Docs

Volcengine Kubernetes Engine

Full stack observation

Metric references

Full-stack observability supports the collection of various network-layer and application-layer metrics. This topic describes supported metrics.

Background

After you turn on full-stack observability and configure metric collection, the system collects the corresponding eBPF metrics based on your configuration. For information about configuring metric collection, see Enabling full-stack observability.

warning

Metrics collected by full-stack observability are stored in VMP workspaces for metric queries, dashboard creation, and alerting. All these metrics are non-basic metrics (that is, chargeable metrics). Configure metric collection based on your actual needs. For VMP billing details, see Billing overview.

Metrics

Topology metrics

The cluster topology feature in full-stack observability depends on the following metric, which is enabled by default and cannot be disabled. For details, see Cluster topology.

Metric description	Metric name	Unit	Metric type	Metric source
Network topology	ms_network_topology	-	gauge	eBPF

Network metrics

Network metrics include TCP and UDP metrics. You can enable or disable the corresponding metrics as needed. The following table describes these metrics.

Metric description	Metric name	Unit	Metric type	Metric source
Number of TCP connections initiated	ms_tcp_conn_open_total	Count	counter	eBPF
Number of TCP connections established	ms_tcp_conn_total	Count	counter	eBPF
TCP connection duration	ms_tcp_conn_duration	Milliseconds	histogram	eBPF
Number of TCP connections closed due to errors	ms_tcp_conn_error_closed_total	Count	counter	eBPF
Inbound TCP traffic	ms_tcp_recv_bytes_total	Bytes	counter	eBPF
Outbound TCP traffic	ms_tcp_sent_bytes_total	Bytes	counter	eBPF
Number of inbound TCP packets	ms_tcp_recv_packets_total	Count	counter	eBPF
Number of outbound TCP packets	ms_tcp_sent_packets_total	Count	counter	eBPF
Number of lost TCP packets	ms_tcp_packets_loss_total	Count	counter	eBPF
TCP retransmission count	ms_tcp_retrans_total	Count	counter	eBPF
Inbound UDP traffic	ms_udp_recv_bytes_total	Bytes	counter	eBPF
Outbound UDP traffic	ms_udp_sent_bytes_total	Bytes	counter	eBPF
Inbound UDP datagrams	ms_udp_recv_datagrams_total	Count	counter	eBPF
Outbound UDP datagrams	ms_udp_sent_datagrams_total	Count	counter	eBPF
Number of UDP datagrams that failed to be received.	ms_udp_recv_datagrams_error_total	Count	counter	eBPF
Number of UDP datagrams that failed to be sent.	ms_udp_sent_datagrams_error_total	Count	counter	eBPF

Application metrics

Application metrics include metrics related to application-layer protocols, such as HTTP and DNS. You can enable or disable the corresponding metrics as needed. The following table describes these metrics.

Metric description	Metric name	Unit	Metric type	Metric source
Application-layer API request count	ms_requests_total	Count	counter	eBPF
Application-layer API request latency	ms_request_duration	Milliseconds	histogram	eBPF
Number of MQ requests (including Redis Pub/Sub)	ms_messaging_io_total	Count	counter	eBPF
MQ request duration (including Redis Pub/Sub)	ms_messaging_io_duration	Milliseconds	histogram	eBPF
Number of application-layer gRPC Streaming messages sent and received	ms_streaming_message_total	Count	counter	eBPF
Application layer GRPC Streaming sent/received message size	ms_streaming_message_size	Bytes	counter	eBPF
Total number of application-layer protocol records successfully parsed and processed	ms_agent_records_process_total	Count	counter	eBPF

tip

Application metrics do not fully match the configurations in the console. The application layer protocol of the metric is displayed only via the traffic_protocol label. For a description of the metric label, see the section below.

You can deselect unnecessary application layer protocols in the console to reduce the number of collected metrics.

Metric labels

Various labels are added to full-stack observability metrics to indicate the dimensions of traffic. For example, the source_pod label indicates the source Pod for the metric in the request. The following table describes metric labels.

tip

The labels supported by different metrics vary slightly.

Label name	Label description
source_workload	Source workload.
source_workload_kind	The kind of the source workload.
source_workload_namespace	The namespace of the source workload.
source_pod	The source pod. This label appears only in client-side metrics.
source_node	The node where the request source is located.
dest_workload	The destination workload.
dest_workload_kind	The kind of the destination workload.
dest_workload_namespace	The namespace of the destination workload.
dest_service	The Kubernetes Service corresponding to the destination of the request. This label appears only when the request destination is a Service.
dest_service_namespace	The namespace of the destination Service. This label appears only when the request destination is a Service.
dest_pod	The destination pod. This label appears only in server-side metrics.
dest_node	The node where the request destination is located.
dest_cluster	The ID of the cluster where the request destination is located.
traffic_protocol	Traffic protocol, for example, http or dns.
traffic_direction	Traffic direction: ingress: Inbound. egress: Outbound.
traffic_role	Traffic role. Possible values: client: Client server: Server
traffic_status_code	HTTP status code. For example, HTTP status codes 2xx, 4xx, and 5xx.
traffic_status	Traffic status. Possible values: ok: Success. error: Fail. unknown: Unknown.
operation	The API name. This label appears only in API metrics.
dns_domain	DNS domain. This label appears only in API metrics.
dns_domain_type	The DNS type, for example, CNAME, A, AAAA, or TXT. This label appears only in API metrics.
is_slow	Whether the traffic is from a slow call. Valid values: true: Slow call. false: Non-slow call.

Last updated: 2026.07.24 10:53:34