[agent] enable basic metric collection and exporter

api/agent/v1alpha2/agent_types.go

@@ -17,6 +17,7 @@ package v1alpha2
 import (
 	"sort"
 
+	"github.com/pkg/errors"
 	"go.githedgehog.com/fabric/api/meta"
 	vpcapi "go.githedgehog.com/fabric/api/vpc/v1alpha2"
 	wiringapi "go.githedgehog.com/fabric/api/wiring/v1alpha2"
@@ -126,14 +127,208 @@ type AgentStatus struct {
 	LastAppliedTime metav1.Time `json:"lastAppliedTime,omitempty"`
 	// Generation of the last successful configuration application
 	LastAppliedGen int64 `json:"lastAppliedGen,omitempty"`
-	// Information about the switch and NOS
+	// Detailed switch state updated with each heartbeat
+	State SwitchState `json:"state,omitempty"`
+	// Information about the switch and NOS updated with each heartbeat
 	NOSInfo NOSInfo `json:"nosInfo,omitempty"`
 	// Status updates from the agent
 	StatusUpdates []ApplyStatusUpdate `json:"statusUpdates,omitempty"`
 	// Conditions of the agent, includes readiness marker for use with kubectl wait
 	Conditions []metav1.Condition `json:"conditions"`
 }
 
