Merge pull request #12527 from zhaoxiangchun/feature/zxc-azure-metric

feat(monitor,cloudmon): add azure rds,redis,lb metric
2026-09-24 16:03:43 +08:00 · 2021-10-26 19:34:20 +08:00
parent 20be215e40 e20c47a0f9
commit ea7540cb14
6 changed files with 157 additions and 36 deletions
@@ -29,6 +29,7 @@ const (
 	METRIC_RES_TYPE_TENANT       = "tenant"
 	METRIC_RES_TYPE_DOMAIN       = "domain"
 	METRIC_RES_TYPE_STORAGE      = "storage"
+	METRIC_RES_TYPE_ELB          = "elb"
 	//ext is prefix！
 	METRIC_RES_TYPE_JENKINS      = "ext_jenkins"
 	METRIC_RES_TYPE_EXT_MYSQL    = "ext_mysql"
@@ -18,7 +18,6 @@ import (
 	"yunion.io/x/onecloud/pkg/apis/compute"
 	"yunion.io/x/onecloud/pkg/cloudmon/collectors/common"
 	"yunion.io/x/onecloud/pkg/mcclient"
-	"yunion.io/x/onecloud/pkg/mcclient/modules"
 )

 func init() {
@@ -50,7 +49,7 @@ type SAzureCloudReport struct {
 }

 func (self *SAzureCloudReport) Report() error {
-	servers, err := self.GetAllserverOfThisProvider(&modules.Servers)
+	servers, err := self.GetResourceByOperator()
 	if err != nil {
 		return err
 	}
@@ -64,19 +63,19 @@ func (self *SAzureCloudReport) Report() error {
 	}
 	for _, region := range regionList {
 		servers := regionServerMap[region.GetGlobalId()]
-		switch self.Operator {
-		case "server":
-			err = common.CollectRegionMetricAsync(self.Args.Batch, region, servers, self)
-			//err = self.collectRegionMetricOfHost(region, servers)
-			//case "redis":
-			//	err = self.collectRegionMetricOfRedis(region, servers)
-			//case "rds":
-			//	err = self.collectRegionMetricOfRds(region, servers)
-			//case "oss":
-			//	err = self.collectRegionMetricOfOss(region, servers)
-			//case "elb":
-			//	err = self.collectRegionMetricOfElb(region, servers)
-		}
+		err = common.CollectRegionMetricAsync(self.Args.Batch, region, servers, self)
+		//switch self.Operator {
+		//case "server":
+		//	//err = self.collectRegionMetricOfHost(region, servers)
+		//	//case "redis":
+		//	//	err = self.collectRegionMetricOfRedis(region, servers)
+		//	//case "rds":
+		//	//	err = self.collectRegionMetricOfRds(region, servers)
+		//	//case "oss":
+		//	//	err = self.collectRegionMetricOfOss(region, servers)
+		//	//case "elb":
+		//	//	err = self.collectRegionMetricOfElb(region, servers)
+		//}
 		if err != nil {
 			return err
 		}
@@ -15,6 +15,7 @@
 package azuremon

 import (
+	"fmt"
 	"strconv"
 	"strings"

@@ -22,6 +23,7 @@ import (
 	"yunion.io/x/log"
 	"yunion.io/x/pkg/errors"

