chore: remaining metrics

frontend/src/container/InfraMonitoringHosts/InfraMonitoring.styles.scss

@@ -190,6 +190,11 @@
 			.ant-table-cell:nth-child(n + 3) {
 				padding-right: 24px;
 			}
+			.status-header {
+				display: flex;
+				align-items: center;
+				gap: 4px;
+			}
 			.memory-usage-header {
 				display: flex;
 				align-items: center;

frontend/src/container/InfraMonitoringHosts/utils.tsx

@@ -135,7 +135,14 @@ export const getHostsListColumns = (): ColumnType<HostRowData>[] => [
 		),
 	},
 	{
-		title: 'Status',
+		title: (
+			<div className="status-header">
+				Status
+				<Tooltip title="Sent system metrics in last 10 mins">
+					<InfoCircleOutlined />
+				</Tooltip>
+			</div>
+		),
 		dataIndex: 'active',
 		key: 'active',
 		width: 100,

pkg/query-service/app/clickhouseReader/reader.go

@@ -3541,6 +3541,45 @@ func (r *ClickHouseReader) GetCountOfThings(ctx context.Context, query string) (
 	return count, nil
 }
 
+func (r *ClickHouseReader) GetActiveHostsFromMetricMetadata(ctx context.Context, metricNames []string, hostNameAttr string, sinceUnixMilli int64) (map[string]bool, error) {
+	activeHosts := map[string]bool{}
+
+	query := fmt.Sprintf(
+		`SELECT DISTINCT attr_string_value
+		FROM %s.%s
+		WHERE metric_name IN @metricNames
+		  AND attr_name = @attrName
+		  AND last_reported_unix_milli >= @sinceUnixMilli`,
+		signozMetricDBName,
+		constants.SIGNOZ_METRICS_METADATA_TABLENAME,
+	)
+
+	rows, err := r.db.Query(ctx, query,
+		clickhouse.Named("metricNames", metricNames),
+		clickhouse.Named("attrName", hostNameAttr),
+		clickhouse.Named("sinceUnixMilli", sinceUnixMilli),
+	)
+	if err != nil {
+		return nil, errorsV2.WrapInternalf(err, errorsV2.CodeInternal, "error querying active hosts")
+	}
+	defer rows.Close()
+
+	for rows.Next() {
+		var hostName string
+		if err := rows.Scan(&hostName); err != nil {
+			return nil, errorsV2.WrapInternalf(err, errorsV2.CodeInternal, "error scanning active host row")
+		}
+		if hostName != "" {
+			activeHosts[hostName] = true
+		}
+	}
+	if err := rows.Err(); err != nil {
+		return nil, errorsV2.WrapInternalf(err, errorsV2.CodeInternal, "error iterating active host rows")
+	}
+
+	return activeHosts, nil
+}
+
 func (r *ClickHouseReader) GetLatestReceivedMetric(
 	ctx context.Context, metricNames []string, labelValues map[string]string,
 ) (*model.MetricStatus, *model.ApiError) {

pkg/query-service/app/inframetrics/common.go

@@ -10,9 +10,47 @@ import (
 )
 
