Merge pull request #575 from replicatedhq/diamonwiggins/sc-40946/ceph-analyzer-health-status

Ceph health status messages in Analzyer result

Author: diamonwiggins

pkg/analyze/ceph.go

@@ -74,7 +74,17 @@ type CephStatus struct {
 }
 
 type HealthStatus struct {
-	Status string `json:"status"`
+	Status string                  `json:"status"`
+	Checks map[string]CheckMessage `json:"checks"`
+}
+
+type CheckMessage struct {
+	Severity string  `json:"severity"`
+	Summary  Summary `json:"summary"`
+}
+
+type Summary struct {
+	Message string `json:"message"`
 }
 
 type OsdMap struct {
@@ -121,6 +131,7 @@ func cephStatus(analyzer *troubleshootv1beta2.CephStatusAnalyze, getCollectedFil
 			if outcome.Fail.When == "" {
 				outcome.Fail.When = string(CephHealthErr)
 			}
+
 			match, err := compareCephStatus(status.Health.Status, outcome.Fail.When)
 			if err != nil {
 				return nil, errors.Wrap(err, "failed to compare ceph status")
@@ -134,6 +145,7 @@ func cephStatus(analyzer *troubleshootv1beta2.CephStatusAnalyze, getCollectedFil
 			if outcome.Warn.When == "" {
 				outcome.Warn.When = string(CephHealthWarn)
 			}
+
 			match, err := compareCephStatus(status.Health.Status, outcome.Warn.When)
 			if err != nil {
 				return nil, errors.Wrap(err, "failed to compare ceph status")
@@ -147,13 +159,15 @@ func cephStatus(analyzer *troubleshootv1beta2.CephStatusAnalyze, getCollectedFil
 			if outcome.Pass.When == "" {
 				outcome.Pass.When = string(CephHealthOK)
 			}
+
 			match, err := compareCephStatus(status.Health.Status, outcome.Pass.When)
 			if err != nil {
 				return nil, errors.Wrap(err, "failed to compare ceph status")
 			} else if match {
 				analyzeResult.IsPass = true
 				analyzeResult.Message = outcome.Pass.Message
 				analyzeResult.URI = outcome.Pass.URI
+
 				return analyzeResult, nil
 			}
 		}
@@ -195,21 +209,33 @@ func compareCephStatus(actual, when string) (bool, error) {
 	}
 }
 
-func detailedCephMessage(msg string, status CephStatus) string {
+func detailedCephMessage(outcomeMessage string, status CephStatus) string {
+	var msg = []string{}
+
+	if outcomeMessage != "" {
+		msg = append(msg, outcomeMessage)
+	}
+
 	if status.OsdMap.OsdMap.NumOsd > 0 {
-		msg = fmt.Sprintf("%s. %v/%v OSDs up", msg, status.OsdMap.OsdMap.NumUpOsd, status.OsdMap.OsdMap.NumOsd)
+		msg = append(msg, fmt.Sprintf("%v/%v OSDs up", status.OsdMap.OsdMap.NumUpOsd, status.OsdMap.OsdMap.NumOsd))
 	}
 
 	if status.OsdMap.OsdMap.Full {
-		msg = fmt.Sprintf("%s. OSD disk is full", msg)
+		msg = append(msg, fmt.Sprintf("OSD disk is full"))
 	} else if status.OsdMap.OsdMap.NearFull {
-		msg = fmt.Sprintf("%s. OSD disk is nearly full", msg)
+		msg = append(msg, fmt.Sprintf("OSD disk is nearly full"))
 	}
 
 	if status.PgMap.TotalBytes > 0 {
 		pgUsage := 100 * float64(status.PgMap.UsedBytes) / float64(status.PgMap.TotalBytes)
-		msg = fmt.Sprintf("%s. PG storage usage is %.1f%%.", msg, pgUsage)
+		msg = append(msg, fmt.Sprintf("PG storage usage is %.1f%%", pgUsage))
+	}
+
+	if status.Health.Checks != nil {
+		for k, v := range status.Health.Checks {
+			msg = append(msg, fmt.Sprintf("%s: %s", k, v.Summary.Message))
+		}
 	}
 
