Merge pull request #316 from TrekkieCoder/main

PR - Updated default liveness check policy

loxinet/loxinet.go

@@ -67,6 +67,7 @@ type loxiNetH struct {
 	wg     sync.WaitGroup
 	bgp    *GoBgpH
 	sumDis bool
+	pProbe bool
 	has    *CIStateH
 	logger *tk.Logger
 	ready  bool
@@ -181,6 +182,7 @@ func loxiNetInit() {
 
 	mh.self = opts.Opts.ClusterSelf
 	mh.sumDis = opts.Opts.CSumDisable
+	mh.pProbe = opts.Opts.PassiveEPProbe
 	mh.sigCh = make(chan os.Signal, 5)
 	signal.Notify(mh.sigCh, os.Interrupt, syscall.SIGCHLD)
 	signal.Notify(mh.sigCh, os.Interrupt, syscall.SIGHUP)

loxinet/rules.go

@@ -181,6 +181,7 @@ type epHostOpts struct {
 	probeDuration     uint32
 	currProbeDuration uint32
 	probePort         uint16
+	probeActivated    bool
 }
 
 type epHost struct {
@@ -770,7 +771,7 @@ func validateXlateEPWeights(servEndPoints []cmn.LbEndPointArg) (int, error) {
 	return 0, nil
 }
 
-func (R *RuleH) modNatEpHost(r *ruleEnt, endpoints []ruleNatEp, doAddOp bool) {
+func (R *RuleH) modNatEpHost(r *ruleEnt, endpoints []ruleNatEp, doAddOp bool, liveCheckEn bool) {
 	var hopts epHostOpts
 	hopts.inActTryThr = DflLbaInactiveTries
 	hopts.probeDuration = DflHostProbeTimeout
@@ -790,6 +791,10 @@ func (R *RuleH) modNatEpHost(r *ruleEnt, endpoints []ruleNatEp, doAddOp bool) {
 			hopts.probeType = HostProbePing
 		}
 
+		if mh.pProbe == true || liveCheckEn {
+			hopts.probeActivated = true
+		}
+
 		epKey := makeEPKey(nep.xIP.String(), hopts.probeType, hopts.probePort)
 
 		if doAddOp {
@@ -977,6 +982,8 @@ func (R *RuleH) AddNatLbRule(serv cmn.LbServiceArg, servSecIPs []cmn.LbSecIpArg,
 		return RuleArgsErr, errors.New("secondaryIP-args len error")
 	}
 
+	activateProbe := false
+
 	for _, k := range servSecIPs {
 		pNetAddr := net.ParseIP(k.SecIP)
 		if pNetAddr == nil {
@@ -998,6 +1005,10 @@ func (R *RuleH) AddNatLbRule(serv cmn.LbServiceArg, servSecIPs []cmn.LbSecIpArg,
 	natActs.sel = serv.Sel
 	natActs.mode = cmn.LBMode(serv.Mode)
 
+	if natActs.mode == cmn.LBModeOneArm || serv.Monitor {
+		activateProbe = true
+	}
+
 	for _, k := range servEndPoints {
 		pNetAddr := net.ParseIP(k.EpIP)
 		if pNetAddr == nil {
@@ -1084,7 +1095,7 @@ func (R *RuleH) AddNatLbRule(serv cmn.LbServiceArg, servSecIPs []cmn.LbSecIpArg,
 		eRule.act.action.(*ruleNatActs).endPoints = eEps
 		eRule.act.action.(*ruleNatActs).mode = natActs.mode
 
-		R.modNatEpHost(eRule, eEps, true)
+		R.modNatEpHost(eRule, eEps, true, activateProbe)
 
 		eRule.sT = time.Now()
 		eRule.iTo = serv.InactiveTimeout
@@ -1118,7 +1129,7 @@ func (R *RuleH) AddNatLbRule(serv cmn.LbServiceArg, servSecIPs []cmn.LbSecIpArg,
 	r.BGP = serv.Bgp
 	r.CI = cmn.CIDefault
 
