AdGuardHome/querylog/querylog_file.go
2019-10-02 15:41:14 +03:00

504 lines
10 KiB
Go

package querylog
import (
"bytes"
"compress/gzip"
"encoding/json"
"fmt"
"io"
"os"
"sync"
"time"
"github.com/AdguardTeam/golibs/log"
"github.com/go-test/deep"
)
var (
fileWriteLock sync.Mutex
)
const enableGzip = false
const maxEntrySize = 1000
// flushLogBuffer flushes the current buffer to file and resets the current buffer
func (l *queryLog) flushLogBuffer(fullFlush bool) error {
l.fileFlushLock.Lock()
defer l.fileFlushLock.Unlock()
// flush remainder to file
l.logBufferLock.Lock()
needFlush := len(l.logBuffer) >= logBufferCap
if !needFlush && !fullFlush {
l.logBufferLock.Unlock()
return nil
}
flushBuffer := l.logBuffer
l.logBuffer = nil
l.flushPending = false
l.logBufferLock.Unlock()
err := l.flushToFile(flushBuffer)
if err != nil {
log.Error("Saving querylog to file failed: %s", err)
return err
}
return nil
}
// flushToFile saves the specified log entries to the query log file
func (l *queryLog) flushToFile(buffer []*logEntry) error {
if len(buffer) == 0 {
log.Debug("querylog: there's nothing to write to a file")
return nil
}
start := time.Now()
var b bytes.Buffer
e := json.NewEncoder(&b)
for _, entry := range buffer {
err := e.Encode(entry)
if err != nil {
log.Error("Failed to marshal entry: %s", err)
return err
}
}
elapsed := time.Since(start)
log.Debug("%d elements serialized via json in %v: %d kB, %v/entry, %v/entry", len(buffer), elapsed, b.Len()/1024, float64(b.Len())/float64(len(buffer)), elapsed/time.Duration(len(buffer)))
err := checkBuffer(buffer, b)
if err != nil {
log.Error("failed to check buffer: %s", err)
return err
}
var zb bytes.Buffer
filename := l.logFile
// gzip enabled?
if enableGzip {
filename += ".gz"
zw := gzip.NewWriter(&zb)
zw.Name = l.logFile
zw.ModTime = time.Now()
_, err = zw.Write(b.Bytes())
if err != nil {
log.Error("Couldn't compress to gzip: %s", err)
zw.Close()
return err
}
if err = zw.Close(); err != nil {
log.Error("Couldn't close gzip writer: %s", err)
return err
}
} else {
zb = b
}
fileWriteLock.Lock()
defer fileWriteLock.Unlock()
f, err := os.OpenFile(filename, os.O_WRONLY|os.O_CREATE|os.O_APPEND, 0644)
if err != nil {
log.Error("failed to create file \"%s\": %s", filename, err)
return err
}
defer f.Close()
n, err := f.Write(zb.Bytes())
if err != nil {
log.Error("Couldn't write to file: %s", err)
return err
}
log.Debug("ok \"%s\": %v bytes written", filename, n)
return nil
}
func checkBuffer(buffer []*logEntry, b bytes.Buffer) error {
l := len(buffer)
d := json.NewDecoder(&b)
i := 0
for d.More() {
entry := &logEntry{}
err := d.Decode(entry)
if err != nil {
log.Error("Failed to decode: %s", err)
return err
}
if diff := deep.Equal(entry, buffer[i]); diff != nil {
log.Error("decoded buffer differs: %s", diff)
return fmt.Errorf("decoded buffer differs: %s", diff)
}
i++
}
if i != l {
err := fmt.Errorf("check fail: %d vs %d entries", l, i)
log.Error("%v", err)
return err
}
log.Debug("check ok: %d entries", i)
return nil
}
func (l *queryLog) rotateQueryLog() error {
from := l.logFile
to := l.logFile + ".1"
if enableGzip {
from = l.logFile + ".gz"
to = l.logFile + ".gz.1"
}
if _, err := os.Stat(from); os.IsNotExist(err) {
// do nothing, file doesn't exist
return nil
}
err := os.Rename(from, to)
if err != nil {
log.Error("Failed to rename querylog: %s", err)
