support flattened osdmap format added in octopus

ceph/exporter.go

@@ -60,11 +60,11 @@ func (exporter *Exporter) getCollectors() []prometheus.Collector {
 
 func (exporter *Exporter) cephVersionCmd() []byte {
 	cmd, err := json.Marshal(map[string]interface{}{
-		"prefix": "Version",
+		"prefix": "version",
 		"format": "json",
 	})
 	if err != nil {
-		exporter.Logger.WithError(err).Panic("failed to marshal ceph Version command")
+		exporter.Logger.WithError(err).Panic("failed to marshal ceph version command")
 	}
 
 	return cmd
@@ -87,6 +87,7 @@ func (exporter *Exporter) setCephVersion() error {
 
 	parsedVersion, err := ParseCephVersion(cephVersion.Version)
 	if err != nil {
+		exporter.Logger.Info("version " + cephVersion.Version)
 		return err
 	}
 
@@ -113,15 +114,15 @@ func (exporter *Exporter) Describe(ch chan<- *prometheus.Desc) {
 // prometheus. Collect could be called several times concurrently
 // and thus its run is protected by a single mutex.
 func (exporter *Exporter) Collect(ch chan<- prometheus.Metric) {
+	exporter.mu.Lock()
+	defer exporter.mu.Unlock()
+
 	err := exporter.setCephVersion()
 	if err != nil {
 		exporter.Logger.WithError(err).Error("failed to set ceph Version")
 		return
 	}
 
-	exporter.mu.Lock()
-	defer exporter.mu.Unlock()
-
 	for _, cc := range exporter.getCollectors() {
 		cc.Collect(ch)
 	}

ceph/health.go

@@ -976,6 +976,13 @@ func (c *ClusterHealthCollector) collectorList() []prometheus.Collector {
 	}
 }
 
+type osdMap struct {
+	NumOSDs        float64 `json:"num_osds"`
+	NumUpOSDs      float64 `json:"num_up_osds"`
+	NumInOSDs      float64 `json:"num_in_osds"`
+	NumRemappedPGs float64 `json:"num_remapped_pgs"`
+}
+
 type cephHealthStats struct {
 	Health struct {
 		Summary []struct {
@@ -991,15 +998,8 @@ type cephHealthStats struct {
 			} `json:"summary"`
 		} `json:"checks"`
 	} `json:"health"`
-	OSDMap struct {
-		OSDMap struct {
-			NumOSDs        float64 `json:"num_osds"`
-			NumUpOSDs      float64 `json:"num_up_osds"`
-			NumInOSDs      float64 `json:"num_in_osds"`
-			NumRemappedPGs float64 `json:"num_remapped_pgs"`
-		} `json:"osdmap"`
-	} `json:"osdmap"`
-	PGMap struct {
+	OSDMap map[string]interface{} `json:"osdmap"`
+	PGMap  struct {
 		NumPGs                  float64 `json:"num_pgs"`
 		TotalObjects            float64 `json:"num_objects"`
 		WriteOpPerSec           float64 `json:"write_op_per_sec"`
@@ -1350,15 +1350,38 @@ func (c *ClusterHealthCollector) collect(ch chan<- prometheus.Metric) error {
 	c.CacheFlushIORate.Set(stats.PGMap.CacheFlushBytePerSec)
 	c.CachePromoteIOOps.Set(stats.PGMap.CachePromoteOpPerSec)
 
-	c.OSDsUp.Set(stats.OSDMap.OSDMap.NumUpOSDs)
-	c.OSDsIn.Set(stats.OSDMap.OSDMap.NumInOSDs)
-	c.OSDsNum.Set(stats.OSDMap.OSDMap.NumOSDs)
+	var actualOsdMap osdMap
+	if c.version.IsAtLeast(Octopus) {
+		if stats.OSDMap != nil {
+			actualOsdMap = osdMap{
+				NumOSDs:        stats.OSDMap["num_osds"].(float64),
+				NumUpOSDs:      stats.OSDMap["num_up_osds"].(float64),
+				NumInOSDs:      stats.OSDMap["num_in_osds"].(float64),
+				NumRemappedPGs: stats.OSDMap["num_remapped_pgs"].(float64),
+			}
+		}
+	} else {
+		if stats.OSDMap != nil {
+			innerMap := stats.OSDMap["osdmap"].(map[string]interface{})
+
+			actualOsdMap = osdMap{
+				NumOSDs:        innerMap["num_osds"].(float64),
+				NumUpOSDs:      innerMap["num_up_osds"].(float64),
+				NumInOSDs:      innerMap["num_in_osds"].(float64),
+				NumRemappedPGs: innerMap["num_remapped_pgs"].(float64),
+			}
+		}
+	}
+
+	c.OSDsUp.Set(actualOsdMap.NumUpOSDs)
+	c.OSDsIn.Set(actualOsdMap.NumInOSDs)
+	c.OSDsNum.Set(actualOsdMap.NumOSDs)
 
 	// Ceph (until v10.2.3) doesn't expose the value of down OSDs
 	// from its status, which is why we have to compute it ourselves.
-	c.OSDsDown.Set(stats.OSDMap.OSDMap.NumOSDs - stats.OSDMap.OSDMap.NumUpOSDs)
+	c.OSDsDown.Set(actualOsdMap.NumOSDs - actualOsdMap.NumUpOSDs)
 
-	c.RemappedPGs.Set(stats.OSDMap.OSDMap.NumRemappedPGs)
+	c.RemappedPGs.Set(actualOsdMap.NumRemappedPGs)
 	c.TotalPGs.Set(stats.PGMap.NumPGs)
 	c.Objects.Set(stats.PGMap.TotalObjects)
 

ceph/health_test.go

@@ -816,6 +816,7 @@ $ sudo ceph -s
 			)
 
 			collector := NewClusterHealthCollector(&Exporter{Conn: conn, Cluster: "ceph", Logger: logrus.New()})
+			collector.version = &Version{Major: 14, Minor: 2, Patch: 0}
 			err := prometheus.Register(collector)
 			require.NoError(t, err)
 			defer prometheus.Unregister(collector)

ceph/version.go

@@ -73,7 +73,9 @@ func ParseCephVersion(cephVersion string) (*Version, error) {
 	}
 
 	otherVersions := strings.Split(someVersions[2], "-")
-	if len(otherVersions) != 3 {
+	if len(otherVersions) == 1 {
+		otherVersions = []string{otherVersions[0], "0", ""}
+	} else if len(otherVersions) != 3 {
 		return nil, ErrInvalidVersion
 	}
 

ceph/version_test.go

@@ -45,6 +45,12 @@ func TestParseCephVersion(t *testing.T) {
 			want:    &Version{Major: 16, Minor: 2, Patch: 3, Revision: 33, Commit: "gcc1e126"},
 			wantErr: false,
 		},
+		{
+			name:    "real pacific",
+			args:    args{cephVersion: "ceph version 16.2.7 (dd0603118f56ab514f133c8d2e3adfc983942503) pacific (stable)"},
+			want:    &Version{Major: 16, Minor: 2, Patch: 7, Revision: 0, Commit: ""},
+			wantErr: false,
+		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {