Merge pull request #139 from NearNodeFlash/drive-event

Notify upper layers that a drive failed

Author: ajfloeder

pkg/manager-message-registry/registries/Nnf.1.0.0.go

@@ -54,7 +54,7 @@
             "Message": "The storage pool '%1' patched, SerialNumber '%2', Namespace '%3' replaced with SerialNumber '%4', Namespace '%5'",
             "Severity": "OK",
             "MessageSeverity": "OK",
-            "NumberOfArgs": 1,
+            "NumberOfArgs": 5,
             "ParamTypes": [
                 "string",
                 "string",
@@ -77,6 +77,30 @@
                 "This argument shall contain the namespace id on the new storage device."
             ],
             "Resolution": "None"
+        },
+        "NvmeStateChange": {
+            "Description": "Indicates that an NVMe device has changed state",
+            "LongDescription": "This message shall be used to indicate that an NVMe device has changed state",
+            "Message": "The Nvme State changed slot '%1', model '%2', serial '%3'",
+            "Severity": "Warning",
+            "MessageSeverity": "Warning",
+            "NumberOfArgs": 3,
+            "ParamTypes": [
+                "number",
+                "string",
+                "string"
+            ],
+            "ArgDescriptions": [
+                "The slot identifier.",
+                "The model number.",
+                "The serial number."
+            ],
+            "ArgLongDescriptions": [
+                "This argument shall contain the NVMe slot number.",
+                "This argument shall contain the NVMe model number.",
+                "This argument shall contain the NVMe serial number."
+            ],
+            "Resolution": "None"
         }
     }
 }

pkg/manager-message-registry/registries/Nnf.1.0.0.json

@@ -172,6 +172,20 @@ type ProvidingVolume struct {
 	VolumeId string
 }
 
+// Calculate a drive's health from its current state
+func resourceHealthFromState(state sf.ResourceState) sf.ResourceHealth {
+	switch state {
+	case sf.ENABLED_RST:
+		return sf.OK_RH
+	case sf.DISABLED_RST:
+		return sf.WARNING_RH
+	case sf.UNAVAILABLE_OFFLINE_RST, sf.ABSENT_RST:
+		return sf.CRITICAL_RH
+	default:
+		return sf.CRITICAL_RH
+	}
+}
+
 // TODO: We may want to put this manager under a resource block
 //   /​redfish/​v1/​ResourceBlocks/​{ResourceBlockId} // <- Rabbit
 //   /​redfish/​v1/​ResourceBlocks/​{ResourceBlockId}/​Systems/{​ComputerSystemId} // <- Also Rabbit & Computes
@@ -510,7 +524,7 @@ func (s *Storage) getStatus() (stat sf.ResourceStatus) {
 		stat.State = sf.UNAVAILABLE_OFFLINE_RST
 		stat.Health = sf.CRITICAL_RH
 	} else {
-		stat.Health = sf.OK_RH
+		stat.Health = resourceHealthFromState(s.state)
 		stat.State = s.state
 	}
 
@@ -645,6 +659,14 @@ func (s *Storage) findVolume(volumeId string) *Volume {
 	return nil
 }
 
+func (s *Storage) notify(newState sf.ResourceState) {
+	if newState != s.state {
+		s.state = newState
+		event.EventManager.Publish(msgreg.NvmeStateChangeNnf(strconv.FormatInt(s.slot, 10), s.modelNumber, s.serialNumber))
+	}
+}
+
+// Getters for common volume operations
 func (v *Volume) Id() string                               { return v.id }
 func (v *Volume) GetOdataId() string                       { return v.storage.OdataId() + "/Volumes/" + v.id }
 func (v *Volume) GetCapacityBytes() uint64                 { return uint64(v.capacityBytes) }

pkg/manager-nvme/manager.go

@@ -90,16 +90,17 @@ func (m *Monitor) checkSmartLog(storage *Storage) {
 	smartLog, err := storage.device.GetSmartLog()
 	if err != nil {
 		log.Error(err, "smartlog request failed", "serial", storage.SerialNumber(), "slot", storage.Slot())
-		storage.state = sf.DISABLED_RST
-	}
+		storage.notify(sf.UNAVAILABLE_OFFLINE_RST)
+	} else {
 
-	// nvme.MangleSmartLog(smartLog)
+		// nvme.MangleSmartLog(smartLog)
 
-	state := nvme.InterpretSmartLog(smartLog)
-	if state != storage.state {
-		log.Info("smartlog state change", "old state", storage.state, "new state", state, "serial", storage.SerialNumber(), "slot", storage.Slot())
-		storage.state = state
+		state := nvme.InterpretSmartLog(smartLog)
+		if state != storage.state {
+			log.Info("smartlog state change", "old state", storage.state, "new state", state, "serial", storage.SerialNumber(), "slot", storage.Slot())
+			storage.notify(state)
 
-		nvme.LogSmartLog(log, smartLog)
+			nvme.LogSmartLog(log, smartLog)
+		}
 	}
 }