// Handle a new connection, it will:
//
// 1. Read input from the connection line by line.
// 2. Parse the lines into metrics.
// 3. Validate the metrics.
//
func (d *Detector) handle(conn net.Conn) {
// New conn established.
addr := conn.RemoteAddr()
health.IncrNumClients(1)
log.Info("conn %s established", addr)
// Read
scanner := bufio.NewScanner(conn)
for scanner.Scan() {
// Read line by line.
if err := scanner.Err(); err != nil {
// Close conn on read error.
log.Error("read error: %v, closing conn..", err)
break
}
line := scanner.Text()
// Parse metric.
m, err := parseMetric(line)
if err != nil {
// Skip invalid input.
log.Error("parse error: %v, skipping..", err)
continue
}
// Validate metric.
if err := validateMetric(m); err != nil {
log.Error("invalid metric: %v, skipping..", err)
continue
}
// Process
d.process(m)
}
// Close conn.
conn.Close()
log.Info("conn %s disconnected", addr)
health.DecrNumClients(1)
}
// Start http server.
func Start(c *config.Config, d *storage.DB) {
// Init globals.
cfg = c
db = d
// Auth
auth := newAuthHandler(cfg.Webapp.Auth[0], cfg.Webapp.Auth[1])
// Routes
router := httprouter.New()
// Api
router.GET("/api/config", auth.handler(getConfig))
router.GET("/api/projects", getProjects)
router.GET("/api/project/:id", getProject)
router.POST("/api/project", auth.handler(createProject))
router.PATCH("/api/project/:id", auth.handler(updateProject))
router.DELETE("/api/projects/:id", auth.handler(deleteProject))
router.GET("/api/project/:id/rules", auth.handler(getProjectRules))
router.GET("/api/project/:id/users", auth.handler(getProjectUsers))
router.POST("/api/project/:id/user", auth.handler(addProjectUser))
router.DELETE("/api/project/:id/user/:user_id", auth.handler(deleteProjectUser))
router.GET("/api/users", auth.handler(getUsers))
router.GET("/api/user/:id", auth.handler(getUser))
router.POST("/api/user", auth.handler(createUser))
router.DELETE("/api/user/:id", auth.handler(deleteUser))
router.PATCH("/api/user/:id", auth.handler(updateUser))
router.POST("/api/rule", auth.handler(createRule))
router.DELETE("/api/rule/:id", auth.handler(deleteRule))
router.GET("/api/metric/indexes", getMetricIndexes)
router.GET("/api/metric/data/:name/:start/:stop", getMetrics)
// Static
router.NotFound = newStaticHandler(http.Dir(cfg.Webapp.Static), auth)
// Serve
addr := fmt.Sprintf("0.0.0.0:%d", cfg.Webapp.Port)
log.Info("webapp is listening and serving on %s..", addr)
http.ListenAndServe(addr, router)
}
// 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)
}
}
}
}
// Start a time ticker and wait to check.
func (c *Cleaner) Start() {
log.Info("start cleaner with interval %.3fs..", c.interval.Seconds())
// Check right now.
c.clean()
for {
// And wait for another interval to check.
<-c.ticker.C
c.clean()
}
}
// 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.Info("%s %d outdated metrics cleaned", idx.Name, n)
}
}
}
}
// Start a time ticker to clean.
func (c *Cleaner) Start() {
log.Info("start cleaner..")
// Clean right now.
c.clean()
// Clean each interval.
ticker := time.NewTicker(time.Duration(c.cfg.Cleaner.Interval) * time.Second)
for {
<-ticker.C
c.clean()
}
}
// Start several goroutines to wait for detected metrics, then check each
// metric with all the rules, the configured shell command will be executed
// once a rule is hit.
func (al *Alerter) Start() {
log.Info("start %d alerter workers..", al.cfg.Alerter.Workers)
for i := 0; i < al.cfg.Alerter.Workers; i++ {
go al.work()
}
go func() {
ticker := time.NewTicker(time.Hour * 24)
for _ = range ticker.C {
al.c.Clear()
}
}()
}
// clean checks all indexes for outdated metrics, states and clean them.
func (c *Cleaner) clean() {
idxs := c.db.Index.All()
now := time.Now()
for _, idx := range idxs {
t := time.Unix(int64(idx.Stamp), 0)
if t.Add(c.expiration).Before(now) {
// Clean outdated.
c.db.State.Delete(idx.Name)
c.db.Metric.DeleteTo(idx.Name, uint32(now.Unix()))
c.db.Index.Delete(idx.Name)
log.Info("%s cleaned", idx.Name)
}
}
}
// Start detector.
func (d *Detector) Start() {
addr := fmt.Sprintf("0.0.0.0:%d", d.cfg.Detector.Port)
ln, err := net.Listen("tcp", addr)
if err != nil {
log.Fatal("failed to bind tcp://%s: %v", addr, err)
}
log.Info("detector is listening on tcp://%s..", addr)
for {
conn, err := ln.Accept()
if err != nil {
log.Fatal("accept conn: %v", err)
}
go d.handle(conn)
}
}
// Start the tcp server.
