Merge pull request #1645 from capone212/failing_cehcks_fix

issue #1636:  start CheckRunner if check exists in consul already

command/agent/consul/check.go

@@ -47,6 +47,13 @@ func (r *CheckRunner) Start() {
 	r.started = true
 }
 
+// Started returns if the check runner has started running
+func (r *CheckRunner) Started() bool {
+	r.startedLock.Lock()
+	defer r.startedLock.Unlock()
+	return r.started
+}
+
 // Stop is used to stop the check.
 func (r *CheckRunner) Stop() {
 	r.stopLock.Lock()

command/agent/consul/syncer.go

@@ -394,7 +394,7 @@ func (c *Syncer) syncChecks() error {
 	}
 
 	// Synchronize checks with Consul
-	missingChecks, _, changedChecks, staleChecks := c.calcChecksDiff(consulChecks)
+	missingChecks, existingChecks, changedChecks, staleChecks := c.calcChecksDiff(consulChecks)
 	for _, check := range missingChecks {
 		if err := c.registerCheck(check); err != nil {
 			mErr.Errors = append(mErr.Errors, err)
@@ -403,6 +403,9 @@ func (c *Syncer) syncChecks() error {
 		c.trackedChecks[consulCheckID(check.ID)] = check
 		c.registryLock.Unlock()
 	}
+	for _, check := range existingChecks {
+		c.ensureCheckRunning(check)
+	}
 	for _, check := range changedChecks {
 		// NOTE(sean@): Do we need to deregister the check before
 		// re-registering it?  Not deregistering to avoid missing the
@@ -684,6 +687,16 @@ func (c *Syncer) registerCheck(chkReg *consul.AgentCheckRegistration) error {
 	return c.client.Agent().CheckRegister(chkReg)
 }
 
+// ensureCheckRunning starts the check runner for a check if it's not already running
+func (c *Syncer) ensureCheckRunning(chk *consul.AgentCheckRegistration) {
+	c.registryLock.RLock()
+	defer c.registryLock.RUnlock()
+	if cr, ok := c.checkRunners[consulCheckID(chk.ID)]; ok && !cr.Started() {
+		c.logger.Printf("[DEBUG] consul.syncer: starting runner for existing check. %v", chk.ID)
+		cr.Start()
+	}
+}
+
 // createCheckReg creates a Check that can be registered with Nomad. It also
 // creates a Nomad check for the check types that it can handle.
 func (c *Syncer) createCheckReg(check *structs.ServiceCheck, serviceReg *consul.AgentServiceRegistration) (*consul.AgentCheckRegistration, error) {