return err
}
log.Debug("Rotated from %s to %s successfully", from, to)
return nil
}
func (l *queryLog) periodicQueryLogRotate() {
for range time.Tick(time.Duration(l.conf.Interval) * time.Hour) {
err := l.rotateQueryLog()
if err != nil {
log.Error("Failed to rotate querylog: %s", err)
// do nothing, continue rotating
}
}
}
// Reader is the DB reader context
type Reader struct {
ql *queryLog
f *os.File
jd *json.Decoder
now time.Time
validFrom int64 // UNIX time (ns)
olderThan int64 // UNIX time (ns)
files []string
ifile int
limit uint64
count uint64 // counter for returned elements
latest bool // return the latest entries
filePrepared bool
searching bool // we're seaching for an entry with exact time stamp
fseeker fileSeeker // file seeker object
fpos uint64 // current file offset
nSeekRequests uint32 // number of Seek() requests made (finding a new line doesn't count)
}
type fileSeeker struct {
target uint64 // target value
pos uint64 // current offset, may be adjusted by user for increased accuracy
lastpos uint64 // the last offset returned
lo uint64 // low boundary offset
hi uint64 // high boundary offset
}
// OpenReader - return reader object
func (l *queryLog) OpenReader() *Reader {
r := Reader{}
r.ql = l
r.now = time.Now()
r.validFrom = r.now.Unix() - int64(l.conf.Interval*60*60)
r.validFrom *= 1000000000
r.files = []string{
r.ql.logFile,
r.ql.logFile + ".1",
}
return &r
}
// Close - close the reader
func (r *Reader) Close() {
elapsed := time.Since(r.now)
var perunit time.Duration
if r.count > 0 {
perunit = elapsed / time.Duration(r.count)
}
log.Debug("querylog: read %d entries in %v, %v/entry, seek-reqs:%d",
r.count, elapsed, perunit, r.nSeekRequests)
if r.f != nil {
r.f.Close()
}
}
// BeginRead - start reading
// olderThan: stop returning entries when an entry with this time is reached
// count: minimum number of entries to return
func (r *Reader) BeginRead(olderThan time.Time, count uint64) {
r.olderThan = olderThan.UnixNano()
r.latest = olderThan.IsZero()
r.limit = count
if r.latest {
r.olderThan = r.now.UnixNano()
}
r.filePrepared = false
r.searching = false
r.jd = nil
}
// BeginReadPrev - start reading the previous data chunk
func (r *Reader) BeginReadPrev(olderThan time.Time, count uint64) {
r.olderThan = olderThan.UnixNano()
r.latest = olderThan.IsZero()
r.limit = count
if r.latest {
r.olderThan = r.now.UnixNano()
}
off := r.fpos - maxEntrySize*(r.limit+1)
if int64(off) < maxEntrySize {
off = 0
}
r.fpos = off
log.Debug("QueryLog: seek: %x", off)
_, err := r.f.Seek(int64(off), io.SeekStart)
if err != nil {
log.Error("file.Seek: %s: %s", r.files[r.ifile], err)
return
}
r.nSeekRequests++
r.seekToNewLine()
r.fseeker.pos = r.fpos
r.filePrepared = true
r.searching = false
r.jd = nil
}
// Perform binary seek
// Return 0: success; 1: seek reqiured; -1: error
func (fs *fileSeeker) seekBinary(cur uint64) int32 {
log.Debug("QueryLog: seek: tgt=%x cur=%x, %x: [%x..%x]", fs.target, cur, fs.pos, fs.lo, fs.hi)
off := uint64(0)
if fs.pos >= fs.lo && fs.pos < fs.hi {
if cur == fs.target {
return 0
} else if cur < fs.target {
fs.lo = fs.pos + 1
} else {
fs.hi = fs.pos
}
off = fs.lo + (fs.hi-fs.lo)/2
} else {
// we didn't find another entry from the last file offset: now return the boundary beginning
off = fs.lo
}
if off == fs.lastpos {
return -1
}
fs.lastpos = off
fs.pos = off
return 1
