fix(go.d/k8sstate): use static list of warning/terminated reasons (ne…

…tdata#18885)
webfutureiorepo · Oct 28, 2024 · e465363 · e465363
1 parent 5204ee2
commit e465363
Show file tree

Hide file tree

Showing 6 changed files with 454 additions and 322 deletions.
diff --git a/src/go/plugin/go.d/modules/k8s_state/charts.go b/src/go/plugin/go.d/modules/k8s_state/charts.go
@@ -681,24 +681,42 @@ var (
 			{ID: "pod_%s_container_%s_state_terminated", Name: "terminated"},
 		},
 	}
-	containersStateWaitingChartTmpl = module.Chart{
-		IDSep:    true,
-		ID:       "pod_%s_container_%s.state_waiting_reason",
-		Title:    "Container waiting state reason",
-		Units:    "state",
-		Fam:      "container waiting reason",
-		Ctx:      "k8s_state.pod_container_waiting_state_reason",
-		Priority: prioPodContainerWaitingStateReason,
-	}
-	containersStateTerminatedChartTmpl = module.Chart{
-		IDSep:    true,
-		ID:       "pod_%s_container_%s.state_terminated_reason",
-		Title:    "Container terminated state reason",
-		Units:    "state",
-		Fam:      "container terminated reason",
-		Ctx:      "k8s_state.pod_container_terminated_state_reason",
-		Priority: prioPodContainerTerminatedStateReason,
-	}
+	containersStateWaitingChartTmpl = func() module.Chart {
+		chart := module.Chart{
+			IDSep:    true,
+			ID:       "pod_%s_container_%s.state_waiting_reason",
+			Title:    "Container waiting state reason",
+			Units:    "state",
+			Fam:      "container waiting reason",
+			Ctx:      "k8s_state.pod_container_waiting_state_reason",
+			Priority: prioPodContainerWaitingStateReason,
+		}
+		for _, v := range containerWaitingStateReasons {
+			chart.Dims = append(chart.Dims, &module.Dim{
+				ID:   "pod_%s_container_%s_state_waiting_reason_" + v,
+				Name: v,
+			})
+		}
+		return chart
+	}()
+	containersStateTerminatedChartTmpl = func() module.Chart {
+		chart := module.Chart{
+			IDSep:    true,
+			ID:       "pod_%s_container_%s.state_terminated_reason",
+			Title:    "Container terminated state reason",
+			Units:    "state",
+			Fam:      "container terminated reason",
+			Ctx:      "k8s_state.pod_container_terminated_state_reason",
+			Priority: prioPodContainerTerminatedStateReason,
+		}
+		for _, v := range containerTerminatedStateReasons {
+			chart.Dims = append(chart.Dims, &module.Dim{
+				ID:   "pod_%s_container_%s_state_terminated_reason_" + v,
+				Name: v,
+			})
+		}
+		return chart
+	}()
 )
 
 func (ks *KubeState) newContainerCharts(ps *podState, cs *containerState) *module.Charts {
@@ -728,42 +746,6 @@ func (ks *KubeState) addContainerCharts(ps *podState, cs *containerState) {
 	}
 }
 
-func (ks *KubeState) addContainerWaitingStateReasonToChart(ps *podState, cs *containerState, reason string) {
-	id := fmt.Sprintf(containersStateWaitingChartTmpl.ID, replaceDots(ps.id()), cs.name)
-	c := ks.Charts().Get(id)
-	if c == nil {
-		ks.Warningf("chart '%s' does not exist", id)
-		return
-	}
-	dim := &module.Dim{
-		ID:   fmt.Sprintf("pod_%s_container_%s_state_waiting_reason_%s", ps.id(), cs.name, reason),
-		Name: reason,
-	}
-	if err := c.AddDim(dim); err != nil {
-		ks.Warning(err)
-		return
-	}
-	c.MarkNotCreated()
-}
-
-func (ks *KubeState) addContainerTerminatedStateReasonToChart(ps *podState, cs *containerState, reason string) {
-	id := fmt.Sprintf(containersStateTerminatedChartTmpl.ID, replaceDots(ps.id()), cs.name)
-	c := ks.Charts().Get(id)
-	if c == nil {
-		ks.Warningf("chart '%s' does not exist", id)
-		return
-	}
-	dim := &module.Dim{
-		ID:   fmt.Sprintf("pod_%s_container_%s_state_terminated_reason_%s", ps.id(), cs.name, reason),
-		Name: reason,
-	}
-	if err := c.AddDim(dim); err != nil {
-		ks.Warning(err)
-		return
-	}
-	c.MarkNotCreated()
-}
-
 var discoveryStatusChart = module.Chart{
 	ID:       "discovery_discoverers_state",
 	Title:    "Running discoverers state",

diff --git a/src/go/plugin/go.d/modules/k8s_state/collect.go b/src/go/plugin/go.d/modules/k8s_state/collect.go
@@ -5,6 +5,7 @@ package k8s_state
 import (
 	"errors"
 	"fmt"
+	"slices"
 	"strings"
 	"time"
 
@@ -15,6 +16,29 @@ import (
 
 const precision = 1000
 
+var (
+	containerWaitingStateReasons = []string{
+		"PodInitializing",
+		"ContainerCreating",
+		"CrashLoopBackOff",
+		"CreateContainerConfigError",
+		"ErrImagePull",
+		"ImagePullBackOff",
+		"CreateContainerError",
+		"InvalidImageName",
+		"Other",
+	}
+	containerTerminatedStateReasons = []string{
+		"OOMKilled",
+		"Completed",
+		"Error",
+		"ContainerCannotRun",
+		"DeadlineExceeded",
+		"Evicted",
+		"Other",
+	}
+)
+
 func (ks *KubeState) collect() (map[string]int64, error) {
 	if ks.discoverer == nil {
 		return nil, errors.New("nil discoverer")
@@ -165,19 +189,25 @@ func (ks *KubeState) collectPodsState(mx map[string]int64) {
 			mx[ppx+"state_terminated"] = boolToInt(cs.stateTerminated)
 			mx[ppx+"readiness"] = boolToInt(cs.ready)
 			mx[ppx+"restarts"] = cs.restarts
-			for _, r := range cs.stateWaitingReasons {
-				if r.new {
-					r.new = false
-					ks.addContainerWaitingStateReasonToChart(ps, cs, r.reason)
+
+			for _, v := range containerWaitingStateReasons {
+				mx[ppx+"state_waiting_reason_"+v] = 0
+			}
+			if v := cs.waitingReason; v != "" {
+				if !slices.Contains(containerWaitingStateReasons, cs.waitingReason) {
+					v = "Other"
 				}
-				mx[ppx+"state_waiting_reason_"+r.reason] = boolToInt(r.active)
+				mx[ppx+"state_waiting_reason_"+v] = 1
+			}
+
+			for _, v := range containerTerminatedStateReasons {
+				mx[ppx+"state_terminated_reason_"+v] = 0
 			}
-			for _, r := range cs.stateTerminatedReasons {
-				if r.new {
-					r.new = false
-					ks.addContainerTerminatedStateReasonToChart(ps, cs, r.reason)
+			if v := cs.terminatedReason; v != "" {
+				if !slices.Contains(containerTerminatedStateReasons, cs.terminatedReason) {
+					v = "Other"
 				}
-				mx[ppx+"state_terminated_reason_"+r.reason] = boolToInt(r.active)
+				mx[ppx+"state_terminated_reason_"+v] = 1
 			}
 		}
 	}