// handle a new connection:
// Steps:
// 1. Read input from connection line by line.
// 2. Parse each line into a metric.
// 3. Validate the metric
// 4. Process the metric.
func (d *Detector) handle(conn net.Conn) {
addr := conn.RemoteAddr()
health.IncrNumClients(1)
defer health.DecrNumClients(1)
log.Infof("conn %s established", addr)
scanner := bufio.NewScanner(conn)
for scanner.Scan() { // Read line by line.
if err := scanner.Err(); err != nil { // Close on read error.
log.Errorf("read error: %v, closing conn..", err)
break
}
line := scanner.Text()
m, err := parseMetric(line) // Parse
if err != nil {
log.Errorf("parse error: %v, skipping..", err)
continue
}
if err = m.Validate(); err != nil {
log.Errorf("invalid metric: %v, skipping..", err)
return
}
d.process(m, true)
}
conn.Close()
log.Infof("conn %s disconnected", addr)
}
// match a metric with rules, and return matched rules.
// Details:
// 1. If no rules matched, return false.
// 2. If any black patterns matched, return false.
// 3. Else, return true and matched rules.
func (d *Detector) match(m *models.Metric) (bool, []*models.Rule) {
// Check rules.
timer := util.NewTimer() // Filter timer
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.Errorf("invalid black pattern: %s, %v", p, err)
continue
}
if ok {
// Hit black pattern.
log.Debugf("%s hit black pattern %s", m.Name, p)
return false, rules
}
}
return true, rules // OK
}
// process the input metric.
// Steps:
// 1. Match metric with all rules.
// 2. Detect the metric with matched rules.
// 3. Output detection results to receivers.
func (d *Detector) process(m *models.Metric, shouldAdjustIdle bool) {
health.IncrNumMetricIncomed(1)
timer := util.NewTimer() // Detection cost timer
// Match
ok, rules := d.match(m)
if !ok {
return
}
if shouldAdjustIdle {
d.adjustIdleM(m, rules)
}
// Detect
evs, err := d.detect(m, rules)
if err != nil {
log.Errorf("detect: %v, skipping..", err)
return
}
health.IncrNumMetricDetected(1)
// Output
for _, ev := range evs {
d.output(ev)
}
// Time end.
elapsed := timer.Elapsed()
if elapsed > float64(d.cfg.Detector.WarningTimeout) {
log.Warnf("detection is slow: %.2fms", elapsed)
}
health.AddDetectionCost(elapsed)
}
// work waits for events to alert.
func (al *Alerter) work() {
for {
ev := <-al.In
ew := models.NewWrapperOfEvent(ev) // Avoid locks
if al.checkAlertAt(ew.Metric) { // Check alert interval
continue
}
if al.checkOneDayAlerts(ew.Metric) { // Check one day limit
continue
}
// Avoid noises by issuing alerts only when same alert has occurred
// predefined times.
if al.checkAlertCount(ew.Metric) {
al.setAlertRecord(ew.Metric)
log.Warnf("Not enough alerts with in `AlertCheckInterval` time skipping..: %v", ew.Metric.Name)
continue
}
al.setAlertRecord(ew.Metric)
al.incrAlertNum(ew.Metric)
// Store event
if err := al.storeEvent(ev); err != nil {
log.Warnf("failed to store event:%v, skipping..", err)
continue
}
// Do alert.
var err error
if ew.Project, err = al.getProjByRule(ew.Rule); err != nil {
continue
}
if al.shoudProjBeSilent(ew.Project) {
continue
}
var users []models.User
if users, err = al.getUsersByProj(ew.Project); err != nil {
continue
}
for _, user := range users {
ew.User = &user
if ew.Rule.Level < user.RuleLevel {
continue
}
if len(al.cfg.Alerter.Command) == 0 {
log.Warnf("alert command not configured")
continue
}
if err = al.execCommand(ew); err != nil { // Execute command
log.Errorf("exec %s: %v", al.cfg.Alerter.Command, err)
continue
}
log.Infof("send to %s with %s ok", user.Name, ew.Metric.Name)
}
if len(users) != 0 {
al.setAlertAt(ew.Metric)
health.IncrNumAlertingEvents(1)
}
}
}
// pushDeled pushes changed rule to listeners.
func (c *rulesCache) pushDeled(rule *models.Rule) {
for _, ch := range c.lnsDel {
select {
case ch <- rule:
default:
log.Errorf("buffered deleted rules chan is full, skipping..")
}
}
}
// output detected metrics to all chans in outs.
// Skip if the target chan is full.
func (d *Detector) output(ev *models.Event) {
for _, ch := range d.outs {
select {
case ch <- ev:
default:
log.Errorf("output channel is full, skipping..")
continue
}
}
}
// Start the tcp server.
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.Fatalf("listen: %v", err)
}
log.Infof("detector is listening on %s", addr)
go d.startIdleTracking()
for {
conn, err := ln.Accept()
if err != nil {
log.Errorf("cannot accept conn: %v, skipping..", err)
continue
}
go d.handle(conn)
}
}
// shouldFill0 returns true if given metric needs to fill blanks with zeros to
// its hidtory values.
// A metric should fill0 if it matches configured fill blank zero patterns and
// the matching rules have no option NeverFillZero set.
func (d *Detector) shouldFill0(m *models.Metric, rules []*models.Rule) bool {
for _, p := range d.cfg.Detector.FillBlankZeros {
ok, err := filepath.Match(p, m.Name)
if err != nil {
// Invalid pattern.
log.Errorf("invalid fillBlankZeros pattern: %s, %v", p, err)
continue
}
if ok {
// Matched the fill zeros patterns, then check its rules.
for _, rule := range rules {
if rule.NeverFillZero {
return false
}
}
return true // OK
}
}
return false
}
// shouldTrackIdle returns true if given metric should be tracked for idle.
// A metric should be tracked for idle states if it matches configured check
// pattern list or its matched rules have an option TrackIdle set.
func (d *Detector) shoudTrackIdle(m *models.Metric, rules []*models.Rule) bool {
for _, rule := range rules {
if rule.TrackIdle {
return true
}
}
isHighLevel := false
for _, rule := range rules { // IdleMetricCheckList only works for high level rules.
if rule.Level == models.RuleLevelHigh {
isHighLevel = true
}
}
for _, p := range d.cfg.Detector.IdleMetricCheckList {
ok, err := filepath.Match(p, m.Name)
if err != nil {
log.Errorf("invalid idleMetricCheck pattern: %s, %v", p, err)
continue
}
if ok && isHighLevel {
return true
}
}
return false
}