Edit

Share via

What is Advanced Network Observability?

  • Article

Advanced Network Observability is a debut feature of the Advanced Container Networking Services suite. It equips you with next-level monitoring and diagnostics tools, providing unparalleled visibility into your containerized workloads. These tools empower you to pinpoint and troubleshoot network issues with ease, ensuring optimal performance for your applications.

Advanced Network Observability is compatible with all Linux workloads seamlessly integrating with Hubble regardless of whether the underlying data plane is Cilium or non-Cilium (both are supported) ensuring flexibility for your container networking needs.

Diagram of Advanced Network Observability.

Note

For Cilium data plane scenarios, Advanced Network Observability is available beginning with Kubernetes version 1.29. For non-Cilium data plane scenarios, Advanced Network Observability is supported on all Linux distributions including Azure Linux beginning with version 2.0.

Features of Advanced Network Observability

Advanced Network Observability offers the following capabilities to monitor network-related issues in your cluster:

  • Node-Level Metrics: Understanding the health of your container network at the node-level is crucial for maintaining optimal application performance. These metrics provide insights into traffic volume, dropped packets, number of connections, etc. by node. The metrics are stored in Prometheus format and, as such, you can view them in Grafana.

  • Hubble Metrics (DNS and Pod-Level Metrics): These Prometheus metrics include source and destination pod information allowing you to pinpoint network-related issues at a granular level. Metrics cover traffic volume, dropped packets, TCP resets, L4/L7 packet flows, etc. There are also DNS metrics (currently only for Non-Cilium data planes), covering DNS errors and DNS requests missing responses.

  • Hubble Flow Logs: Flow logs provide deep visibility into your cluster's network activity. All communications to and from pods are logged allowing you to investigate connectivity issues over time. Flow logs help answer questions such as: did the server receive the client's request? What is the round-trip latency between the client's request and server's response?

    • Hubble CLI: The Hubble Command-Line Interface (CLI) can retrieve flow logs across the entire cluster with customizable filtering and formatting.

    • Hubble UI: Hubble UI is a user-friendly browser-based interface for exploring cluster network activity. It creates a service-connection graph based on flow logs, and displays flow logs for the selected namespace. Users are responsible for provisioning and managing the infrastructure required to run Hubble UI.

Key Benefits of Advanced Network Observability

  • CNI-Agnostic: Supported on all Azure CNI variants including kubenet.

  • Cilium and Non-Cilium: Provides a uniform, seamless experience across both Cilium and non-Cilium data planes.

  • eBPF-Based Network Observability: Leverages eBPF (extended Berkeley Packet Filter) for performance and scalability to identify potential bottlenecks and congestion issues before they impact application performance. Gain insights into key network health indicators, including traffic volume, dropped packets, and connection information.

  • Deep Visibility into Network Activity: Understand how your applications are communicating with each other through detailed network flow logs.

  • Simplified Metrics Storage and Visualization Options: Choose between:

    • Azure Managed Prometheus and Grafana: Azure manages the infrastructure and maintenance, allowing users to focus on configuring metrics and visualizing metrics.
    • Bring Your Own (BYO) Prometheus and Grafana: Users deploy and configure their own instances and manage the underlying infrastructure.

Metrics

Node-Level Metrics

The following metrics are aggregated per node. All metrics include labels:

  • cluster
  • instance (Node name)

For non-Cilium data plane scenarios, Advanced Network Observability provides metrics for both Linux and Windows operating systems. The table below outlines the different metrics generated.

Metric Name Description Extra Labels Linux Windows
networkobservability_forward_count Total forwarded packet count direction ✅ ✅
networkobservability_forward_bytes Total forwarded byte count direction ✅ ✅
networkobservability_drop_count Total dropped packet count direction, reason ✅ ✅
networkobservability_drop_bytes Total dropped byte count direction, reason ✅ ✅
networkobservability_tcp_state TCP currently active socket count by TCP state. state ✅ ✅
networkobservability_tcp_connection_remote TCP currently active socket count by remote IP/port. address (IP), port ✅ ❌
networkobservability_tcp_connection_stats TCP connection statistics. (ex: Delayed ACKs, TCPKeepAlive, TCPSackFailures) statistic ✅ ✅
networkobservability_tcp_flag_counters TCP packets count by flag. flag ❌ ✅
networkobservability_ip_connection_stats IP connection statistics. statistic ✅ ❌
networkobservability_udp_connection_stats UDP connection statistics. statistic ✅ ❌
networkobservability_udp_active_sockets UDP currently active socket count ✅ ❌
networkobservability_interface_stats Interface statistics. InterfaceName, statistic ✅ ✅

Pod-Level Metrics (Hubble Metrics)

The following metrics are aggregated per pod (node information is preserved). All metrics include labels:

  • cluster
  • instance (Node name)
  • source or destination

For outgoing traffic, there will be a source label with source pod namespace/name. For incoming traffic, there will be a destination label with destination pod namespace/name.

Metric Name Description Extra Labels Linux Windows
hubble_dns_queries_total Total DNS requests by query source or destination, query, qtypes (query type) ✅ ❌
hubble_dns_responses_total Total DNS responses by query/response source or destination, query, qtypes (query type), rcode (return code), ips_returned (number of IPs) ✅ ❌
hubble_drop_total Total dropped packet count source or destination, protocol, reason ✅ ❌
hubble_tcp_flags_total Total TCP packets count by flag. source or destination, flag ✅ ❌
hubble_flows_processed_total Total network flows processed (L4/L7 traffic) source or destination, protocol, verdict, type, subtype ✅ ❌

Limitations

  • Pod-level metrics are available only on Linux.
  • Cilium data plane is supported starting with Kubernetes version 1.29.
  • Metric labels may have subtle differences between Cilium and non-Cilium clusters.
  • Cilium data plane does not currently support DNS metrics.

Scale

Azure managed Prometheus and Grafana impose service-specific scale limitations. For more information, see Scrape Prometheus metrics at scale in Azure Monitor

Next steps