func (d *Detector) Start() {
// Listen
addr := fmt.Sprintf("0.0.0.0:%d", d.cfg.Detector.Port)
ln, err := net.Listen("tcp", addr)
if err != nil {
log.Fatal("listen: %v", err)
}
log.Info("detector is listening on %s..", addr)
// Accept
for {
conn, err := ln.Accept()
if err != nil {
log.Error("cannot accept conn: %v, skipping..", err)
continue
}
go d.handle(conn)
}
}
// Start several goroutines to wait for detected metrics, then check each
// metric with all the rules, the configured shell command will be executed
// once a rule is hit.
func (al *Alerter) Start() {
log.Info("start %d alerter workers..", al.cfg.Alerter.Workers)
for i := 0; i < al.cfg.Alerter.Workers; i++ {
go al.work()
}
}
// work waits for detected metrics, then check each metric with all the
// rules, the configured shell command will be executed once a rule is hit.
func (al *Alerter) work() {
for {
ev := <-al.In
// Check interval.
v, ok := al.m.Get(ev.Metric.Name)
if ok && ev.Metric.Stamp-v.(uint32) < al.cfg.Alerter.Interval {
continue
}
// Check alert times in one day
v, ok = al.c.Get(ev.Metric.Name)
if ok && atomic.LoadUint32(v.(*uint32)) > al.cfg.Alerter.OneDayLimit {
log.Warn("%s hit alerting one day limit, skipping..", ev.Metric.Name)
continue
}
if !ok {
var newCounter uint32
newCounter = 1
al.c.Set(ev.Metric.Name, &newCounter)
} else {
atomic.AddUint32(v.(*uint32), 1)
}
// Universals
var univs []models.User
if err := al.db.Admin.DB().Where("universal = ?", true).Find(&univs).Error; err != nil {
log.Error("get universal users: %v, skiping..", err)
continue
}
for _, rule := range ev.Metric.TestedRules {
ev.Rule = rule
ev.TranslateRuleComment()
// Project
proj := &models.Project{}
if err := al.db.Admin.DB().Model(rule).Related(proj).Error; err != nil {
log.Error("project, %v, skiping..", err)
continue
}
ev.Project = proj
// Silent
if al.shouldSilent(proj) {
continue
}
// Users
var users []models.User
if err := al.db.Admin.DB().Model(proj).Related(&users, "Users").Error; err != nil {
log.Error("get users: %v, skiping..", err)
continue
}
users = append(users, univs...)
// Send
for _, user := range users {
ev.User = &user
if rule.Level < user.RuleLevel {
continue
}
// Exec
if len(al.cfg.Alerter.Command) == 0 {
log.Warn("alert command not configured")
continue
}
if err := al.execCommand(ev); err != nil {
log.Error("exec %s: %v", al.cfg.Alerter.Command, err)
continue
}
log.Info("send message to %s with %s ok", user.Name, ev.Metric.Name)
}
if len(users) != 0 {
al.m.Set(ev.Metric.Name, ev.Metric.Stamp)
health.IncrNumAlertingEvents(1)
}
}
}
}
// work waits for detected metrics, then check each metric with all the
// rules, the configured shell command will be executed once a rule is hit.
func (al *Alerter) work() {
for {
metric := <-al.In
// Check interval.
v, ok := al.m.Get(metric.Name)
if ok && metric.Stamp-v.(uint32) < al.cfg.Alerter.Interval {
continue
}
// Check alert times in one day
v, ok = al.c.Get(metric.Name)
if ok && atomic.LoadUint32(v.(*uint32)) > al.cfg.Alerter.OneDayLimit {
log.Warn("%s hit alerting one day limit, skipping..", metric.Name)
continue
}
if !ok {
var newCounter uint32
newCounter = 1
al.c.Set(metric.Name, &newCounter)
} else {
atomic.AddUint32(v.(*uint32), 1)
}
// Universals
var univs []models.User
if err := al.db.Admin.DB().Where("universal = ?", true).Find(&univs).Error; err != nil {
log.Error("get universal users: %v, skiping..", err)
continue
}
for _, rule := range metric.TestedRules {
// Project
proj := &models.Project{}
if err := al.db.Admin.DB().Model(rule).Related(proj).Error; err != nil {
log.Error("project, %v, skiping..", err)
continue
}
// Users
var users []models.User
if err := al.db.Admin.DB().Model(proj).Related(&users, "Users").Error; err != nil {
log.Error("get users: %v, skiping..", err)
continue
}
users = append(users, univs...)
// Send
for _, user := range users {
d := &msg{
Project: proj,
Metric: metric,
User: &user,
Rule: rule,
}
// Exec
if len(al.cfg.Alerter.Command) == 0 {
log.Warn("alert command not configured")
continue
}
b, _ := json.Marshal(d)
cmd := exec.Command(al.cfg.Alerter.Command, string(b))
if err := cmd.Run(); err != nil {
log.Error("exec %s: %v", al.cfg.Alerter.Command, err)
continue
}
log.Info("send message to %s with %s ok", user.Name, metric.Name)
}
if len(users) != 0 {
al.m.Set(metric.Name, metric.Stamp)
health.IncrNumAlertingEvents(1)
}
}
}
}