-	return msg
+	return strings.Join(msg, "\n")
 }

pkg/analyze/ceph_test.go

@@ -56,7 +56,7 @@ func Test_cephStatus(t *testing.T) {
 				IsWarn:  true,
 				IsFail:  false,
 				Title:   "Ceph Status",
-				Message: "Ceph status is HEALTH_WARN. 5/5 OSDs up. OSD disk is nearly full. PG storage usage is 85.0%.",
+				Message: "Ceph status is HEALTH_WARN\n5/5 OSDs up\nOSD disk is nearly full\nPG storage usage is 85.0%",
 				URI:     "https://rook.io/docs/rook/v1.4/ceph-common-issues.html",
 				IconKey: "rook",
 				IconURI: "https://troubleshoot.sh/images/analyzer-icons/rook.svg?w=11&h=16",
@@ -89,7 +89,7 @@ func Test_cephStatus(t *testing.T) {
 				IsWarn:  false,
 				IsFail:  true,
 				Title:   "Ceph Status",
-				Message: "Ceph status is HEALTH_ERR. 4/5 OSDs up. OSD disk is full. PG storage usage is 95.0%.",
+				Message: "Ceph status is HEALTH_ERR\n4/5 OSDs up\nOSD disk is full\nPG storage usage is 95.0%",
 				URI:     "https://rook.io/docs/rook/v1.4/ceph-common-issues.html",
 				IconKey: "rook",
 				IconURI: "https://troubleshoot.sh/images/analyzer-icons/rook.svg?w=11&h=16",
@@ -172,7 +172,7 @@ func Test_cephStatus(t *testing.T) {
 				IsWarn:  false,
 				IsFail:  true,
 				Title:   "Ceph Status",
-				Message: "custom message WARN. 5/5 OSDs up. OSD disk is nearly full. PG storage usage is 85.0%.",
+				Message: "custom message WARN\n5/5 OSDs up\nOSD disk is nearly full\nPG storage usage is 85.0%",
 				URI:     "custom uri WARN",
 				IconKey: "rook",
 				IconURI: "https://troubleshoot.sh/images/analyzer-icons/rook.svg?w=11&h=16",
@@ -218,6 +218,43 @@ func Test_cephStatus(t *testing.T) {
 				}
 			}`,
 		},
+		{
+			name:     "warn case with multiple health status messages",
+			analyzer: troubleshootv1beta2.CephStatusAnalyze{},
+			expectResult: AnalyzeResult{
+				IsPass:  false,
+				IsWarn:  true,
+				IsFail:  false,
+				Title:   "Ceph Status",
+				Message: "Ceph status is HEALTH_WARN\nPOOL_NO_REDUNDANCY: 11 pool(s) have no replicas configured\nPOOL_PG_NUM_NOT_POWER_OF_TWO: 8 pool(s) have non-power-of-two pg_num",
+				URI:     "https://rook.io/docs/rook/v1.4/ceph-common-issues.html",
+				IconKey: "rook",
+				IconURI: "https://troubleshoot.sh/images/analyzer-icons/rook.svg?w=11&h=16",
+			},
+			filePath: "ceph/status.json",
+			file: `{
+				"fsid": "96a8178c-6aa2-4adf-a309-9e8869a79611",
+				"health": {
+					"status": "HEALTH_WARN",
+					"checks": {
+						"POOL_NO_REDUNDANCY": {
+							"severity": "HEALTH_WARN",
+							"summary": {
+								"message": "11 pool(s) have no replicas configured",
+								"count": 11
+							},
+							"muted": false
+						},
+						"POOL_PG_NUM_NOT_POWER_OF_TWO": {
+							"severity": "HEALTH_WARN",
+							"summary": {
+								"message": "8 pool(s) have non-power-of-two pg_num"
+							}
+						}
+					}
+				}
+			}`,
+		},
 	}
 
 	for _, test := range tests {