+type SwitchState struct {
+	// Switch interfaces state (incl. physical, management and port channels)
+	Interfaces map[string]SwitchStateInterface `json:"interfaces,omitempty"` // TODO add LLDP to interface?
+	// Switch transceivers state (port -> transceiver)
+	Transceivers map[string]SwitchStateTransceiver `json:"transceivers,omitempty"` // TODO move to interface?
+	// State of all BGP neighbors (VRF -> neighbor address -> state)
+	BGPNeighbors map[string]map[string]SwitchStateBGPNeighbor `json:"bgpNeighbors,omitempty"`
+
+	// TODO add FAN info, PSU info, Temperature info
+}
+
+type SwitchStateInterface struct {
+	Enabled     bool                         `json:"enabled,omitempty"`
+	AdminStatus AdminStatus                  `json:"adminStatus,omitempty"`
+	OperStatus  OperStatus                   `json:"operStatus,omitempty"`
+	MAC         string                       `json:"mac,omitempty"`
+	LastChange  metav1.Time                  `json:"lastChanged,omitempty"`
+	Counters    SwitchStateInterfaceCounters `json:"counters,omitempty"`
+}
+
+type SwitchStateInterfaceCounters struct {
+	InBitsPerSecond  float64     `json:"inBitsPerSecond,omitempty"`
+	InDiscards       uint64      `json:"inDiscards,omitempty"`
+	InErrors         uint64      `json:"inErrors,omitempty"`
+	InPktsPerSecond  float64     `json:"inPktsPerSecond,omitempty"`
+	InUtilization    uint8       `json:"inUtilization,omitempty"`
+	LastClear        metav1.Time `json:"lastClear,omitempty"`
+	OutBitsPerSecond float64     `json:"outBitsPerSecond,omitempty"`
+	OutDiscards      uint64      `json:"outDiscards,omitempty"`
+	OutErrors        uint64      `json:"outErrors,omitempty"`
+	OutPktsPerSecond float64     `json:"outPktsPerSecond,omitempty"`
+	OutUtilization   uint8       `json:"outUtilization,omitempty"`
+}
+
+type AdminStatus string
+
+const (
+	AdminStatusUnset   AdminStatus = ""
+	AdminStatusUp      AdminStatus = "up"
+	AdminStatusDown    AdminStatus = "down"
+	AdminStatusTesting AdminStatus = "testing"
+)
+
+func (a AdminStatus) ID() (uint8, error) {
+	switch a {
+	case AdminStatusUnset:
+		return 0, nil
+	case AdminStatusUp:
+		return 1, nil
+	case AdminStatusDown:
+		return 2, nil
+	case AdminStatusTesting:
+		return 3, nil
+	default:
+		return 0, errors.Errorf("unknown AdminStatus %s", a)
+	}
+}
+
+type OperStatus string
+
+const (
+	OperStatusUnset          OperStatus = ""
+	OperStatusUp             OperStatus = "up"
+	OperStatusDown           OperStatus = "down"
+	OperStatusTesting        OperStatus = "testing"
+	OperStatusUnknown        OperStatus = "unknown"
+	OperStatusDormant        OperStatus = "dormant"
+	OperStatusNotPresent     OperStatus = "notPresent"
+	OperStatusLowerLayerDown OperStatus = "lowerLayerDown"
+)
+
+func (o OperStatus) ID() (uint8, error) {
+	switch o {
+	case OperStatusUnset:
+		return 0, nil
+	case OperStatusUp:
+		return 1, nil
+	case OperStatusDown:
+		return 2, nil
+	case OperStatusTesting:
+		return 3, nil
+	case OperStatusUnknown:
+		return 4, nil
+	case OperStatusDormant:
+		return 5, nil
+	case OperStatusNotPresent:
+		return 6, nil
+	case OperStatusLowerLayerDown:
+		return 7, nil
+	default:
+		return 0, errors.Errorf("unknown OperStatus %s", o)
+	}
+}
+
+type SwitchStateTransceiver struct {
+	CableClass  string  `json:"cable-class,omitempty"`
+	Temperature float64 `json:"temperature,omitempty"`
+	Voltage     float64 `json:"voltage,omitempty"`
+
+	// TODO add vendor, serial number, etc
+	// Vendor           *string  `json:"vendor,omitempty" module:"openconfig-platform-diagnostics"`
+	// VendorPart       *string  `json:"vendor-part,omitempty" module:"openconfig-platform-diagnostics"`
+}
+
+type SwitchStateBGPNeighbor struct {
+	ConnectionsDropped     uint64                  `json:"connectionsDropped,omitempty"`
+	Enabled                bool                    `json:"enabled,omitempty"`
+	EstablishedTransitions uint64                  `json:"establishedTransitions,omitempty"`
+	LastEstablished        metav1.Time             `json:"lastEstablished,omitempty"`
+	LastRead               metav1.Time             `json:"lastRead,omitempty"`
+	LastResetReason        string                  `json:"lastResetReason,omitempty"`
+	LastResetTime          metav1.Time             `json:"lastResetTime,omitempty"`
+	LastWrite              metav1.Time             `json:"lastWrite,omitempty"`
+	LocalAS                uint32                  `json:"localAS,omitempty"`
+	Messages               BGPMessages             `json:"messages,omitempty"`
+	PeerAS                 uint32                  `json:"peerAS,omitempty"`
+	PeerGroup              string                  `json:"peerGroup,omitempty"`
+	PeerPort               uint16                  `json:"peerPort,omitempty"`
+	PeerType               BGPPeerType             `json:"peerType,omitempty"`
+	RemoteRouterID         string                  `json:"remoteRouterID,omitempty"`
+	SessionState           BGPNeighborSessionState `json:"sessionState,omitempty"`
+	ShutdownMessage        string                  `json:"shutdownMessage,omitempty"`
+}
+
+type BGPNeighborSessionState string
+
+const (
+	BGPNeighborSessionStateUnset       BGPNeighborSessionState = ""
+	BGPNeighborSessionStateIdle        BGPNeighborSessionState = "idle"
+	BGPNeighborSessionStateConnect     BGPNeighborSessionState = "connect"
+	BGPNeighborSessionStateActive      BGPNeighborSessionState = "active"
+	BGPNeighborSessionStateOpenSent    BGPNeighborSessionState = "openSent"
+	BGPNeighborSessionStateOpenConfirm BGPNeighborSessionState = "openConfirm"
+	BGPNeighborSessionStateEstablished BGPNeighborSessionState = "established"
+)
+
+func (b BGPNeighborSessionState) ID() (uint8, error) {
+	switch b {
+	case BGPNeighborSessionStateUnset:
+		return 0, nil
+	case BGPNeighborSessionStateIdle:
+		return 1, nil
+	case BGPNeighborSessionStateConnect:
+		return 2, nil
+	case BGPNeighborSessionStateActive:
+		return 3, nil
+	case BGPNeighborSessionStateOpenSent:
+		return 4, nil
+	case BGPNeighborSessionStateOpenConfirm:
+		return 5, nil
+	case BGPNeighborSessionStateEstablished:
+		return 6, nil
+	default:
+		return 0, errors.Errorf("unknown BGPNeighborSessionState %s", b)
+	}
+}
+
+type BGPPeerType string
+
+const (
+	BGPPeerTypeUnset    BGPPeerType = ""
+	BGPPeerTypeInternal BGPPeerType = "internal"
+	BGPPeerTypeExternal BGPPeerType = "external"
+)
+
+func (b BGPPeerType) ID() (uint8, error) {
+	switch b {
+	case BGPPeerTypeUnset:
+		return 0, nil
+	case BGPPeerTypeInternal:
+		return 1, nil
+	case BGPPeerTypeExternal:
+		return 2, nil
+	default:
+		return 0, errors.Errorf("unknown BGPPeerType %s", b)
+	}
+}
+
+type BGPMessages struct {
+	Received BGPMessagesCounters `json:"received,omitempty"`
+	Sent     BGPMessagesCounters `json:"sent,omitempty"`
+}
+
+type BGPMessagesCounters struct {
+	Capability   uint64 `json:"capability,omitempty"`
+	Keepalive    uint64 `json:"keepalive,omitempty"`
+	Notification uint64 `json:"notification,omitempty"`
+	Open         uint64 `json:"open,omitempty"`
+	RouteRefresh uint64 `json:"routeRefresh,omitempty"`
+	Update       uint64 `json:"update,omitempty"`
+}
+
 // NOSInfo contains information about the switch and NOS received from the switch itself by the agent
 type NOSInfo struct {
 	// ASIC name, such as "broadcom" or "vs"