+	com_api "yunion.io/x/onecloud/pkg/apis/compute"
 	"yunion.io/x/onecloud/pkg/cloudmon/collectors/common"
 	"yunion.io/x/onecloud/pkg/cloudprovider"
 	"yunion.io/x/onecloud/pkg/multicloud/azure"
@@ -50,8 +52,7 @@ func (self *SAzureCloudReport) collectRegionMetricOfHost(region cloudprovider.IC
 			continue
 		}
 		classicKey := "microsoft.classiccompute/virtualmachines"
-		ns := "Microsoft.Compute/virtualMachines"
-		metricSpecs := azureMetricSpecs
+		ns, metricSpecs := self.getMetricSpecs(server)
 		if strings.Contains(strings.ToLower(external_id), classicKey) {
 			ns = classicKey
 			metricSpecs = azureClassicMetricsSpec
@@ -75,11 +76,13 @@ func (self *SAzureCloudReport) collectRegionMetricOfHost(region cloudprovider.IC
 			for _, value := range *rtnMetrics.Value {
 				if value.Name.LocalizedValue != nil {
 					if metricName == *(value.Name.LocalizedValue) || (value.Name.Value != nil && *value.Name.Value == metricName) {
-						metric, err := common.FillVMCapacity(server.(*jsonutils.JSONDict))
-						if err != nil {
-							return errors.Wrapf(err, "fill vm %q capacity", srvPrefix)
+						if self.Operator == string(common.SERVER) {
+							metric, err := common.FillVMCapacity(server.(*jsonutils.JSONDict))
+							if err != nil {
+								return errors.Wrapf(err, "fill vm %q capacity", srvPrefix)
+							}
+							dataList = append(dataList, metric)
 						}
-						dataList = append(dataList, metric)
 						if value.Timeseries != nil {
 							for _, timeserie := range *value.Timeseries {
 								serverMetric, err := self.collectMetricFromThisServer(server, timeserie, influxDbSpecs)
@@ -106,7 +109,7 @@ func (self *SAzureCloudReport) collectMetricFromThisServer(server jsonutils.JSON
 	rtnMetric azure.TimeSeriesElement, influxDbSpecs []string) ([]influxdb.SMetricData, error) {
 	datas := make([]influxdb.SMetricData, 0)
 	for _, data := range *rtnMetric.Data {
-		metric, err := common.JsonToMetric(server.(*jsonutils.JSONDict), "", common.ServerTags, make([]string, 0))
+		metric, err := self.NewMetricFromJson(server)
 		if err != nil {
 			return nil, err
 		}
@@ -153,3 +156,39 @@ func (self *SAzureCloudReport) collectMetricFromThisServer(server jsonutils.JSON

 	return datas, nil
 }
+
+func (self *SAzureCloudReport) getMetricSpecs(res jsonutils.JSONObject) (string, map[string][]string) {
+	switch common.MonType(self.Operator) {
+	case common.SERVER:
+		return SERVER_METRIC_NAMESPACE, azureMetricSpecs
+	case common.REDIS:
+		return REDIS_METRIC_NAMESPACE, azureRedisMetricsSpec
+	case common.RDS:
+		return self.getRdsMetricSpecsByEngine(res)
+	case common.ELB:
+		return ELB_METRIC_NAMESPACE, azureElbMetricSpecs
+	default:
+		return SERVER_METRIC_NAMESPACE, azureMetricSpecs
+	}
+}
+
+func (self *SAzureCloudReport) getRdsMetricSpecsByEngine(res jsonutils.JSONObject) (string, map[string][]string) {
+	engine, _ := res.GetString("engine")
+	externalId, _ := res.GetString("external_id")
+	suffix := "servers"
+	if strings.Contains(externalId, "flexible") {
+		suffix = "flexibleServers"
+	}
+	switch engine {
+	case com_api.DBINSTANCE_TYPE_SQLSERVER:
+		return "Microsoft.Sql/servers/databases", azureRdsMetricsSpec
+	case com_api.DBINSTANCE_TYPE_MYSQL:
+		return fmt.Sprintf("Microsoft.DBforMySQL/%s", suffix), azureRdsMetricsSpec
+	case com_api.DBINSTANCE_TYPE_POSTGRESQL:
+		return fmt.Sprintf("Microsoft.DBforPostgreSQL/%s", suffix), azureRdsMetricsSpec
+	case com_api.DBINSTANCE_TYPE_MARIADB:
+		return "Microsoft.DBforMariaDB/servers", azureRdsMetricsSpec
+	default:
+		return fmt.Sprintf("Microsoft.DBforMySQL/%s", suffix), azureRdsMetricsSpec
+	}
+}
@@ -27,6 +27,7 @@ const (
 	UNIT_MEM           = "byte"
 	UNIT_MSEC          = "ms"
 	UNIT_COUNT_SEC     = "count/s"
+	UNIT_BYTES         = "byte"

 	//ESC监控指标
 	INFLUXDB_FIELD_CPU_USAGE           = "vm_cpu.usage_active"
@@ -58,8 +59,11 @@ const (
 	INFLUXDB_FIELD_RDS_DISK_USAGE           = "rds_disk.used_percent"
 	INFLUXDB_FIELD_RDS_DISK_READ_BPS        = "rds_diskio.read_bps"
 	INFLUXDB_FIELD_RDS_DISK_WRITE_BPS       = "rds_diskio.write_bps"
+	INFLUXDB_FIELD_RDS_DISK_IO_PERSENT      = "rds_diskio.used_percent"
 	INFLUXDB_FIELD_RDS_CONN_COUNT           = "rds_conn.used_count"
 	INFLUXDB_FIELD_RDS_CONN_USAGE           = "rds_conn.used_percent"
+	INFLUXDB_FIELD_RDS_CONN_ACTIVE          = "rds_conn.active_count"
+	INFLUXDB_FIELD_RDS_CONN_FAILED          = "rds_conn.failed_count"

