Send device scan status (#31894)

cmd/agent/subcommands/snmp/command.go

@@ -9,10 +9,6 @@ package snmp
 import (
 	"errors"
 	"fmt"
-	"net"
-	"os"
-	"strconv"
-
 	"github.com/DataDog/datadog-agent/cmd/agent/command"
 	"github.com/DataDog/datadog-agent/comp/aggregator"
 	"github.com/DataDog/datadog-agent/comp/aggregator/demultiplexer/demultiplexerimpl"
@@ -30,11 +26,15 @@ import (
 	compressionfx "github.com/DataDog/datadog-agent/comp/serializer/compression/fx"
 	snmpscan "github.com/DataDog/datadog-agent/comp/snmpscan/def"
 	snmpscanfx "github.com/DataDog/datadog-agent/comp/snmpscan/fx"
+	"github.com/DataDog/datadog-agent/pkg/networkdevice/metadata"
 	"github.com/DataDog/datadog-agent/pkg/snmp/snmpparse"
 	"github.com/DataDog/datadog-agent/pkg/util/fxutil"
-
 	"github.com/spf13/cobra"
 	"go.uber.org/fx"
+	"net"
+	"os"
+	"strconv"
+	"time"
 )
 
 const (
@@ -302,16 +302,64 @@ func scanDevice(connParams *snmpparse.SNMPConfig, args argsType, snmpScanner snm
 		// newSNMP only returns config errors, so any problem is a usage error
 		return configErr{err}
 	}
-	if err := snmp.Connect(); err != nil {
+	namespace := conf.GetString("network_devices.namespace")
+	deviceID := namespace + ":" + connParams.IPAddress
+	// Since the snmp connection can take a while, start by sending an in progress status for the start of the scan
+	// before connecting to the agent
+	inProgressStatusPayload := metadata.NetworkDevicesMetadata{
+		DeviceScanStatus: &metadata.ScanStatusMetadata{
+			DeviceID:   deviceID,
+			ScanStatus: metadata.ScanStatusInProgress,
+		},
+		CollectTimestamp: time.Now().Unix(),
+		Namespace:        namespace,
+	}
+	if err = snmpScanner.SendPayload(inProgressStatusPayload); err != nil {
+		return fmt.Errorf("unable to send in progress status: %v", err)
+	}
+	if err = snmp.Connect(); err != nil {
+		// Send an error status if we can't connect to the agent
+		errorStatusPayload := metadata.NetworkDevicesMetadata{
+			DeviceScanStatus: &metadata.ScanStatusMetadata{
+				DeviceID:   deviceID,
+				ScanStatus: metadata.ScanStatusError,
+			},
+			CollectTimestamp: time.Now().Unix(),
+			Namespace:        namespace,
+		}
+		if err = snmpScanner.SendPayload(errorStatusPayload); err != nil {
+			return fmt.Errorf("unable to send error status: %v", err)
+		}
 		return fmt.Errorf("unable to connect to SNMP agent on %s:%d: %w", snmp.LocalAddr, snmp.Port, err)
 	}
-
-	namespace := conf.GetString("network_devices.namespace")
-
-	err = snmpScanner.RunDeviceScan(snmp, namespace, connParams.IPAddress)
+	err = snmpScanner.RunDeviceScan(snmp, namespace, deviceID)
 	if err != nil {
+		// Send an error status if we can't scan the device
+		errorStatusPayload := metadata.NetworkDevicesMetadata{
+			DeviceScanStatus: &metadata.ScanStatusMetadata{
+				DeviceID:   deviceID,
+				ScanStatus: metadata.ScanStatusError,
+			},
+			CollectTimestamp: time.Now().Unix(),
+			Namespace:        namespace,
+		}
+		if err = snmpScanner.SendPayload(errorStatusPayload); err != nil {
+			return fmt.Errorf("unable to send error status: %v", err)
+		}
 		return fmt.Errorf("unable to perform device scan: %v", err)
 	}
+	// Send a completed status if the scan was successful
+	completedStatusPayload := metadata.NetworkDevicesMetadata{
+		DeviceScanStatus: &metadata.ScanStatusMetadata{
+			DeviceID:   deviceID,
+			ScanStatus: metadata.ScanStatusCompleted,
+		},
+		CollectTimestamp: time.Now().Unix(),
+		Namespace:        namespace,
+	}
+	if err = snmpScanner.SendPayload(completedStatusPayload); err != nil {
+		return fmt.Errorf("unable to send completed status: %v", err)
+	}
 	return nil
 }
 

comp/snmpscan/def/component.go

@@ -7,6 +7,7 @@
 package snmpscan
 
 import (
+	"github.com/DataDog/datadog-agent/pkg/networkdevice/metadata"
 	"github.com/gosnmp/gosnmp"
 )
 
@@ -15,6 +16,7 @@ import (
 // Component is the component type.
 type Component interface {
 	// Triggers a device scan
-	RunDeviceScan(snmpConection *gosnmp.GoSNMP, deviceNamespace string, deviceIPAddress string) error
+	RunDeviceScan(snmpConection *gosnmp.GoSNMP, deviceNamespace string, deviceID string) error
 	RunSnmpWalk(snmpConection *gosnmp.GoSNMP, firstOid string) error
+	SendPayload(payload metadata.NetworkDevicesMetadata) error
 }

comp/snmpscan/impl/devicescan.go

@@ -6,23 +6,20 @@
 package snmpscanimpl
 
 import (
-	"encoding/json"
 	"time"
 
