// Open a DB by fileName and options.
func Open(fileName string) (*DB, error) {
// Create if not exist
_, err := os.Stat(fileName)
if os.IsNotExist(err) {
log.Debug("create dir %s", fileName)
err := os.Mkdir(fileName, filemode)
if err != nil {
return nil, err
}
}
// Admindb.
db := new(DB)
db.Admin, err = admindb.Open(path.Join(fileName, admindbFileName))
if err != nil {
return nil, err
}
// Indexdb.
db.Index, err = indexdb.Open(path.Join(fileName, indexdbFileName))
if err != nil {
return nil, err
}
// Metricdb.
db.Metric, err = metricdb.Open(path.Join(fileName, metricdbFileName))
if err != nil {
return nil, err
}
log.Debug("storage is opened successfully")
return db, nil
}
// Match a metric with rules, and return matched rules.
//
// If no rules matched, return false.
// If any black patterns matched, return false.
// Else, return true and matched rules.
//
func (d *Detector) match(m *models.Metric) (bool, []*models.Rule) {
// Check rules.
timer := util.NewTimer()
rules := d.flt.MatchedRules(m)
elapsed := timer.Elapsed()
health.AddFilterCost(elapsed)
if len(rules) == 0 {
// Hit no rules.
return false, rules
}
// Check blacklist.
for _, p := range d.cfg.Detector.BlackList {
ok, err := filepath.Match(p, m.Name)
if err != nil {
// Invalid black pattern.
log.Error("invalid black pattern: %s, %v", p, err)
continue
}
if ok {
// Hit black pattern.
log.Debug("%s hit black pattern %s", m.Name, p)
return false, rules
}
}
// Ok
return true, rules
}
// migrate db schema.
func (db *DB) migrate() error {
log.Debug("migrate sql schemas..")
rule := &models.Rule{}
user := &models.User{}
proj := &models.Project{}
return db.db.AutoMigrate(rule, user, proj).Error
}
// Test whether a metric matches the rules.
func (d *Detector) match(m *models.Metric) bool {
// Check rules.
rules := d.filter.MatchedRules(m)
if len(rules) == 0 {
log.Debug("%s hit no rules", m.Name)
return false
}
// Check blacklist.
for _, pattern := range d.cfg.Detector.BlackList {
ok, err := filepath.Match(pattern, m.Name)
if err == nil && ok {
log.Debug("%s hit black pattern %s", m.Name, pattern)
return false
}
}
return true
}
func initLog() {
log.SetName("banshee")
if *debug {
log.SetLevel(log.DEBUG)
}
goVs := runtime.Version()
nCPU := runtime.GOMAXPROCS(-1)
vers := version.Version
log.Debug("banshee%s %s %d cpu", vers, goVs, nCPU)
}
// UpdateWithJSONFile update the config from a json file.
func (c *Config) UpdateWithJSONFile(fileName string) error {
log.Debug("read config from %s..", fileName)
b, err := ioutil.ReadFile(fileName)
if err != nil {
return err
}
err = json.Unmarshal(b, c)
if err != nil {
return err
}
return err
}
func main() {
// Arguments
fileName := flag.String("c", "config.json", "config file")
debug := flag.Bool("d", false, "debug mode")
vers := flag.Bool("v", false, "version")
flag.Parse()
// Version
if *vers {
fmt.Fprintln(os.Stdout, version.Version)
os.Exit(1)
}
// Logging
log.SetName("banshee")
if *debug {
log.SetLevel(log.DEBUG)
}
log.Debug("using %s, max %d cpu", runtime.Version(), runtime.GOMAXPROCS(-1))
// Config
cfg := config.New()
if flag.NFlag() == 0 || (flag.NFlag() == 1 && *debug == true) {
log.Warn("no config file specified, using default..")
} else {
err := cfg.UpdateWithJSONFile(*fileName)
if err != nil {
log.Fatal("failed to load %s, %s", *fileName, err)
}
}
// Storage
options := &storage.Options{
NumGrid: cfg.Period[0],
GridLen: cfg.Period[1],
}
db, err := storage.Open(cfg.Storage.Path, options)
if err != nil {
log.Fatal("failed to open %s: %v", cfg.Storage.Path, err)
}
// Cleaner
cleaner := cleaner.New(db, cfg.Period[0]*cfg.Period[1])
go cleaner.Start()
// Filter
filter := filter.New()
filter.Init(db)
// Alerter
alerter := alerter.New(cfg, db, filter)
alerter.Start()
// Webapp
go webapp.Start(cfg, db)
// Detector
detector := detector.New(cfg, db, filter)
detector.Out(alerter.In)
detector.Start()
}
// Init from db.
func (f *Filter) Init(db *storage.DB) {
log.Debug("init filter's rules from cache..")