 	INFLUXDB_FIELD_RDS_QPS              = "rds_qps.query_qps"
 	INFLUXDB_FIELD_RDS_TPS              = "rds_tps.trans_qps"
@@ -76,6 +80,8 @@ const (
 	INFLUXDB_FIFLD_REDIS_CACHE_KEYS     = "dcs_cachekeys.key_count"
 	INFLUXDB_FIFLD_REDIS_CACHE_EXP_KEYS = INFLUXDB_FIFLD_REDIS_CACHE_KEYS + ",exp=expire"
 	INFLUXDB_FIFLD_REDIS_DATA_MEM_USAGE = "dcs_datamem.used_byte"
+	INFLUXDB_FIFLD_REDIS_SERVER_LOAD    = "dcs_cpu.server_load"
+	INFLUXDB_FIFLD_REDIS_CONN_ERRORS    = "dcs_conn.errors"

 	//对象存储OSS监控指标
 	INFLUXDB_FIELD_OSS_NET_BPS_RX          = "oss_netio.bps_recv"
@@ -94,20 +100,22 @@ const (
 	INFLUXDB_FIELD_OSS_REQ_COUNT_4XX       = INFLUXDB_FIELD_OSS_REQ_COUNT + ",request=4xx"

 	//负载均衡监控指标
-	INFLUXDB_FIELD_ELB_NET_BPS_RX     = "haproxy.bin"
-	INFLUXDB_FIELD_ELB_NET_BPS_TX     = "haproxy.bout"
-	INFLUXDB_FIELD_ELB_REQ_RATE       = "haproxy.req_rate,request=http"
-	INFLUXDB_FIELD_ELB_CONN_RATE      = "haproxy.conn_rate,request=tcp"
-	INFLUXDB_FIELD_ELB_DREQ_COUNT     = "haproxy.dreq,request=http"
-	INFLUXDB_FIELD_ELB_DCONN_COUNT    = "haproxy.dcon,request=tcp"
-	INFLUXDB_FIELD_ELB_HRSP_COUNT     = "haproxy.hrsp_Nxx"
-	INFLUXDB_FIELD_ELB_HRSP_COUNT_2XX = INFLUXDB_FIELD_ELB_HRSP_COUNT + ",request=2xx"
-	INFLUXDB_FIELD_ELB_HRSP_COUNT_3XX = INFLUXDB_FIELD_ELB_HRSP_COUNT + ",request=3xx"
-	INFLUXDB_FIELD_ELB_HRSP_COUNT_4XX = INFLUXDB_FIELD_ELB_HRSP_COUNT + ",request=4xx"
-	INFLUXDB_FIELD_ELB_HRSP_COUNT_5XX = INFLUXDB_FIELD_ELB_HRSP_COUNT + ",request=5xx"
-	INFLUXDB_FIELD_ELB_CHC_STATUS     = "haproxy.check_status"
-	INFLUXDB_FIELD_ELB_CHC_CODE       = "haproxy.check_code"
-	INFLUXDB_FIELD_ELB_LAST_CHC       = "haproxy.last_chk"
+	INFLUXDB_FIELD_ELB_NET_BPS_RX      = "haproxy.bin"
+	INFLUXDB_FIELD_ELB_NET_BPS_TX      = "haproxy.bout"
+	INFLUXDB_FIELD_ELB_REQ_RATE        = "haproxy.req_rate,request=http"
+	INFLUXDB_FIELD_ELB_CONN_RATE       = "haproxy.conn_rate,request=tcp"
+	INFLUXDB_FIELD_ELB_DREQ_COUNT      = "haproxy.dreq,request=http"
+	INFLUXDB_FIELD_ELB_DCONN_COUNT     = "haproxy.dcon,request=tcp"
+	INFLUXDB_FIELD_ELB_HRSP_COUNT      = "haproxy.hrsp_Nxx"
+	INFLUXDB_FIELD_ELB_HRSP_COUNT_2XX  = INFLUXDB_FIELD_ELB_HRSP_COUNT + ",request=2xx"
+	INFLUXDB_FIELD_ELB_HRSP_COUNT_3XX  = INFLUXDB_FIELD_ELB_HRSP_COUNT + ",request=3xx"
+	INFLUXDB_FIELD_ELB_HRSP_COUNT_4XX  = INFLUXDB_FIELD_ELB_HRSP_COUNT + ",request=4xx"
+	INFLUXDB_FIELD_ELB_HRSP_COUNT_5XX  = INFLUXDB_FIELD_ELB_HRSP_COUNT + ",request=5xx"
+	INFLUXDB_FIELD_ELB_CHC_STATUS      = "haproxy.check_status"
+	INFLUXDB_FIELD_ELB_CHC_CODE        = "haproxy.check_code"
+	INFLUXDB_FIELD_ELB_LAST_CHC        = "haproxy.last_chk"
+	INFLUXDB_FIELD_ELB_SNAT_PORT       = "haproxy.used_snat_port"
+	INFLUXDB_FIELD_ELB_SNAT_CONN_COUNT = "haproxy.snat_conn_count"

