Files
qb/database/database.go

468 lines
11 KiB
Go
Raw Normal View History

2026-02-10 14:02:11 +00:00
package database
import (
"errors"
"fmt"
"hash/crc32"
"io"
"log"
"net"
"os"
"path/filepath"
"sync"
"time"
"gordenko.dev/dima/qb/atree"
"gordenko.dev/dima/qb/bin"
"gordenko.dev/dima/qb/freelist"
"gordenko.dev/dima/qb/txlog"
)
func JoinSnapshotFileName(dir string, logNumber int) string {
return filepath.Join(dir, fmt.Sprintf("%d.snapshot", logNumber))
}
2026-05-31 20:01:28 +00:00
// type metricLockEntry struct {
// XLock bool
// RLocks int
// WaitQueue []any
// }
2026-02-10 14:02:11 +00:00
type Database struct {
2026-05-10 19:15:11 +00:00
mutex sync.Mutex
workerSignalCh chan struct{}
workerQueue []any
rLocksToRelease []uint32
metrics map[uint32]*_metric
//metricLockEntries map[uint32]*metricLockEntry
2026-05-31 20:01:28 +00:00
dataFreeList *freelist.FreeList
indexFreeList *freelist.FreeList
dir string
databaseName string
txlog *txlog.Writer
atree *atree.Atree
tcpPort int
logfile *os.File
logger *log.Logger
exitCh chan struct{}
waitGroup *sync.WaitGroup
2026-02-10 14:02:11 +00:00
}
type Options struct {
TCPPort int
Dir string
DatabaseName string
RedoDir string
Logfile *os.File
ExitCh chan struct{}
WaitGroup *sync.WaitGroup
}
func New(opt Options) (_ *Database, err error) {
if opt.TCPPort <= 0 {
return nil, errors.New("TCPPort option is required")
}
if opt.Dir == "" {
return nil, errors.New("Dir option is required")
}
if opt.DatabaseName == "" {
return nil, errors.New("DatabaseName option is required")
}
if opt.RedoDir == "" {
return nil, errors.New("RedoDir option is required")
}
if opt.Logfile == nil {
return nil, errors.New("Logfile option is required")
}
if opt.ExitCh == nil {
return nil, errors.New("ExitCh option is required")
}
if opt.WaitGroup == nil {
return nil, errors.New("WaitGroup option is required")
}
2026-05-31 20:01:28 +00:00
// deltaFreeListFile, err := os.OpenFile(
// filepath.Join(opt.Dir, opt.DatabaseName+".deltafree"),
// os.O_RDWR|os.O_CREATE,
// 0666,
// )
// if err != nil {
// return nil, err
// }
2026-05-21 23:38:52 +03:00
2026-05-31 20:01:28 +00:00
dataFreeList, err := freelist.New(freelist.Options{
PageSize: 2048,
BaseFilePath: filepath.Join(opt.Dir, opt.DatabaseName+".free_data_base"),
DeltaFilePath: filepath.Join(opt.Dir, opt.DatabaseName+".free_data_delta"),
})
2026-05-21 23:38:52 +03:00
2026-05-31 20:01:28 +00:00
indexFreeList, err := freelist.New(freelist.Options{
PageSize: 2048,
BaseFilePath: filepath.Join(opt.Dir, opt.DatabaseName+".free_index_base"),
DeltaFilePath: filepath.Join(opt.Dir, opt.DatabaseName+".free_index_delta"),
2026-05-21 23:38:52 +03:00
})
2026-02-10 14:02:11 +00:00
s := &Database{
2026-05-10 19:15:11 +00:00
workerSignalCh: make(chan struct{}, 1),
dir: opt.Dir,
databaseName: opt.DatabaseName,
metrics: make(map[uint32]*_metric),
//metricLockEntries: make(map[uint32]*metricLockEntry),
2026-05-31 20:01:28 +00:00
dataFreeList: dataFreeList,
indexFreeList: indexFreeList,
tcpPort: opt.TCPPort,
logfile: opt.Logfile,
logger: log.New(opt.Logfile, "", log.LstdFlags),
exitCh: opt.ExitCh,
waitGroup: opt.WaitGroup,
2026-02-10 14:02:11 +00:00
}
return s, nil
}
func (s *Database) ListenAndServe() (err error) {
listener, err := net.Listen("tcp", fmt.Sprintf(":%d", s.tcpPort))
if err != nil {
return fmt.Errorf("net.Listen: %s; port=%d", err, s.tcpPort)
}
s.atree, err = atree.New(atree.Options{
2026-02-23 19:37:05 +00:00
Dir: s.dir,
DatabaseName: s.databaseName,
2026-02-10 14:02:11 +00:00
})
if err != nil {
return fmt.Errorf("atree.New: %s", err)
}
s.atree.Run()
go s.worker()
s.recovery()
s.logger.Println("database started")
for {
// Listen for an incoming connection.
