AdGuardHome/internal/stats/stats.go
Ainar Garipov 167b112511 Pull request: 5035-more-clients-netip-addr
Updates #5035.

Squashed commit of the following:

commit 1934ea14299921760e9fcf6dd9053bd3155cb40e
Author: Ainar Garipov <A.Garipov@AdGuard.COM>
Date:   Wed Nov 9 14:19:54 2022 +0300

    all: move more client code to netip.Addr
2022-11-09 14:37:07 +03:00

549 lines
13 KiB
Go

// Package stats provides units for managing statistics of the filtering DNS
// server.
package stats
import (
"fmt"
"io"
"net"
"net/netip"
"os"
"sync"
"sync/atomic"
"time"
"github.com/AdguardTeam/AdGuardHome/internal/aghhttp"
"github.com/AdguardTeam/golibs/errors"
"github.com/AdguardTeam/golibs/log"
"go.etcd.io/bbolt"
)
// DiskConfig is the configuration structure that is stored in file.
type DiskConfig struct {
// Interval is the number of days for which the statistics are collected
// before flushing to the database.
Interval uint32 `yaml:"statistics_interval"`
}
// checkInterval returns true if days is valid to be used as statistics
// retention interval. The valid values are 0, 1, 7, 30 and 90.
func checkInterval(days uint32) (ok bool) {
return days == 0 || days == 1 || days == 7 || days == 30 || days == 90
}
// Config is the configuration structure for the statistics collecting.
type Config struct {
// UnitID is the function to generate the identifier for current unit. If
// nil, the default function is used, see newUnitID.
UnitID UnitIDGenFunc
// ConfigModified will be called each time the configuration changed via web
// interface.
ConfigModified func()
// HTTPRegister is the function that registers handlers for the stats
// endpoints.
HTTPRegister aghhttp.RegisterFunc
// Filename is the name of the database file.
Filename string
// LimitDays is the maximum number of days to collect statistics into the
// current unit.
LimitDays uint32
}
// Interface is the statistics interface to be used by other packages.
type Interface interface {
// Start begins the statistics collecting.
Start()
io.Closer
// Update collects the incoming statistics data.
Update(e Entry)
// GetTopClientIP returns at most limit IP addresses corresponding to the
// clients with the most number of requests.
TopClientsIP(limit uint) []netip.Addr
// WriteDiskConfig puts the Interface's configuration to the dc.
WriteDiskConfig(dc *DiskConfig)
}
// StatsCtx collects the statistics and flushes it to the database. Its default
// flushing interval is one hour.
//
// TODO(e.burkov): Use atomic.Pointer for accessing db in go1.19.
type StatsCtx struct {
// limitHours is the maximum number of hours to collect statistics into the
// current unit.
//
// It is of type uint32 to be accessed by atomic. It's arranged at the
// beginning of the structure to keep 64-bit alignment.
limitHours uint32
// currMu protects curr.
currMu *sync.RWMutex
// curr is the actual statistics collection result.
curr *unit
// dbMu protects db.
dbMu *sync.Mutex
// db is the opened statistics database, if any.
db *bbolt.DB
// unitIDGen is the function that generates an identifier for the current
// unit. It's here for only testing purposes.
unitIDGen UnitIDGenFunc
// httpRegister is used to set HTTP handlers.
httpRegister aghhttp.RegisterFunc
// configModified is called whenever the configuration is modified via web
// interface.
configModified func()
// filename is the name of database file.
filename string
}
// New creates s from conf and properly initializes it. Don't use s before
// calling it's Start method.
func New(conf Config) (s *StatsCtx, err error) {
defer withRecovered(&err)
s = &StatsCtx{
currMu: &sync.RWMutex{},
dbMu: &sync.Mutex{},
filename: conf.Filename,
configModified: conf.ConfigModified,
httpRegister: conf.HTTPRegister,
}
if s.limitHours = conf.LimitDays * 24; !checkInterval(conf.LimitDays) {
s.limitHours = 24
}
if s.unitIDGen = newUnitID; conf.UnitID != nil {
s.unitIDGen = conf.UnitID
}
// TODO(e.burkov): Move the code below to the Start method.
err = s.openDB()
if err != nil {
return nil, fmt.Errorf("opening database: %w", err)
}
var udb *unitDB
id := s.unitIDGen()
tx, err := s.db.Begin(true)
if err != nil {
return nil, fmt.Errorf("stats: opening a transaction: %w", err)
}
deleted := deleteOldUnits(tx, id-s.limitHours-1)