 	KEY_VMS   = "vms"
 	KEY_CPUS  = "cpus"
@@ -133,6 +141,10 @@ const (
 	MetricKeyClassicNetworkOut   = "Network Out"
 	MetricKeyClassicDiskReadBPS  = "Disk Read Bytes/Sec"
 	MetricKeyClassicDiskWriteBPS = "Disk Write Bytes/Sec"
+
+	SERVER_METRIC_NAMESPACE = "Microsoft.Compute/virtualMachines"
+	REDIS_METRIC_NAMESPACE  = "Microsoft.Cache/redis"
+	ELB_METRIC_NAMESPACE    = "Microsoft.Network/loadBalancers"
 )

 var azureMetricSpecs = map[string][]string{
@@ -154,3 +166,39 @@ var azureClassicMetricsSpec = map[string][]string{
 	MetricKeyDiskReadOPS:         {DEFAULT_STATISTICS, UNIT_COUNT_SEC, INFLUXDB_FIELD_DISK_READ_IOPS},
 	MetricKeyDiskWriteOPS:        {DEFAULT_STATISTICS, UNIT_COUNT_SEC, INFLUXDB_FIELD_DISK_WRITE_IOPS},
 }
+
+var azureRedisMetricsSpec = map[string][]string{
+	"percentProcessorTime": {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIELD_REDIS_CPU_USAGE},
+	"usedmemorypercentage": {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIELD_REDIS_MEM_USAGE},
+	"connectedclients":     {DEFAULT_STATISTICS, UNIT_COUNT, INFLUXDB_FIFLD_REDIS_CONN_USAGE},
+	"operationsPerSecond":  {DEFAULT_STATISTICS, UNIT_COUNT, INFLUXDB_FIFLD_REDIS_OPT_SES},
+	"alltotalkeys":         {DEFAULT_STATISTICS, UNIT_COUNT, INFLUXDB_FIFLD_REDIS_CACHE_KEYS},
+	"expiredkeys":          {DEFAULT_STATISTICS, UNIT_COUNT, INFLUXDB_FIFLD_REDIS_CACHE_EXP_KEYS},
+	"usedmemory":           {DEFAULT_STATISTICS, UNIT_BYTES, INFLUXDB_FIFLD_REDIS_DATA_MEM_USAGE},
+	"serverLoad":           {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIFLD_REDIS_SERVER_LOAD},
+	"errors":               {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIFLD_REDIS_CONN_ERRORS},
+}
+
+//mariadb,mysql,postgresql
+var azureRdsMetricsSpec = map[string][]string{
+	"cpu_percent":            {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIELD_RDS_CPU_USAGE},
+	"memory_percent":         {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIELD_RDS_MEM_USAGE},
+	"storage_percent":        {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIELD_RDS_DISK_USAGE},
+	"network_bytes_ingress":  {DEFAULT_STATISTICS, UNIT_BYTES, INFLUXDB_FIELD_RDS_NET_BPS_RX},
+	"network_bytes_egress":   {DEFAULT_STATISTICS, UNIT_BYTES, INFLUXDB_FIELD_RDS_NET_BPS_TX},
+	"io_consumption_percent": {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIELD_RDS_DISK_IO_PERSENT},
+	"active_connections":     {DEFAULT_STATISTICS, UNIT_COUNT, INFLUXDB_FIELD_RDS_CONN_ACTIVE},
+	"connections_failed":     {DEFAULT_STATISTICS, UNIT_COUNT, INFLUXDB_FIELD_RDS_CONN_FAILED},
+}
+
+var azureRdsMetricsSpecSqlserver = map[string][]string{
+	"cpu_percent":          {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIELD_RDS_CPU_USAGE},
+	"memory_usage_percent": {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIELD_RDS_MEM_USAGE},
+	"storage_percent":      {DEFAULT_STATISTICS, UNIT_PERCENT, INFLUXDB_FIELD_RDS_DISK_USAGE},
+	"connections_failed":   {DEFAULT_STATISTICS, UNIT_COUNT, INFLUXDB_FIELD_RDS_CONN_FAILED},
+}
+
+var azureElbMetricSpecs = map[string][]string{
+	"SnatConnectionCount": {DEFAULT_STATISTICS, UNIT_COUNT, INFLUXDB_FIELD_ELB_SNAT_PORT},
+	"UsedSnatPorts":       {DEFAULT_STATISTICS, UNIT_COUNT, INFLUXDB_FIELD_ELB_SNAT_CONN_COUNT},
+}
@@ -47,6 +47,24 @@ func (self *CloudReportBase) Report() error {
 	return fmt.Errorf("No Implment the method")
 }