 var dotMetricMap = map[string]string{
-	"system_cpu_time":                       "system.cpu.time",
-	"system_memory_usage":                   "system.memory.usage",
-	"system_cpu_load_average_15m":           "system.cpu.load_average.15m",
+	"system_uptime":                  "system.uptime",
+	"system_cpu_physical_count":      "system.cpu.physical.count",
+	"system_cpu_logical_count":       "system.cpu.logical.count",
+	"system_cpu_time":                "system.cpu.time",
+	"system_cpu_frequency":           "system.cpu.frequency",
+	"system_cpu_utilization":         "system.cpu.utilization",
+	"system_cpu_load_average_15m":    "system.cpu.load_average.15m",
+	"system_memory_usage":            "system.memory.usage",
+	"system_memory_limit":            "system.memory.limit",
+	"system_memory_utilization":      "system.memory.utilization",
+	"system_memory_linux_available":  "system.memory.linux.available",
+	"system_memory_linux_shared":     "system.memory.linux.shared",
+	"system_memory_linux_slab_usage": "system.memory.linux.slab.usage",
+	"system_paging_usage":            "system.paging.usage",
+	"system_paging_utilization":      "system.paging.utilization",
+	"system_paging_faults":           "system.paging.faults",
+	"system_paging_operations":       "system.paging.operations",
+	"system_disk_io":                 "system.disk.io",
+	"system_disk_operations":         "system.disk.operations",
+	"system_disk_io_time":            "system.disk.io_time",
+	"system_disk_operation_time":     "system.disk.operation_time",
+	"system_disk_merged":             "system.disk.merged",
+	"system_disk_limit":              "system.disk.limit",
+	"system_filesystem_usage":        "system.filesystem.usage",
+	"system_filesystem_utilization":  "system.filesystem.utilization",
+	"system_filesystem_limit":        "system.filesystem.limit",
+	"system_network_packet_dropped":  "system.network.packet.dropped",
+	"system_network_packet_count":    "system.network.packet.count",
+	"system_network_errors":          "system.network.errors",
+	"system_network_io":              "system.network.io",
+	"system_network_connection_count": "system.network.connection.count",
+	"system_process_count":            "system.process.count",
+	"system_process_created":          "system.process.created",
+	"system_disk_pending_operations":  "system.disk.pending_operations",
+	"system_disk_weighted_io_time":    "system.disk.weighted_io_time",
+	"system_filesystem_inodes_usage":  "system.filesystem.inodes.usage",
+	"system_network_conntrack_count":  "system.network.conntrack.count",
+	"system_network_conntrack_max":    "system.network.conntrack.max",
+	"system_cpu_load_average_1m":      "system.cpu.load_average.1m",
+	"system_cpu_load_average_5m":      "system.cpu.load_average.5m",
+
 	"host_name":                             "host.name",
 	"k8s_cluster_name":                      "k8s.cluster.name",
 	"k8s_node_name":                         "k8s.node.name",

pkg/query-service/app/inframetrics/hosts.go

@@ -72,6 +72,48 @@ var (
 		"load15": GetDotMetrics("system_cpu_load_average_15m"),
 		"wait":   GetDotMetrics("system_cpu_time"),
 	}
+	uniqueMetricNamesForHosts = []string{
+		GetDotMetrics("system_uptime"),
+		GetDotMetrics("system_cpu_physical_count"),
+		GetDotMetrics("system_cpu_logical_count"),
+		GetDotMetrics("system_cpu_time"),
+		GetDotMetrics("system_cpu_frequency"),
+		GetDotMetrics("system_cpu_utilization"),
+		GetDotMetrics("system_cpu_load_average_15m"),
+		GetDotMetrics("system_memory_usage"),
+		GetDotMetrics("system_memory_limit"),
+		GetDotMetrics("system_memory_utilization"),
+		GetDotMetrics("system_memory_linux_available"),
+		GetDotMetrics("system_memory_linux_shared"),
+		GetDotMetrics("system_memory_linux_slab_usage"),
+		GetDotMetrics("system_paging_usage"),
+		GetDotMetrics("system_paging_utilization"),
+		GetDotMetrics("system_paging_faults"),
+		GetDotMetrics("system_paging_operations"),
+		GetDotMetrics("system_disk_io"),
+		GetDotMetrics("system_disk_operations"),
+		GetDotMetrics("system_disk_io_time"),
+		GetDotMetrics("system_disk_operation_time"),
+		GetDotMetrics("system_disk_merged"),
+		GetDotMetrics("system_disk_limit"),
+		GetDotMetrics("system_filesystem_usage"),
+		GetDotMetrics("system_filesystem_utilization"),
+		GetDotMetrics("system_filesystem_limit"),
+		GetDotMetrics("system_network_packet_dropped"),
+		GetDotMetrics("system_network_packet_count"),
+		GetDotMetrics("system_network_errors"),
+		GetDotMetrics("system_network_io"),
+		GetDotMetrics("system_network_connection_count"),
+		GetDotMetrics("system_process_count"),
+		GetDotMetrics("system_process_created"),
+		GetDotMetrics("system_disk_pending_operations"),
+		GetDotMetrics("system_disk_weighted_io_time"),
+		GetDotMetrics("system_filesystem_inodes_usage"),
+		GetDotMetrics("system_network_conntrack_count"),
+		GetDotMetrics("system_network_conntrack_max"),
+		GetDotMetrics("system_cpu_load_average_1m"),
+		GetDotMetrics("system_cpu_load_average_5m"),
+	}
 )
 
 func NewHostsRepo(reader interfaces.Reader, querierV2 interfaces.Querier) *HostsRepo {
@@ -130,62 +172,9 @@ func (h *HostsRepo) GetHostAttributeValues(ctx context.Context, req v3.FilterAtt
 	return &v3.FilterAttributeValueResponse{StringAttributeValues: hostNames}, nil
 }
 