// Listen rules changes.
db.Admin.RulesCache.OnAdd(f.addRuleCh)
db.Admin.RulesCache.OnDel(f.delRuleCh)
go f.addRules()
go f.delRules()
// Add rules from cache
rules := db.Admin.RulesCache.All()
for _, rule := range rules {
f.addRule(rule)
}
}
// Open a DB by fileName and options.
func Open(fileName string, options *Options) (*DB, error) {
// Create if not exist
_, err := os.Stat(fileName)
if os.IsNotExist(err) {
log.Debug("create dir %s", fileName)
err := os.Mkdir(fileName, filemode)
if err != nil {
return nil, err
}
}
// Admindb.
db := new(DB)
db.Admin, err = admindb.Open(path.Join(fileName, admindbFileName))
if err != nil {
return nil, err
}
// Indexdb.
db.Index, err = indexdb.Open(path.Join(fileName, indexdbFileName))
if err != nil {
return nil, err
}
// Metricdb.
db.Metric, err = metricdb.Open(path.Join(fileName, metricdbFileName))
if err != nil {
return nil, err
}
name := fmt.Sprintf("%s-%dx%d", statedbFileName, options.NumGrid, options.GridLen)
opts := &statedb.Options{NumGrid: options.NumGrid, GridLen: options.GridLen}
// Statedb.
db.State, err = statedb.Open(path.Join(fileName, name), opts)
if err != nil {
return nil, err
}
log.Debug("storage is opened successfully")
return db, nil
}
// Init cache from db.
func (c *rulesCache) Init(db *gorm.DB) error {
log.Debug("init rules from admindb..")
// Query
var rules []models.Rule
err := db.Find(&rules).Error
if err != nil {
return err
}
// Load
for _, rule := range rules {
// Share
r := &rule
r.Share()
c.rules.Set(rule.ID, r)
}
return nil
}
// Handle a connection, it will filter the mertics by rules and detect whether
// the metrics are anomalies.
func (d *Detector) handle(conn net.Conn) {
// New conn
addr := conn.RemoteAddr()
defer func() {
conn.Close()
log.Info("conn %s disconnected", addr)
}()
log.Info("conn %s established", addr)
// Scan line by line.
scanner := bufio.NewScanner(conn)
for scanner.Scan() {
if err := scanner.Err(); err != nil {
log.Info("read conn: %v, closing it..", err)
break
}
startAt := time.Now()
// Parse
line := scanner.Text()
m, err := parseMetric(line)
if err != nil {
if len(line) > 10 {
line = line[:10]
}
log.Error("parse '%s': %v, skipping..", line, err)
continue
}
// Filter
if d.match(m) {
// Detect
err = d.detect(m)
if err != nil {
log.Error("failed to detect: %v, skipping..", err)
continue
}
elapsed := time.Since(startAt)
log.Debug("%dμs %s %.3f", elapsed.Nanoseconds()/1000, m.Name, m.Score)
// Output
d.output(m)
// Store
if err := d.store(m); err != nil {
log.Error("store metric %s: %v, skiping..", m.Name, err)
}
}
}
}
// load indexes from db to cache.
func (db *DB) load() {
log.Debug("init index from indexdb..")
// Scan values to memory.
iter := db.db.NewIterator(nil, nil)
for iter.Next() {
// Decode
key := iter.Key()
value := iter.Value()
idx := &models.Index{}
idx.Name = string(key)
err := decode(value, idx)
if err != nil {
// Skip corrupted values
continue
}
db.m.Set(idx.Name, idx)
}
}
// clean checks all indexes and do cleaning.
func (c *Cleaner) clean() {
idxs := c.db.Index.All()
// Use local server time and uint32 is enough for further 90 years
now := uint32(time.Now().Unix())
for _, idx := range idxs {
if idx.Stamp+c.cfg.Cleaner.Threshold < now {
// Long time no data, clean all.
c.db.Index.Delete(idx.Name)
c.db.Metric.DeleteTo(idx.Name, idx.Stamp+1) // DeleteTo is right closed
log.Info("%s fully cleaned", idx.Name)
} else {
// Clean outdated metrics.
n, _ := c.db.Metric.DeleteTo(idx.Name, now-c.cfg.Expiration)
if n > 0 {
log.Debug("%s %d outdated metrics cleaned", idx.Name, n)
}
}
}
}