conn, err := listener.Accept()
if err != nil {
s.logger.Printf("listener.Accept: %s\n", err)
time.Sleep(time.Second)
} else {
go s.handleTCPConn(conn)
}
}
}
func (s *Database) recovery() {
2026-05-31 20:01:28 +00:00
// FIX
// advisor, err := recovery.NewRecoveryAdvisor(recovery.RecoveryAdvisorOptions{
// Dir: s.dir,
// VerifySnapshot: s.verifySnapshot,
// })
// if err != nil {
// panic(err)
// }
2026-02-10 14:02:11 +00:00
2026-05-31 20:01:28 +00:00
// recipe, err := advisor.GetRecipe()
// if err != nil {
// qb.Abort(qb.GetRecoveryRecipeFailed, err)
// }
2026-02-10 14:02:11 +00:00
2026-05-31 20:01:28 +00:00
// var logNumber int
2026-02-10 14:02:11 +00:00
2026-05-31 20:01:28 +00:00
// if recipe != nil {
// if recipe.Snapshot != "" {
// err = s.loadSnapshot(recipe.Snapshot)
// if err != nil {
// qb.Abort(qb.LoadSnapshotFailed, err)
// }
// }
// for _, changesFileName := range recipe.Changes {
// err = s.replayChanges(changesFileName)
// if err != nil {
// qb.Abort(qb.ReplayChangesFailed, err)
// }
// }
// logNumber = recipe.LogNumber
// }
2026-02-10 14:02:11 +00:00
2026-05-31 20:01:28 +00:00
// s.txlog, err = txlog.NewWriter(txlog.WriterOptions{
// Dir: s.dir,
// LogNumber: logNumber,
// AppendToWorkerQueue: s.appendJobToWorkerQueue,
// FreeList: s.freeList,
// Atree: s.atree,
// ExitCh: s.exitCh,
// WaitGroup: s.waitGroup,
// })
2026-02-23 19:37:05 +00:00
// if err != nil {
2026-05-31 20:01:28 +00:00
// qb.Abort(qb.CreateChangesWriterFailed, err)
2026-02-23 19:37:05 +00:00
// }
2026-05-31 20:01:28 +00:00
// go s.txlog.Run()
// // fileNames, err := s.searchREDOFiles()
// // if err != nil {
// // qb.Abort(qb.SearchREDOFilesFailed, err)
// // }
// // if len(fileNames) > 0 {
// // for _, fileName := range fileNames {
// // err = s.replayREDOFile(fileName)
// // if err != nil {
// // qb.Abort(qb.ReplayREDOFileFailed, err)
// // }
// // }
// // for _, fileName := range fileNames {
// // err = os.Remove(fileName)
// // if err != nil {
// // qb.Abort(qb.RemoveREDOFileFailed, err)
// // }
// // }
// // }
2026-02-23 19:37:05 +00:00
2026-05-31 20:01:28 +00:00
// if recipe != nil {
// if recipe.CompleteSnapshot {
// err = s.dumpSnapshot(logNumber)
2026-02-23 19:37:05 +00:00
// if err != nil {
2026-05-31 20:01:28 +00:00
// qb.Abort(qb.DumpSnapshotFailed, err)
2026-02-23 19:37:05 +00:00
// }
// }
2026-05-31 20:01:28 +00:00
// for _, fileName := range recipe.ToDelete {
2026-02-23 19:37:05 +00:00
// err = os.Remove(fileName)
// if err != nil {
2026-05-31 20:01:28 +00:00
// qb.Abort(qb.RemoveRecipeFileFailed, err)
2026-02-23 19:37:05 +00:00
// }
// }
// }
2026-02-10 14:02:11 +00:00
}
2026-02-23 19:37:05 +00:00
// func (s *Database) searchREDOFiles() ([]string, error) {
// var (
// reREDO = regexp.MustCompile(`a\d+\.redo`)