+func (self *CloudReportBase) GetResourceByOperator() ([]jsonutils.JSONObject, error) {
+	var servers []jsonutils.JSONObject
+	var err error
+	switch MonType(self.Operator) {
+	case REDIS:
+		servers, err = self.GetAllserverOfThisProvider(&modules.ElasticCache)
+	case RDS:
+		servers, err = self.GetAllserverOfThisProvider(&modules.DBInstance)
+	case OSS:
+		servers, err = self.GetAllserverOfThisProvider(&modules.Buckets)
+	case ELB:
+		servers, err = self.GetAllserverOfThisProvider(&modules.Loadbalancers)
+	default:
+		servers, err = self.GetAllserverOfThisProvider(&modules.Servers)
+	}
+	return servers, err
+}
+
 func (self *CloudReportBase) GetAllserverOfThisProvider(manager modulebase.Manager) ([]jsonutils.JSONObject, error) {
 	query := jsonutils.NewDict()

@@ -229,10 +229,18 @@ func init() {
 			newMetricFieldCreateInput("used_percent", "Percentage of used disks", monitor.METRIC_UNIT_PERCENT, 1),
 		})

+	// rds_conn
+	RegistryMetricCreateInput("rds_conn", "Rds connection", monitor.METRIC_RES_TYPE_RDS,
+		monitor.METRIC_DATABASE_TELE, 5, []monitor.MetricFieldCreateInput{
+			newMetricFieldCreateInput("active_count", "active connection count", monitor.METRIC_UNIT_COUNT, 1),
+			newMetricFieldCreateInput("failed_count", "failed connection count", monitor.METRIC_UNIT_COUNT, 2),
+		})
+
 	// dcs_cpu
 	RegistryMetricCreateInput("dcs_cpu", "Redis CPU usage", monitor.METRIC_RES_TYPE_REDIS,
 		monitor.METRIC_DATABASE_TELE, 1, []monitor.MetricFieldCreateInput{
 			newMetricFieldCreateInput("usage_percent", "CPU active state utilization rate", monitor.METRIC_UNIT_PERCENT, 1),
+			newMetricFieldCreateInput("server_load", "server load", monitor.METRIC_UNIT_PERCENT, 2),
 		})

 	// dcs_mem
@@ -253,6 +261,7 @@ func init() {
 	RegistryMetricCreateInput("dcs_conn", "Redis connect", monitor.METRIC_RES_TYPE_REDIS,
 		monitor.METRIC_DATABASE_TELE, 5, []monitor.MetricFieldCreateInput{
 			newMetricFieldCreateInput("used_percent", "Connection usage", monitor.METRIC_UNIT_PERCENT, 1),
+			newMetricFieldCreateInput("errors", "Connection errors", monitor.METRIC_UNIT_COUNT, 1),
 		})

 	// dcs_instantopt
@@ -280,6 +289,13 @@ func init() {
 			newMetricFieldCreateInput("balance", "balance", monitor.METRIC_UNIT_RMB, 1),
 		})

+	// haproxy
+	RegistryMetricCreateInput("haproxy", "load balance", monitor.METRIC_RES_TYPE_ELB,
+		monitor.METRIC_DATABASE_TELE, 1, []monitor.MetricFieldCreateInput{
+			newMetricFieldCreateInput("used_snat_port", "used SNAT port count", monitor.METRIC_UNIT_COUNT, 1),
+			newMetricFieldCreateInput("snat_conn_count", "SNAT connection count", monitor.METRIC_UNIT_COUNT, 1),
+		})
+
 	// agent_cpu
 	RegistryMetricCreateInput("agent_cpu", "CPU usage", monitor.METRIC_RES_TYPE_AGENT, monitor.METRIC_DATABASE_TELE, 1,
 		[]monitor.MetricFieldCreateInput{