From 49d4ce13bb2f1f1d34e5a9ae8713e21de67daf82 Mon Sep 17 00:00:00 2001
From: Sami Alhaddad <rootsami@gmail.com>
Date: Thu, 16 Feb 2023 00:38:09 +0100
Subject: [PATCH] feat: add more metric to detect failures (#18)

---
 pkg/metric/metric.go   | 14 ++++++++++++++
 pkg/server/handlers.go | 11 ++++++++++-
 2 files changed, 24 insertions(+), 1 deletion(-)

diff --git a/pkg/metric/metric.go b/pkg/metric/metric.go
index 3de8c72..623bb82 100644
--- a/pkg/metric/metric.go
+++ b/pkg/metric/metric.go
@@ -10,6 +10,8 @@ type Metrics struct {
 	ChangeResources  *prometheus.GaugeVec
 	AddResources     *prometheus.GaugeVec
 	DestroyResources *prometheus.GaugeVec
+	PlanFailure      *prometheus.GaugeVec
+	GitPullFailure   *prometheus.GaugeVec
 }
 
 func NewMetrics(reg prometheus.Registerer) *Metrics {
@@ -29,10 +31,22 @@ func NewMetrics(reg prometheus.Registerer) *Metrics {
 			Name:      "plan_destroy_resources",
 			Help:      "Number of resources to be destroyed based on tf plan",
 		}, []string{"stack"}),
+		PlanFailure: prometheus.NewGaugeVec(prometheus.GaugeOpts{
+			Namespace: "terradrift",
+			Name:      "plan_failure",
+			Help:      "Status of the last scan of a stack",
+		}, []string{"stack"}),
+		GitPullFailure: prometheus.NewGaugeVec(prometheus.GaugeOpts{
+			Namespace: "terradrift",
+			Name:      "git_pull_failure",
+			Help:      "Status of the last git pull",
+		}, []string{}),
 	}
 	reg.MustRegister(m.AddResources)
 	reg.MustRegister(m.ChangeResources)
 	reg.MustRegister(m.DestroyResources)
+	reg.MustRegister(m.PlanFailure)
+	reg.MustRegister(m.GitPullFailure)
 
 	return m
 }
diff --git a/pkg/server/handlers.go b/pkg/server/handlers.go
index daadc87..2b79bcf 100644
--- a/pkg/server/handlers.go
+++ b/pkg/server/handlers.go
@@ -17,6 +17,9 @@ func (s Server) scanHandler(c *gin.Context) {
 	name := c.Query("stack")
 	planResp, err := tfstack.StackScan(name, s.Workdir, s.ConfigPath, s.ExtraBackendVars)
 
+	// Reset the plan failure metric
+	promMetrics.PlanFailure.With(prometheus.Labels{"stack": name}).Set(0)
+
 	if err == nil {
 
 		// Record metrics for drifts in resources
@@ -35,11 +38,13 @@ func (s Server) scanHandler(c *gin.Context) {
 			c.JSON(404, errorMessage)
 		} else if strings.Contains(errorMessage, "error acquiring the state lock") {
 
+			promMetrics.PlanFailure.With(prometheus.Labels{"stack": name}).Set(1)
 			// When there's a current terrafom plan in progress, terraform locks the state till it's finished.
 			c.JSON(502, "Another plan is in-progress for the requested stack, please try again in few minutes.")
 
 		} else {
 
+			promMetrics.PlanFailure.With(prometheus.Labels{"stack": name}).Set(1)
 			c.JSON(500, errorMessage)
 		}
 	}
@@ -48,9 +53,13 @@ func (s Server) scanHandler(c *gin.Context) {
 // gitHandler is a handler function for git sync endpoint
 func (s Server) gitHandler(c *gin.Context) {
 
+	// Reset the git failure metric
+	promMetrics.GitPullFailure.With(prometheus.Labels{}).Set(0)
+
 	status, err := git.GitPull(s.Workdir, s.GitToken, s.GitTimeout)
 	if err != nil {
-		c.JSON(500, err)
+		promMetrics.GitPullFailure.With(prometheus.Labels{}).Set(1)
+		c.JSON(500, error.Error(err))
 	} else {
 		c.JSON(200, status)
 	}