// fileNames []string
// )
// entries, err := os.ReadDir(s.redoDir)
// if err != nil {
// return nil, err
// }
// for _, entry := range entries {
// if entry.Type().IsRegular() {
// baseName := entry.Name()
// if reREDO.MatchString(baseName) {
// fileNames = append(fileNames, filepath.Join(s.redoDir, baseName))
// }
// }
// }
// return fileNames, nil
// }
// func (s *Database) replayREDOFile(fileName string) error {
// redoFile, err := redo.ReadREDOFile(redo.ReadREDOFileReq{
// FileName: fileName,
// DataPageSize: atree.DataPageSize,
// IndexPageSize: atree.IndexPageSize,
// })
// if err != nil {
// return fmt.Errorf("can't read REDO file %s: %s", fileName, err)
// }
// metric, ok := s.metrics[redoFile.MetricID]
// if !ok {
// return fmt.Errorf("has REDOFile, metric %d not found", redoFile.MetricID)
// }
// if metric.Until < redoFile.Timestamp {
// waitCh := make(chan struct{})
// s.atree.ApplyREDO(atree.WriteTask{
// DataPage: redoFile.DataPage,
// IndexPages: redoFile.IndexPages,
// })
// <-waitCh
// waitCh = s.txlog.WriteAppendedMeasureWithOverflow(
// txlog.AppendedMeasureWithOverflow{
// MetricID: redoFile.MetricID,
// Timestamp: redoFile.Timestamp,
// Value: redoFile.Value,
// IsDataPageReused: redoFile.IsDataPageReused,
// DataPageNo: redoFile.DataPage.PageNo,
// IsRootChanged: redoFile.IsRootChanged,
// RootPageNo: redoFile.RootPageNo,
// ReusedIndexPages: redoFile.ReusedIndexPages,
// },
// fileName,
// false,
// )
// <-waitCh
// }
// return nil
// }
2026-02-10 14:02:11 +00:00
func (s *Database) verifySnapshot(fileName string) (_ bool, err error) {
file, err := os.Open(fileName)
if err != nil {
return
}
defer file.Close()
stat, err := file.Stat()
if err != nil {
return
}
if stat.Size() <= 4 {
return false, nil
}
var (
payloadSize = stat.Size() - 4
hash = crc32.NewIEEE()
)
_, err = io.CopyN(hash, file, payloadSize)
if err != nil {
return
}
calculatedCRC := hash.Sum32()
storedCRC, err := bin.ReadUint32(file)
if err != nil {
return
}
if storedCRC != calculatedCRC {
return false, fmt.Errorf("strored CRC %d not equal calculated CRC %d",
storedCRC, calculatedCRC)
}
return true, nil
}
func (s *Database) replayChanges(fileName string) error {
walReader, err := txlog.NewReader(txlog.ReaderOptions{
FileName: fileName,
BufferSize: 1024 * 1024,
})
if err != nil {
return err
}
for {
lsn, records, done, err := walReader.ReadPacket()
if err != nil {
return err
}
_ = lsn
if done {
return nil
}
for _, record := range records {
if err = s.replayChangesRecord(record); err != nil {
return err
}
}
}
}
func (s *Database) replayChangesRecord(untyped any) error {
2026-05-21 23:38:52 +03:00
// switch rec := untyped.(type) {
// case txlog.AddedMetric:
// var (
// values qb.ValueCompressor
// timestampsBuf = conbuf.New(nil)
// valuesBuf = conbuf.New(nil)
