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                 reader, err = os.Open(filepath.Join(j.dir, journalDayName(place.day)+".jsonl"))
  85                 if err != nil {
  86                         return nil, fmt.Errorf("open day file: %w", err)
  87                 }
  88
  89                 if j.readers == nil {
  90                         j.readers = make(map[uint32]*os.File)
  91                 }
  92
  93                 j.readers[place.day] = reader
  94         }
  95
  96         line := make([]byte, place.length)
  97         _, err := reader.ReadAt(line, int64(place.offset))
  98         if err != nil {
  99                 return nil, fmt.Errorf("read line of %s: %w", journalDayName(place.day), err)
 100         }
 101
 102         if len(line) == 0 || line[len(line)-1] != '\n' {
 103                 return nil, fmt.Errorf("line of %s at %d does not end its record", journalDayName(place.day), place.offset)
 104         }
 105
 106         return line[:len(line)-1], nil
 107 }
 108
 109 // Close closes the open day files.
 110 func (j *Journal) Close() error {
 111         err := j.closeWriter()
 112         for day, reader := range j.readers {
 113                 err = errors.Join(err, reader.Close())
 114                 delete(j.readers, day)
 115         }
 116
 117         return err
 118 }
 119
 120 // Scan gives each line of the day files from place on to add, with its
 121 // position, oldest day first. The zero position starts at the first line. A
 122 // crash during a write can leave a part of a line at the end of the last day
 123 // file. Its write never returned, so Scan cuts it off.
 124 func (j *Journal) Scan(ctx context.Context, place Position, add func([]byte, Position) error) error {
 125         paths, err := filepath.Glob(filepath.Join(j.dir, "*.jsonl"))
 126         if err != nil {
 127                 return fmt.Errorf("list day files: %w", err)
 128         }
 129
 130         slices.Sort(paths)
 131         for n, path := range paths {
 132                 day, err := journalDayNumber(strings.TrimSuffix(filepath.Base(path), ".jsonl"))
 133                 if err != nil {
 134                         return err
 135                 }
 136
 137                 if day < place.day {
 138                         continue
 139                 }
 140
 141                 offset := uint64(0)
 142                 if day == place.day {
 143                         offset = place.offset
 144                 }
 145
 146                 err = journalScanFile(ctx, path, day, offset, n == len(paths)-1, add)
 147                 if err != nil {
 148                         return err
 149                 }
 150         }
 151
 152         return nil
 153 }
 154
 155 func (j *Journal) closeWriter() error {
 156         if j.file == nil {
 157                 return nil
 158         }
 159
 160         err := j.file.Close()
 161         j.file = nil
 162         j.day = 0
 163         if err != nil {
 164                 return fmt.Errorf("close day file: %w", err)
 165         }
 166
 167         return nil
 168 }
 169
 170 // journalDayNumber returns the day YYYY-MM-DD as the number YYYYMMDD.
 171 func journalDayNumber(day string) (uint32, error) {
 172         number, err := strconv.ParseUint(strings.ReplaceAll(day, "-", ""), 10, 32)
 173         if err != nil || len(day) != len("2006-01-02") {
 174                 return 0, fmt.Errorf("invalid day %q", day)
 175         }
 176
 177         return uint32(number), nil
 178 }
 179
 180 // journalDayName returns the day number YYYYMMDD as YYYY-MM-DD.
 181 func journalDayName(day uint32) string {
 182         return fmt.Sprintf("%04d-%02d-%02d", day/10000, day/100%100, day%100)
 183 }
 184
 185 // journalScanFile gives each line of the day file at path from offset on to
 186 // add. In the last day file, it cuts a part of a line at the end.
 187 func journalScanFile(ctx context.Context, path string, day uint32, offset uint64, last bool, add func([]byte, Position) error) (err error) {
 188         file, err := os.Open(path)
 189         if err != nil {
 190                 return fmt.Errorf("open day file: %w", err)
 191         }
 192
 193         defer func() {
 194                 err = errors.Join(err, file.Close())
 195         }()
 196
 197         _, err = file.Seek(int64(offset), io.SeekStart)
 198         if err != nil {
 199                 return fmt.Errorf("seek day file: %w", err)
 200         }
 201
 202         reader := bufio.NewReader(file)
 203         for {
 204                 if ctx.Err() != nil {
 205                         return ctx.Err()
 206                 }
 207
 208                 line, err := reader.ReadBytes('\n')
 209                 if errors.Is(err, io.EOF) && len(line) == 0 {
 210                         return nil
 211                 }
 212
 213                 if errors.Is(err, io.EOF) && last {
 214                         return journalCutFile(path, offset)
 215                 }
 216
 217                 if errors.Is(err, io.EOF) {
 218                         return fmt.Errorf("%s ends in an incomplete line at %d", filepath.Base(path), offset)
 219                 }
 220
 221                 if err != nil {
 222                         return fmt.Errorf("read %s: %w", filepath.Base(path), err)
 223                 }
 224
 225                 place := Position{
 226                         day:    day,
 227                         offset: offset,
 228                         length: uint32(len(line)),
 229                 }
 230
 231                 err = add(line[:len(line)-1], place)
 232                 if err != nil {
 233                         return fmt.Errorf("%s at %d: %w", filepath.Base(path), offset, err)
 234                 }
 235
 236                 offset += uint64(len(line))
 237         }
 238 }
 239
 240 // journalCutFile cuts the file at path to size bytes.
 241 func journalCutFile(path string, size uint64) error {
 242         err := os.Truncate(path, int64(size))
 243         if err != nil {
 244                 return fmt.Errorf("cut incomplete line: %w", err)
 245         }
 246
 247         return nil
 248 }