AdGuardHome/internal/querylog/search.go
Stanislav Chzhen 1e0ff4d437 Pull request 2119: 6570-querylog-size-memory
Updates #6570.

Squashed commit of the following:

commit 92b2723ac1b2a78138a55cb82a3222f66119bbda
Merge: 2da12283b 0143c3aac
Author: Stanislav Chzhen <s.chzhen@adguard.com>
Date:   Tue Jan 9 17:07:23 2024 +0300

    Merge branch 'master' into 6570-querylog-size-memory

commit 2da12283b5f504fa77f08fa6026fa9a57b806b38
Author: Stanislav Chzhen <s.chzhen@adguard.com>
Date:   Tue Jan 9 17:04:54 2024 +0300

    all: imp tests

commit 1cb404c65d4e9981b709d689fd281253eca01f82
Merge: 5f7d20516 94d437d40
Author: Stanislav Chzhen <s.chzhen@adguard.com>
Date:   Thu Dec 28 20:18:00 2023 +0300

    Merge branch 'master' into 6570-querylog-size-memory

commit 5f7d20516934867e1a141c19a97edb49407884ee
Author: Stanislav Chzhen <s.chzhen@adguard.com>
Date:   Wed Dec 27 15:07:54 2023 +0300

    all: imp docs

commit 0b17cfc4243a88606c62e4b1ae893a09149655b5
Author: Stanislav Chzhen <s.chzhen@adguard.com>
Date:   Mon Dec 25 20:06:09 2023 +0300