// )
// if rec.MetricType == qb.Cumulative {
// values = chunkenc.NewReverseCumulativeDeltaCompressor(
// valuesBuf, 0, byte(rec.FracDigits))
// } else {
// values = chunkenc.NewReverseInstantDeltaCompressor(
// valuesBuf, 0, byte(rec.FracDigits))
// }
2026-02-10 14:02:11 +00:00
2026-05-21 23:38:52 +03:00
// s.metrics[rec.MetricID] = &_metric{
// MetricType: rec.MetricType,
// FracDigits: byte(rec.FracDigits),
// TimestampsBuf: timestampsBuf,
// ValuesBuf: valuesBuf,
// Timestamps: chunkenc.NewReverseTimeDeltaCompressor(timestampsBuf, 0),
// Values: values,
// }
2026-02-10 14:02:11 +00:00
2026-05-21 23:38:52 +03:00
// case txlog.DeletedMetric:
// delete(s.metrics, rec.MetricID)
// if len(rec.FreePageNumbers) > 0 {
// s.freeList.AddPages(rec.FreePageNumbers)
// }
2026-02-10 14:02:11 +00:00
2026-05-21 23:38:52 +03:00
// case txlog.AppendedMeasure:
// metric, ok := s.metrics[rec.MetricID]
// if ok {
// metric.Timestamps.Append(rec.Timestamp)
// metric.Values.Append(rec.Value)
2026-02-10 14:02:11 +00:00
2026-05-21 23:38:52 +03:00
// if metric.Since == 0 {
// metric.Since = rec.Timestamp
// metric.SinceValue = rec.Value
// }
2026-02-10 14:02:11 +00:00
2026-05-21 23:38:52 +03:00
// metric.Until = rec.Timestamp
// metric.UntilValue = rec.Value
2026-05-10 00:59:47 +00:00
// }
2026-05-21 23:38:52 +03:00
// case txlog.AppendedMeasures:
// metric, ok := s.metrics[rec.MetricID]
// if ok {
// for _, measure := range rec.Measures {
// metric.Timestamps.Append(measure.Timestamp)
// metric.Values.Append(measure.Value)
// if metric.Since == 0 {
// metric.Since = measure.Timestamp
// metric.SinceValue = measure.Value
// }
// metric.Until = measure.Timestamp
// metric.UntilValue = measure.Value
// }
2026-05-10 00:59:47 +00:00
// }
2026-02-10 14:02:11 +00:00
2026-05-21 23:38:52 +03:00
// // case txlog.AppendedMeasureWithOverflow:
// // metric, ok := s.metrics[rec.MetricID]
// // if ok {
// // metric.ReinitBy(rec.Timestamp, rec.Value)
// // if rec.IsRootChanged {
// // metric.RootPageNo = rec.RootPageNo
// // }
// // metric.LastPageNo = rec.DataPageNo
// // // delete free pages
// // if rec.IsDataPageReused {
// // s.freeList.DeleteReservedPages([]uint32{
// // rec.DataPageNo,
// // })
// // }
// // if len(rec.ReusedIndexPages) > 0 {
// // s.freeList.DeleteReservedPages(rec.ReusedIndexPages)
// // }
// // }
// case txlog.DeletedMeasures:
// metric, ok := s.metrics[rec.MetricID]
// if ok {
// metric.DeleteMeasures()
// if len(rec.FreePageNumbers) > 0 {
// s.freeList.AddPages(rec.FreePageNumbers)
// }
// }
2026-02-10 14:02:11 +00:00
2026-05-21 23:38:52 +03:00
// default:
// qb.Abort(qb.UnknownTxLogRecordTypeBug,
// fmt.Errorf("bug: unknown record type %T in TransactionLog", rec))
// }
2026-02-10 14:02:11 +00:00
return nil
}