-func (h *HostsRepo) getActiveHosts(ctx context.Context, orgID valuer.UUID, req model.HostListRequest) (map[string]bool, error) {
-	activeStatus := map[string]bool{}
-	step := common.MinAllowedStepInterval(req.Start, req.End)
-
-	hasHostName := false
-	for _, key := range req.GroupBy {
-		if key.Key == hostNameAttrKey {
-			hasHostName = true
-		}
-	}
-
-	if !hasHostName {
-		req.GroupBy = append(req.GroupBy, v3.AttributeKey{Key: hostNameAttrKey})
-	}
-
-	params := v3.QueryRangeParamsV3{
-		Start: time.Now().Add(-time.Minute * 10).UTC().UnixMilli(),
-		End:   time.Now().UTC().UnixMilli(),
-		Step:  step,
-		CompositeQuery: &v3.CompositeQuery{
-			BuilderQueries: map[string]*v3.BuilderQuery{
-				"A": {
-					QueryName:    "A",
-					StepInterval: step,
-					DataSource:   v3.DataSourceMetrics,
-					AggregateAttribute: v3.AttributeKey{
-						Key:      metricToUseForHostAttributes,
-						DataType: v3.AttributeKeyDataTypeFloat64,
-					},
-					Temporality:      v3.Unspecified,
-					Filters:          req.Filters,
-					GroupBy:          req.GroupBy,
-					Expression:       "A",
-					TimeAggregation:  v3.TimeAggregationAvg,
-					SpaceAggregation: v3.SpaceAggregationAvg,
-					Disabled:         false,
-				},
-			},
-			QueryType: v3.QueryTypeBuilder,
-			PanelType: v3.PanelTypeGraph,
-		},
-	}
-
-	queryResponse, _, err := h.querierV2.QueryRange(ctx, orgID, &params)
-	if err != nil {
-		return nil, err
-	}
-
-	for _, result := range queryResponse {
-		for _, series := range result.Series {
-			name := series.Labels[hostNameAttrKey]
-			activeStatus[name] = true
-		}
-	}
-
-	return activeStatus, nil
+func (h *HostsRepo) getActiveHosts(ctx context.Context) (map[string]bool, error) {
+	tenMinAgo := time.Now().Add(-10 * time.Minute).UTC().UnixMilli()
+	return h.reader.GetActiveHostsFromMetricMetadata(ctx, uniqueMetricNamesForHosts, hostNameAttrKey, tenMinAgo)
 }
 
 func (h *HostsRepo) getMetadataAttributes(ctx context.Context, req model.HostListRequest) (map[string]map[string]string, error) {
@@ -441,7 +430,7 @@ func (h *HostsRepo) GetHostList(ctx context.Context, orgID valuer.UUID, req mode
 		return resp, err
 	}
 
-	activeHosts, err := h.getActiveHosts(ctx, orgID, req)
+	activeHosts, err := h.getActiveHosts(ctx)
 	if err != nil {
 		return resp, err
 	}

pkg/query-service/constants/constants.go

@@ -125,6 +125,7 @@ const (
 	SIGNOZ_TIMESERIES_v4_6HRS_TABLENAME        = "distributed_time_series_v4_6hrs"
 	SIGNOZ_ATTRIBUTES_METADATA_TABLENAME       = "distributed_attributes_metadata"
 	SIGNOZ_ATTRIBUTES_METADATA_LOCAL_TABLENAME = "attributes_metadata"
+	SIGNOZ_METRICS_METADATA_TABLENAME          = "distributed_metadata"
 )
 
 // alert related constants

pkg/query-service/interfaces/interface.go

@@ -99,6 +99,7 @@ type Reader interface {
 	SubscribeToQueryProgress(queryId string) (<-chan model.QueryProgress, func(), *model.ApiError)
 
 	GetCountOfThings(ctx context.Context, query string) (uint64, error)
+	GetActiveHostsFromMetricMetadata(ctx context.Context, metricNames []string, hostNameAttr string, sinceUnixMilli int64) (map[string]bool, error)
 
 	//trace
 	GetTraceFields(ctx context.Context) (*model.GetFieldsResponse, *model.ApiError)