-	R.modNatEpHost(r, natActs.endPoints, true)
+	R.modNatEpHost(r, natActs.endPoints, true, activateProbe)
 
 	tk.LogIt(tk.LogDebug, "nat lb-rule added - %d:%s-%s\n", r.ruleNum, r.tuples.String(), r.act.String())
 
@@ -1174,7 +1185,11 @@ func (R *RuleH) DeleteNatLbRule(serv cmn.LbServiceArg) (int, error) {
 	defer R.Tables[RtLB].Mark.PutCounter(rule.ruleNum)
 
 	eEps := rule.act.action.(*ruleNatActs).endPoints
-	R.modNatEpHost(rule, eEps, false)
+	activatedProbe := false
+	if rule.act.action.(*ruleNatActs).mode == cmn.LBModeOneArm || rule.ActChk {
+		activatedProbe = true
+	}
+	R.modNatEpHost(rule, eEps, false, activatedProbe)
 
 	delete(R.Tables[RtLB].eMap, rt.ruleKey())
 	if rule.ruleNum < RtMaximumLbs {
@@ -1408,9 +1423,13 @@ func (R *RuleH) GetEpHosts() ([]cmn.EndPointMod, error) {
 		// Make end-point
 		ret.HostName = data.hostName
 		ret.Name = data.epKey
-		ret.InActTries = data.opts.inActTryThr
-		ret.ProbeType = data.opts.probeType
-		ret.ProbeDuration = data.opts.probeDuration
+		if !data.opts.probeActivated {
+			ret.ProbeType = HostProbeNone
+		} else {
+			ret.ProbeType = data.opts.probeType
+			ret.ProbeDuration = data.opts.probeDuration
+			ret.InActTries = data.opts.inActTryThr
+		}
 		ret.ProbeReq = data.opts.probeReq
 		ret.ProbeResp = data.opts.probeResp
 		ret.ProbePort = data.opts.probePort
@@ -1483,6 +1502,10 @@ func makeEPKey(hostName string, probeType string, probePort uint16) string {
 func (R *RuleH) AddEPHost(apiCall bool, hostName string, name string, args epHostOpts) (int, error) {
 	var epKey string
 
+	if apiCall && args.probeType != HostProbeNone {
+		args.probeActivated = true
+	}
+
 	R.epMx.Lock()
 	defer R.epMx.Unlock()
 
@@ -1624,6 +1647,12 @@ func (R *RuleH) epCheckNow(ep *epHost) {
 		sName = fmt.Sprintf("[%s]:%d", ep.hostName, ep.opts.probePort)
 	}
 
+	if !ep.opts.probeActivated {
+		ep.inactive = false
+		ep.inActTries = 0
+		return
+	}
+
 	if ep.opts.probeType == HostProbeConnectTcp ||
 		ep.opts.probeType == HostProbeConnectUdp ||
 		ep.opts.probeType == HostProbeConnectSctp {
@@ -1739,7 +1768,7 @@ func epTicker(R *RuleH, helper int) {
 			if idx > 0 {
 				idx = 0
 				// We restart the sweep from beginning while taking a short break
-				// Due to how goLang range, works we would be sweeping eps mostly randomly
+				// Due to how goLang range works, we would be sweeping eps mostly randomly
 				R.epMx.Unlock()
 				break
 			}

options/options.go

@@ -22,4 +22,5 @@ var Opts struct {
 	CPUProfile        string         `long:"cpuprofile" description:"Enable cpu profiling and specify file to use" default:"none" env:"CPUPROF"`
 	NoPrometheus      bool           `short:"p" long:"nopro" description:" Do not run prometheus thread"`
 	CSumDisable       bool           `long:"disable-csum" description:"Disable checksum update(experimental)"`
+	PassiveEPProbe    bool           `long:"passive-probe" description:"Enable passive liveness probes(experimental)"`
 }