         return err
 164                 }
 165         }
 166
 167         return nil
 168 }
 169
 170 // scanFile gives each line of the day file at path from offset on to add. In
 171 // the last day file, it cuts a part of a line at the end.
 172 func scanFile(ctx context.Context, path string, day uint32, offset uint64, last bool, add func([]byte, position) error) (err error) {
 173         file, err := os.Open(path)
 174         if err != nil {
 175                 return fmt.Errorf("open day file: %w", err)
 176         }
 177
 178         defer func() {
 179                 err = errors.Join(err, file.Close())
 180         }()
 181
 182         _, err = file.Seek(int64(offset), io.SeekStart)
 183         if err != nil {
 184                 return fmt.Errorf("seek day file: %w", err)
 185         }
 186
 187         reader := bufio.NewReader(file)
 188         for {
 189                 if ctx.Err() != nil {
 190                         return ctx.Err()
 191                 }
 192
 193                 line, err := reader.ReadBytes('\n')
 194                 if errors.Is(err, io.EOF) && len(line) == 0 {
 195                         return nil
 196                 }
 197
 198                 if errors.Is(err, io.EOF) && last {
 199                         return cutFile(path, offset)
 200                 }
 201
 202                 if errors.Is(err, io.EOF) {
 203                         return fmt.Errorf("%s ends in an incomplete line at %d", filepath.Base(path), offset)
 204                 }
 205
 206                 if err != nil {
 207                         return fmt.Errorf("read %s: %w", filepath.Base(path), err)
 208                 }
 209
 210                 place := position{
 211                         day:    day,
 212                         offset: offset,
 213                         length: uint32(len(line)),
 214                 }
 215
 216                 err = add(line[:len(line)-1], place)
 217                 if err != nil {
 218                         return fmt.Errorf("%s at %d: %w", filepath.Base(path), offset, err)
 219                 }
 220
 221                 offset += uint64(len(line))
 222         }
 223 }
 224
 225 // cutFile cuts the file at path to size bytes.
 226 func cutFile(path string, size uint64) error {
 227         err := os.Truncate(path, int64(size))
 228         if err != nil {
 229                 return fmt.Errorf("cut incomplete line: %w", err)
 230         }
 231
 232         return nil
 233 }
 234
 235 // dayNumber returns the day YYYY-MM-DD as the number YYYYMMDD.
 236 func dayNumber(day string) (uint32, error) {
 237         number, err := strconv.ParseUint(strings.ReplaceAll(day, "-", ""), 10, 32)
 238         if err != nil || len(day) != len("2006-01-02") {
 239                 return 0, fmt.Errorf("invalid day %q", day)
 240         }
 241
 242         return uint32(number), nil
 243 }
 244
 245 // dayName returns the day number YYYYMMDD as YYYY-MM-DD.
 246 func dayName(day uint32) string {
 247         return fmt.Sprintf("%04d-%02d-%02d", day/10000, day/100%100, day%100)
 248 }