}
// Seek to a new line
func (r *Reader) seekToNewLine() bool {
b := make([]byte, maxEntrySize*2)
_, err := r.f.Read(b)
if err != nil {
log.Error("QueryLog: file.Read: %s: %s", r.files[r.ifile], err)
return false
}
off := bytes.IndexByte(b, '\n') + 1
if off == 0 {
log.Error("QueryLog: Can't find a new line: %s", r.files[r.ifile])
return false
}
r.fpos += uint64(off)
log.Debug("QueryLog: seek: %x (+%d)", r.fpos, off)
_, err = r.f.Seek(int64(r.fpos), io.SeekStart)
if err != nil {
log.Error("QueryLog: file.Seek: %s: %s", r.files[r.ifile], err)
return false
}
return true
}
// Open a file
func (r *Reader) openFile() bool {
var err error
fn := r.files[r.ifile]
r.f, err = os.Open(fn)
if err != nil {
if !os.IsNotExist(err) {
log.Error("QueryLog: Failed to open file \"%s\": %s", fn, err)
}
return false
}
return true
}
// Seek to the needed position
func (r *Reader) prepareRead() bool {
fn := r.files[r.ifile]
fi, err := r.f.Stat()
if err != nil {
log.Error("QueryLog: file.Stat: %s: %s", fn, err)
return false
}
fsize := uint64(fi.Size())
off := uint64(0)
if r.latest {
// read data from the end of file
off = fsize - maxEntrySize*(r.limit+1)
if int64(off) < maxEntrySize {
off = 0
}
r.fpos = off
log.Debug("QueryLog: seek: %x", off)
_, err = r.f.Seek(int64(off), io.SeekStart)
if err != nil {
log.Error("QueryLog: file.Seek: %s: %s", fn, err)
return false
}
} else {
// start searching in file: we'll read the first chunk of data from the middle of file
r.searching = true
r.fseeker = fileSeeker{}
r.fseeker.target = uint64(r.olderThan)
r.fseeker.hi = fsize
rc := r.fseeker.seekBinary(0)
r.fpos = r.fseeker.pos
if rc == 1 {
_, err = r.f.Seek(int64(r.fpos), io.SeekStart)
if err != nil {
log.Error("QueryLog: file.Seek: %s: %s", fn, err)
return false
}
}
}
r.nSeekRequests++
if !r.seekToNewLine() {
return false
}
r.fseeker.pos = r.fpos
return true
}
// Next - return the next entry or nil if reading is finished
func (r *Reader) Next() *logEntry { // nolint
var err error
for {
// open file if needed
if r.f == nil {
if r.ifile == len(r.files) {
return nil
}
if !r.openFile() {
r.ifile++
continue
}
}
if !r.filePrepared {
if !r.prepareRead() {
return nil
}
r.filePrepared = true
}
// open decoder if needed
if r.jd == nil {
r.jd = json.NewDecoder(r.f)
}
// check if there's data
if !r.jd.More() {
r.jd = nil
return nil
}
// read data
var entry logEntry
err = r.jd.Decode(&entry)
if err != nil {
log.Error("QueryLog: Failed to decode: %s", err)
r.jd = nil
return nil
}
t := entry.Time.UnixNano()
if r.searching {
r.jd = nil
rr := r.fseeker.seekBinary(uint64(t))
r.fpos = r.fseeker.pos
if rr < 0 {
log.Error("QueryLog: File seek error: can't find the target entry: %s", r.files[r.ifile])
return nil
} else if rr == 0 {
// We found the target entry.
// We'll start reading the previous chunk of data.
r.searching = false
off := r.fpos - (maxEntrySize * (r.limit + 1))
if int64(off) < maxEntrySize {
off = 0
}
r.fpos = off
}
_, err = r.f.Seek(int64(r.fpos), io.SeekStart)
if err != nil {
log.Error("QueryLog: file.Seek: %s: %s", r.files[r.ifile], err)
return nil
}
r.nSeekRequests++
if !r.seekToNewLine() {
return nil
}
r.fseeker.pos = r.fpos
continue
}
if t < r.validFrom {
continue
}
if t >= r.olderThan {
return nil
}
r.count++
return &entry
}
}
// Total returns the total number of items
func (r *Reader) Total() int {
return 0
}