udb = loadUnitFromDB(tx, id)
err = finishTxn(tx, deleted > 0)
if err != nil {
log.Error("stats: %s", err)
}
s.curr = newUnit(id)
s.curr.deserialize(udb)
log.Debug("stats: initialized")
return s, nil
}
// withRecovered turns the value recovered from panic if any into an error and
// combines it with the one pointed by orig. orig must be non-nil.
func withRecovered(orig *error) {
p := recover()
if p == nil {
return
}
var err error
switch p := p.(type) {
case error:
err = fmt.Errorf("panic: %w", p)
default:
err = fmt.Errorf("panic: recovered value of type %[1]T: %[1]v", p)
}
*orig = errors.WithDeferred(*orig, err)
}
// type check
var _ Interface = (*StatsCtx)(nil)
// Start implements the Interface interface for *StatsCtx.
func (s *StatsCtx) Start() {
s.initWeb()
go s.periodicFlush()
}
// Close implements the io.Closer interface for *StatsCtx.
func (s *StatsCtx) Close() (err error) {
defer func() { err = errors.Annotate(err, "stats: closing: %w") }()
db := s.swapDatabase(nil)
if db == nil {
return nil
}
defer func() {
cerr := db.Close()
if cerr == nil {
log.Debug("stats: database closed")
}
err = errors.WithDeferred(err, cerr)
}()
tx, err := db.Begin(true)
if err != nil {
return fmt.Errorf("opening transaction: %w", err)
}
defer func() { err = errors.WithDeferred(err, finishTxn(tx, err == nil)) }()
s.currMu.RLock()
defer s.currMu.RUnlock()
udb := s.curr.serialize()
return udb.flushUnitToDB(tx, s.curr.id)
}
// Update implements the Interface interface for *StatsCtx.
func (s *StatsCtx) Update(e Entry) {
if atomic.LoadUint32(&s.limitHours) == 0 {
return
}
if e.Result == 0 || e.Result >= resultLast || e.Domain == "" || e.Client == "" {
log.Debug("stats: malformed entry")
return
}
s.currMu.Lock()
defer s.currMu.Unlock()
if s.curr == nil {
log.Error("stats: current unit is nil")
return
}
clientID := e.Client
if ip := net.ParseIP(clientID); ip != nil {
clientID = ip.String()
}
s.curr.add(e.Result, e.Domain, clientID, uint64(e.Time))
}
// WriteDiskConfig implements the Interface interface for *StatsCtx.
func (s *StatsCtx) WriteDiskConfig(dc *DiskConfig) {
dc.Interval = atomic.LoadUint32(&s.limitHours) / 24
}
// TopClientsIP implements the [Interface] interface for *StatsCtx.
func (s *StatsCtx) TopClientsIP(maxCount uint) (ips []netip.Addr) {
limit := atomic.LoadUint32(&s.limitHours)
if limit == 0 {
return nil
}
units, _ := s.loadUnits(limit)
if units == nil {
return nil
}
// Collect data for all the clients to sort and crop it afterwards.
m := map[string]uint64{}
for _, u := range units {
for _, it := range u.Clients {
m[it.Name] += it.Count
}
}
a := convertMapToSlice(m, int(maxCount))
ips = []netip.Addr{}
for _, it := range a {
ip, err := netip.ParseAddr(it.Name)
if err == nil {
ips = append(ips, ip)
}
}
return ips
}
// database returns the database if it's opened. It's safe for concurrent use.
func (s *StatsCtx) database() (db *bbolt.DB) {
s.dbMu.Lock()
defer s.dbMu.Unlock()
return s.db
}
// swapDatabase swaps the database with another one and returns it. It's safe
// for concurrent use.
func (s *StatsCtx) swapDatabase(with *bbolt.DB) (old *bbolt.DB) {
s.dbMu.Lock()
defer s.dbMu.Unlock()
old, s.db = s.db, with
return old
}
// deleteOldUnits walks the buckets available to tx and deletes old units. It
// returns the number of deletions performed.
func deleteOldUnits(tx *bbolt.Tx, firstID uint32) (deleted int) {
log.Debug("stats: deleting old units until id %d", firstID)
// TODO(a.garipov): See if this is actually necessary. Looks like a rather
// bizarre solution.
const errStop errors.Error = "stop iteration"
walk := func(name []byte, _ *bbolt.Bucket) (err error) {
nameID, ok := unitNameToID(name)
if ok && nameID >= firstID {
return errStop
}
err = tx.DeleteBucket(name)
if err != nil {
log.Debug("stats: deleting bucket: %s", err)
return nil
}
log.Debug("stats: deleted unit %d (name %x)", nameID, name)
deleted++
return nil
}
err := tx.ForEach(walk)
if err != nil && !errors.Is(err, errStop) {
log.Debug("stats: deleting units: %s", err)
}
return deleted
}
// openDB returns an error if the database can't be opened from the specified
// file. It's safe for concurrent use.
func (s *StatsCtx) openDB() (err error) {
log.Debug("stats: opening database")
var db *bbolt.DB