-	"github.com/DataDog/datadog-agent/comp/forwarder/eventplatform"
-	"github.com/DataDog/datadog-agent/pkg/logs/message"
 	"github.com/DataDog/datadog-agent/pkg/networkdevice/metadata"
 	"github.com/DataDog/datadog-agent/pkg/snmp/gosnmplib"
 	"github.com/gosnmp/gosnmp"
 )
 
-func (s snmpScannerImpl) RunDeviceScan(snmpConnection *gosnmp.GoSNMP, deviceNamespace string, deviceIPAddress string) error {
+func (s snmpScannerImpl) RunDeviceScan(snmpConnection *gosnmp.GoSNMP, deviceNamespace string, deviceID string) error {
+	// execute the scan
 	pdus, err := gatherPDUs(snmpConnection)
 	if err != nil {
 		return err
 	}
 
-	deviceID := deviceNamespace + ":" + deviceIPAddress
 	var deviceOids []*metadata.DeviceOID
 	for _, pdu := range pdus {
 		record, err := metadata.DeviceOIDFromPDU(deviceID, pdu)
@@ -35,15 +32,8 @@ func (s snmpScannerImpl) RunDeviceScan(snmpConnection *gosnmp.GoSNMP, deviceName
 
 	metadataPayloads := metadata.BatchDeviceScan(deviceNamespace, time.Now(), metadata.PayloadMetadataBatchSize, deviceOids)
 	for _, payload := range metadataPayloads {
-		payloadBytes, err := json.Marshal(payload)
+		err := s.SendPayload(payload)
 		if err != nil {
-			s.log.Errorf("Error marshalling device metadata: %v", err)
-			continue
-		}
-		m := message.NewMessage(payloadBytes, nil, "", 0)
-		s.log.Debugf("Device OID metadata payload is %d bytes", len(payloadBytes))
-		s.log.Tracef("Device OID metadata payload: %s", string(payloadBytes))
-		if err := s.epforwarder.SendEventPlatformEventBlocking(m, eventplatform.EventTypeNetworkDevicesMetadata); err != nil {
 			return err
 		}
 	}

comp/snmpscan/impl/sendpayload.go

@@ -0,0 +1,28 @@
+// Unless explicitly stated otherwise all files in this repository are licensed
+// under the Apache License Version 2.0.
+// This product includes software developed at Datadog (https://www.datadoghq.com/).
+// Copyright 2024-present Datadog, Inc.
+
+package snmpscanimpl
+
+import (
+	"encoding/json"
+	"github.com/DataDog/datadog-agent/comp/forwarder/eventplatform"
+	"github.com/DataDog/datadog-agent/pkg/logs/message"
+	"github.com/DataDog/datadog-agent/pkg/networkdevice/metadata"
+)
+
+func (s snmpScannerImpl) SendPayload(payload metadata.NetworkDevicesMetadata) error {
+	payloadBytes, err := json.Marshal(payload)
+	if err != nil {
+		s.log.Errorf("Error marshalling device metadata: %v", err)
+		return nil
+	}
+	m := message.NewMessage(payloadBytes, nil, "", 0)
+	s.log.Debugf("Device metadata payload is %d bytes", len(payloadBytes))
+	s.log.Tracef("Device metadata payload: %s", string(payloadBytes))
+	if err := s.epforwarder.SendEventPlatformEventBlocking(m, eventplatform.EventTypeNetworkDevicesMetadata); err != nil {
+		return err
+	}
+	return nil
+}

comp/snmpscan/mock/mock.go

@@ -13,6 +13,7 @@ import (
 
 	log "github.com/DataDog/datadog-agent/comp/core/log/def"
 	snmpscan "github.com/DataDog/datadog-agent/comp/snmpscan/def"
+	"github.com/DataDog/datadog-agent/pkg/networkdevice/metadata"
 	"github.com/gosnmp/gosnmp"
 )
 
@@ -38,3 +39,6 @@ func (m mock) RunDeviceScan(_ *gosnmp.GoSNMP, _ string, _ string) error {
 func (m mock) RunSnmpWalk(_ *gosnmp.GoSNMP, _ string) error {
 	return nil
 }
+func (m mock) SendPayload(_ metadata.NetworkDevicesMetadata) error {
+	return nil
+}

pkg/networkdevice/metadata/payload.go

@@ -43,6 +43,7 @@ type NetworkDevicesMetadata struct {
 	NetflowExporters []NetflowExporter      `json:"netflow_exporters,omitempty"`
 	Diagnoses        []DiagnosisMetadata    `json:"diagnoses,omitempty"`
 	DeviceOIDs       []DeviceOID            `json:"device_oids,omitempty"`
+	DeviceScanStatus *ScanStatusMetadata    `json:"scan_status,omitempty"`
 	CollectTimestamp int64                  `json:"collect_timestamp"`
 }
 
@@ -81,6 +82,24 @@ type DeviceOID struct {
 	Value    string `json:"value"`
 }
 
+// ScanStatus type for the different possible scan statuses
+type ScanStatus string
+
+const (
+	// ScanStatusInProgress represents a scan in progress
+	ScanStatusInProgress ScanStatus = "in progress"
+	// ScanStatusCompleted represents a completed scan
+	ScanStatusCompleted ScanStatus = "completed"
+	// ScanStatusError represents a scan error
+	ScanStatusError ScanStatus = "error"
+)
+
+// ScanStatusMetadata contains scan status metadata
+type ScanStatusMetadata struct {
+	DeviceID   string     `json:"device_id"`
+	ScanStatus ScanStatus `json:"scan_status"`
+}
+
 // InterfaceMetadata contains interface metadata
 type InterfaceMetadata struct {
 	DeviceID      string        `json:"device_id"`