    all: querylog size memory
2024-01-11 18:38:30 +03:00

340 lines
8.4 KiB
Go

package querylog
import (
"fmt"
"io"
"time"
"github.com/AdguardTeam/golibs/errors"
"github.com/AdguardTeam/golibs/log"
"golang.org/x/exp/slices"
)
// client finds the client info, if any, by its ClientID and IP address,
// optionally checking the provided cache. It will use the IP address
// regardless of if the IP anonymization is enabled now, because the
// anonymization could have been disabled in the past, and client will try to
// find those records as well.
func (l *queryLog) client(clientID, ip string, cache clientCache) (c *Client, err error) {
cck := clientCacheKey{clientID: clientID, ip: ip}
var ok bool
if c, ok = cache[cck]; ok {
return c, nil
}
var ids []string
if clientID != "" {
ids = append(ids, clientID)
}
if ip != "" {
ids = append(ids, ip)
}
c, err = l.findClient(ids)
if err != nil {
return nil, err
}
// Cache all results, including negative ones, to prevent excessive and
// expensive client searching.
cache[cck] = c
return c, nil
}
// searchMemory looks up log records which are currently in the in-memory
// buffer. It optionally uses the client cache, if provided. It also returns
// the total amount of records in the buffer at the moment of searching.
// l.confMu is expected to be locked.
func (l *queryLog) searchMemory(params *searchParams, cache clientCache) (entries []*logEntry, total int) {
// We use this configuration check because a buffer can contain a single log
// record. See [newQueryLog].
if l.conf.MemSize == 0 {
return nil, 0
}
l.bufferLock.Lock()
defer l.bufferLock.Unlock()
l.buffer.ReverseRange(func(entry *logEntry) (cont bool) {
// A shallow clone is enough, since the only thing that this loop
// modifies is the client field.
e := entry.shallowClone()
var err error
e.client, err = l.client(e.ClientID, e.IP.String(), cache)
if err != nil {
msg := "querylog: enriching memory record at time %s" +
" for client %q (clientid %q): %s"
log.Error(msg, e.Time, e.IP, e.ClientID, err)
// Go on and try to match anyway.
}
if params.match(e) {
entries = append(entries, e)
}
return true
})
return entries, int(l.buffer.Len())
}
// search searches log entries in memory buffer and log file using specified
// parameters and returns the list of entries found and the time of the oldest
// entry. l.confMu is expected to be locked.
func (l *queryLog) search(params *searchParams) (entries []*logEntry, oldest time.Time) {
start := time.Now()
if params.limit == 0 {
return []*logEntry{}, time.Time{}
}
cache := clientCache{}
memoryEntries, bufLen := l.searchMemory(params, cache)
log.Debug("querylog: got %d entries from memory", len(memoryEntries))
fileEntries, oldest, total := l.searchFiles(params, cache)
log.Debug("querylog: got %d entries from files", len(fileEntries))
total += bufLen
totalLimit := params.offset + params.limit
// now let's get a unified collection
entries = append(memoryEntries, fileEntries...)
if len(entries) > totalLimit {
// remove extra records
entries = entries[:totalLimit]
}
// Resort entries on start time to partially mitigate query log looking
// weird on the frontend.
//
// See https://github.com/AdguardTeam/AdGuardHome/issues/2293.
slices.SortStableFunc(entries, func(a, b *logEntry) (res int) {
return -a.Time.Compare(b.Time)
})
if params.offset > 0 {
if len(entries) > params.offset {
entries = entries[params.offset:]
} else {
entries = make([]*logEntry, 0)
oldest = time.Time{}
}
}
if len(entries) > 0 {
// Update oldest after merging in the memory buffer.
oldest = entries[len(entries)-1].Time
}
log.Debug(
"querylog: prepared data (%d/%d) older than %s in %s",
len(entries),
total,
params.olderThan,
time.Since(start),
)
return entries, oldest
}
// seekRecord changes the current position to the next record older than the
// provided parameter.
func (r *qLogReader) seekRecord(olderThan time.Time) (err error) {
if olderThan.IsZero() {
return r.SeekStart()
}
err = r.seekTS(olderThan.UnixNano())
if err == nil {
// Read to the next record, because we only need the one that goes
// after it.
_, err = r.ReadNext()
}
return err
}
// setQLogReader creates a reader with the specified files and sets the
// position to the next record older than the provided parameter.
func (l *queryLog) setQLogReader(olderThan time.Time) (qr *qLogReader, err error) {
files := []string{
l.logFile + ".1",
l.logFile,
}
r, err := newQLogReader(files)
if err != nil {
return nil, fmt.Errorf("opening qlog reader: %s", err)
}
err = r.seekRecord(olderThan)
if err != nil {
defer func() { err = errors.WithDeferred(err, r.Close()) }()
log.Debug("querylog: cannot seek to %s: %s", olderThan, err)
return nil, nil
}
return r, nil
}
// readEntries reads entries from the reader to totalLimit. By default, we do
// not scan more than maxFileScanEntries at once. The idea is to make search
// calls faster so that the UI could handle it and show something quicker.
// This behavior can be overridden if maxFileScanEntries is set to 0.
func (l *queryLog) readEntries(
r *qLogReader,
params *searchParams,
cache clientCache,
totalLimit int,
) (entries []*logEntry, oldestNano int64, total int) {
for total < params.maxFileScanEntries || params.maxFileScanEntries <= 0 {
ent, ts, rErr := l.readNextEntry(r, params, cache)
if rErr != nil {
if rErr == io.EOF {
oldestNano = 0
break
}
log.Error("querylog: reading next entry: %s", rErr)
}
oldestNano = ts
total++
if ent == nil {
continue
}
entries = append(entries, ent)
if len(entries) == totalLimit {
break
}
}
return entries, oldestNano, total
}
// searchFiles looks up log records from all log files. It optionally uses the
// client cache, if provided. searchFiles does not scan more than
// maxFileScanEntries so callers may need to call it several times to get all
// the results. oldest and total are the time of the oldest processed entry
// and the total number of processed entries, including discarded ones,
// correspondingly.
func (l *queryLog) searchFiles(
params *searchParams,
cache clientCache,
) (entries []*logEntry, oldest time.Time, total int) {
r, err := l.setQLogReader(params.olderThan)
if err != nil {
log.Error("querylog: %s", err)
}
if r == nil {
return entries, oldest, 0
}
defer func() {
if closeErr := r.Close(); closeErr != nil {
log.Error("querylog: closing file: %s", closeErr)
}
}()
totalLimit := params.offset + params.limit
entries, oldestNano, total := l.readEntries(r, params, cache, totalLimit)
if oldestNano != 0 {
oldest = time.Unix(0, oldestNano)
}
return entries, oldest, total
}
// quickMatchClientFinder is a wrapper around the usual client finding function
// to make it easier to use with quick matches.
type quickMatchClientFinder struct {
client func(clientID, ip string, cache clientCache) (c *Client, err error)
cache clientCache
}
// findClient is a method that can be used as a quickMatchClientFinder.
func (f quickMatchClientFinder) findClient(clientID, ip string) (c *Client) {
var err error
c, err = f.client(clientID, ip, f.cache)
if err != nil {
log.Error(
"querylog: enriching file record for quick search: for client %q (clientid %q): %s",
ip,
clientID,
err,
)
}
return c
}
// readNextEntry reads the next log entry and checks if it matches the search
// criteria. It optionally uses the client cache, if provided. e is nil if
// the entry doesn't match the search criteria. ts is the timestamp of the
// processed entry.
func (l *queryLog) readNextEntry(
r *qLogReader,
params *searchParams,
cache clientCache,
) (e *logEntry, ts int64, err error) {
var line string
line, err = r.ReadNext()
if err != nil {
return nil, 0, err
}
clientFinder := quickMatchClientFinder{
client: l.client,
cache: cache,
}
if !params.quickMatch(line, clientFinder.findClient) {
ts = readQLogTimestamp(line)
return nil, ts, nil
}
e = &logEntry{}
decodeLogEntry(e, line)
if l.isIgnored(e.QHost) {
return nil, ts, nil
}
e.client, err = l.client(e.ClientID, e.IP.String(), cache)
if err != nil {
log.Error(
"querylog: enriching file record at time %s for client %q (clientid %q): %s",
e.Time,
e.IP,
e.ClientID,
err,
)
// Go on and try to match anyway.
}
if e.client != nil && e.client.IgnoreQueryLog {
return nil, ts, nil
}
ts = e.Time.UnixNano()
if !params.match(e) {
return nil, ts, nil
}
return e, ts, nil
}