db, err = bbolt.Open(s.filename, 0o644, nil)
if err != nil {
if err.Error() == "invalid argument" {
log.Error("AdGuard Home cannot be initialized due to an incompatible file system.\nPlease read the explanation here: https://github.com/AdguardTeam/AdGuardHome/wiki/Getting-Started#limitations")
}
return err
}
// Use defer to unlock the mutex as soon as possible.
defer log.Debug("stats: database opened")
s.dbMu.Lock()
defer s.dbMu.Unlock()
s.db = db
return nil
}
func (s *StatsCtx) flush() (cont bool, sleepFor time.Duration) {
id := s.unitIDGen()
s.currMu.Lock()
defer s.currMu.Unlock()
ptr := s.curr
if ptr == nil {
return false, 0
}
limit := atomic.LoadUint32(&s.limitHours)
if limit == 0 || ptr.id == id {
return true, time.Second
}
db := s.database()
if db == nil {
return true, 0
}
isCommitable := true
tx, err := db.Begin(true)
if err != nil {
log.Error("stats: opening transaction: %s", err)
return true, 0
}
defer func() {
if err = finishTxn(tx, isCommitable); err != nil {
log.Error("stats: %s", err)
}
}()
s.curr = newUnit(id)
flushErr := ptr.serialize().flushUnitToDB(tx, ptr.id)
if flushErr != nil {
log.Error("stats: flushing unit: %s", flushErr)
isCommitable = false
}
delErr := tx.DeleteBucket(idToUnitName(id - limit))
if delErr != nil {
// TODO(e.burkov): Improve the algorithm of deleting the oldest bucket
// to avoid the error.
if errors.Is(delErr, bbolt.ErrBucketNotFound) {
log.Debug("stats: warning: deleting unit: %s", delErr)
} else {
isCommitable = false
log.Error("stats: deleting unit: %s", delErr)
}
}
return true, 0
}
// periodicFlush checks and flushes the unit to the database if the freshly
// generated unit ID differs from the current's ID. Flushing process includes:
// - swapping the current unit with the new empty one;
// - writing the current unit to the database;
// - removing the stale unit from the database.
func (s *StatsCtx) periodicFlush() {
for cont, sleepFor := true, time.Duration(0); cont; time.Sleep(sleepFor) {
cont, sleepFor = s.flush()
}
log.Debug("periodic flushing finished")
}
func (s *StatsCtx) setLimit(limitDays int) {
atomic.StoreUint32(&s.limitHours, uint32(24*limitDays))
if limitDays == 0 {
if err := s.clear(); err != nil {
log.Error("stats: %s", err)
}
}
log.Debug("stats: set limit: %d days", limitDays)
}
// Reset counters and clear database
func (s *StatsCtx) clear() (err error) {
defer func() { err = errors.Annotate(err, "clearing: %w") }()
db := s.swapDatabase(nil)
if db != nil {
var tx *bbolt.Tx
tx, err = db.Begin(true)
if err != nil {
log.Error("stats: opening a transaction: %s", err)
} else if err = finishTxn(tx, false); err != nil {
// Don't wrap the error since it's informative enough as is.
return err
}
// Active transactions will continue using database, but new ones won't
// be created.
err = db.Close()
if err != nil {
return fmt.Errorf("closing database: %w", err)
}
// All active transactions are now closed.
log.Debug("stats: database closed")
}
err = os.Remove(s.filename)
if err != nil {
log.Error("stats: %s", err)
}
err = s.openDB()
if err != nil {
log.Error("stats: opening database: %s", err)
}
// Use defer to unlock the mutex as soon as possible.
defer log.Debug("stats: cleared")
s.currMu.Lock()
defer s.currMu.Unlock()
s.curr = newUnit(s.unitIDGen())
return nil
}
func (s *StatsCtx) loadUnits(limit uint32) (units []*unitDB, firstID uint32) {
db := s.database()
if db == nil {
return nil, 0
}
// Use writable transaction to ensure any ongoing writable transaction is
// taken into account.
tx, err := db.Begin(true)
if err != nil {
log.Error("stats: opening transaction: %s", err)
return nil, 0
}
s.currMu.RLock()
defer s.currMu.RUnlock()
cur := s.curr
var curID uint32
if cur != nil {
curID = cur.id
} else {
curID = s.unitIDGen()
}
// Per-hour units.
units = make([]*unitDB, 0, limit)
firstID = curID - limit + 1
for i := firstID; i != curID; i++ {
u := loadUnitFromDB(tx, i)
if u == nil {
u = &unitDB{NResult: make([]uint64, resultLast)}
}
units = append(units, u)
}
err = finishTxn(tx, false)
if err != nil {
log.Error("stats: %s", err)
}
if cur != nil {
units = append(units, cur.serialize())
}
if unitsLen := len(units); unitsLen != int(limit) {
log.Fatalf("loaded %d units whilst the desired number is %d", unitsLen, limit)
}
return units, firstID
}