Merge pull request #768 from fabxc/fabxc/rule-err

main: exit on invalid rule files on startup.

main.go

@@ -225,12 +225,13 @@ func (p *prometheus) reloadConfig() bool {
 		log.Errorf("Note: The configuration format has changed with version 0.14. Please see the documentation (http://prometheus.io/docs/operating/configuration/) and the provided configuration migration tool (https://github.com/prometheus/migrate).")
 		return false
 	}
+	success := true
 
-	p.webService.StatusHandler.ApplyConfig(conf)
+	success = success && p.webService.StatusHandler.ApplyConfig(conf)
-	p.targetManager.ApplyConfig(conf)
+	success = success && p.targetManager.ApplyConfig(conf)
-	p.ruleManager.ApplyConfig(conf)
+	success = success && p.ruleManager.ApplyConfig(conf)
 
-	return true
+	return success
 }
 
 // Serve starts the Prometheus server. It returns after the server has been shut

retrieval/targetmanager.go

@@ -284,7 +284,8 @@ func (tm *TargetManager) Pools() map[string][]*Target {
 
 // ApplyConfig resets the manager's target providers and job configurations as defined
 // by the new cfg. The state of targets that are valid in the new configuration remains unchanged.
-func (tm *TargetManager) ApplyConfig(cfg *config.Config) {
+// Returns true on success.
+func (tm *TargetManager) ApplyConfig(cfg *config.Config) bool {
 	tm.m.RLock()
 	running := tm.running
 	tm.m.RUnlock()
@@ -305,6 +306,7 @@ func (tm *TargetManager) ApplyConfig(cfg *config.Config) {
 
 	tm.globalLabels = cfg.GlobalConfig.Labels
 	tm.providers = providers
+	return true
 }
 
 // targetsFromGroup builds targets based on the given TargetGroup and config.

rules/manager.go

@@ -271,11 +271,12 @@ func (m *Manager) runIteration() {
 }
 
 // ApplyConfig updates the rule manager's state as the config requires. If
-// loading the new rules failed the old rule set is restored.
+// loading the new rules failed the old rule set is restored. Returns true on success.
-func (m *Manager) ApplyConfig(conf *config.Config) {
+func (m *Manager) ApplyConfig(conf *config.Config) bool {
 	m.Lock()
 	defer m.Unlock()
 
+	success := true
 	m.interval = time.Duration(conf.GlobalConfig.EvaluationInterval)
 
 	rulesSnapshot := make([]Rule, len(m.rules))
@@ -288,6 +289,7 @@ func (m *Manager) ApplyConfig(conf *config.Config) {
 		if err != nil {
 			// The only error can be a bad pattern.
 			log.Errorf("Error retrieving rule files for %s: %s", pat, err)
+			success = false
 		}
 		files = append(files, fs...)
 	}
@@ -295,7 +297,9 @@ func (m *Manager) ApplyConfig(conf *config.Config) {
 		// If loading the new rules failed, restore the old rule set.
 		m.rules = rulesSnapshot
 		log.Errorf("Error loading rules, previous rule set restored: %s", err)
+		success = false
 	}
+	return success
 }
 
 // loadRuleFiles loads alerting and recording rules from the given files.

web/status.go

@@ -54,8 +54,10 @@ func (h *PrometheusStatusHandler) ServeHTTP(w http.ResponseWriter, r *http.Reque
 }
 
 // ApplyConfig updates the status handler's state as the new config requires.
-func (h *PrometheusStatusHandler) ApplyConfig(conf *config.Config) {
+// Returns true on success.
+func (h *PrometheusStatusHandler) ApplyConfig(conf *config.Config) bool {
 	h.mu.Lock()
 	h.Config = conf.String()
 	h.mu.Unlock()
+	return true
 }