Network Monitoring

Network Monitoring provides visibility into communication between your services, hosts, and containers. Use it to identify which resources communicate, how much traffic they exchange, and which connections are slow or failing.

Cloud Network Monitoring collects TCP/UDP traffic and DNS activity through the Middleware Agent using eBPF on the Linux kernel. Network collection does not require application code changes.

With Cloud Network Monitoring, you can:

  • Monitor traffic volume, throughput, connection failures, retransmissions, and latency.
  • Explore service dependencies using Table, Sankey, and Map views.
  • Investigate slow or failed DNS lookups.
  • Review related logs, traces, processes, and alerts from connection details.
  • Investigate network findings with OpsAI.

Configure Network Monitoring#

Enable Network Monitoring on the Middleware Agent for the hosts or clusters you want to monitor.

Prerequisites#

  • Linux hosts, Docker on Linux, or Kubernetes with a Linux kernel that supports eBPF.
  • Middleware Agent version 2.0.1 with Network Monitoring enabled.
  • Your Middleware API key and target URL.
  • Root access for host installation, or permission to install the Helm chart for Kubernetes.
  • A cluster name when configuring Kubernetes.

Enable Data Collection#

Install or update the Middleware Agent for your environment:

Use the installation script for your DEB- or RPM-based distribution.

bash
MW_API_KEY="<MW_API_KEY>" MW_TARGET="https://<MW_UID>.middleware.io:443" MW_AGENT_FEATURES_CLOUD_NETWORK_MONITORING=true bash -c "$(curl -L https://install.middleware.io/scripts/deb-install.sh)"

On RPM-based distributions, use rpm-install.sh instead of deb-install.sh.

Use the Docker Agent installation script.

bash
MW_API_KEY="<MW_API_KEY>" MW_TARGET="https://<MW_UID>.middleware.io:443" MW_AGENT_FEATURES_CLOUD_NETWORK_MONITORING=true bash -c "$(curl -L https://install.middleware.io/scripts/docker-install.sh)"

Deploy the Middleware Agent using the Helm chart.

bash
helm repo add jetstack https://charts.jetstack.io --force-update
helm install cert-manager jetstack/cert-manager --namespace cert-manager --create-namespace --version v1.14.5 --set installCRDs=true

helm repo add middleware-labs https://helm.middleware.io
helm install mw-agent middleware-labs/mw-kube-agent-v3 \
  --set global.mw.apiKey=<MW_API_KEY> \
  --set global.mw.target=https://<MW_UID>.middleware.io:443 \
  --set opsai.enabled=true \
  --set mw-autoinstrumentation.enabled=true \
  --set global.clusterMetadata.name=my-k8s-cluster \
  --set cloudNetworkMonitoring.enabled=true \
  -n mw-agent-ns --create-namespace

Replace my-k8s-cluster with your cluster name.

Then:

  1. Restart the agent after updating its configuration.
  2. Open Network Monitoring → Cloud Network to verify that traffic appears.

Data typically starts appearing within 1–2 minutes after collection begins.

Cloud Network Monitoring#

Go to Network Monitoring → Cloud Network.

The page provides three main areas:

AreaWhat you can view
AnalyticsTCP/UDP connections and DNS lookups in Table, Sankey, or Map views
HealthHost coverage, network findings, and failing DNS lookups
DashboardsNetwork traffic, connectivity, performance, and application metrics

Select the time range you want to investigate. The available data depends on the hosts collecting network telemetry and the filters applied to the view.

TermMeaning
Client / ServerThe side that opened the connection / the side that accepted it.
ConnectionTraffic between a client and a server in the selected time range. One table row.
Volume / ThroughputTotal data moved / data moved per second.
TCP failuresConnection attempts that failed: timeout, refused, reset, unreachable or canceled.
RetransmitsPackets sent again because they were lost. A rise usually means network trouble.
RTT / LatencyTime to reach the server and come back / half of that.
DNS errorsNXDOMAIN: the name doesn't exist. SERVFAIL: the DNS server couldn't answer. Timeout: no answer.
N/A (unresolved)A side Middleware couldn't name, such as an external IP. Hidden unless Show Unresolved is on.

Analytics#

Use Analytics to explore traffic between clients and servers.

A client initiates communication, and a server receives it. The selected grouping determines how each side is represented, such as by service, host, container, availability zone, or IP address.

At the top of the page:

  • Select TCP/UDP or DNS traffic.
  • Choose Table, Sankey, or Map.
  • Set the grouping for the client and server.
  • Apply filters to focus on the resources you want to investigate.

The Summary Graphs show trends for the busiest connections. A problem banner identifies how many connections in the current view have detected issues.

Cloud Network Analytics in Table view with summary graphs and a problem banner

Search and Filter Traffic#

Use the search bar or Quick Filter panel on the left to narrow traffic by available resource attributes.

For example, filter by a service and group the destination by host to inspect which hosts receive its traffic.

Additional controls include:

ControlPurpose
Group ByChoose how client and server data is aggregated
Show UnresolvedInclude traffic where one side could not be identified and appears as N/A
Query LimitControl how many connections are loaded
Column SelectorChoose the fields displayed in the table

The page URL preserves the selected view, grouping, filters, and time range. Copy the link to share the same investigation view with a teammate.

Table View#

The Table view displays traffic between each client and server grouping. A row can aggregate multiple underlying network flows.

Available fields include:

FieldDescription
Client / ServerResources communicating with each other, based on the selected grouping
VolumeTotal data transferred during the selected time range
ThroughputData transferred per second
PacketsPacket counts for the selected traffic
TCP FailuresRecorded TCP failures, such as timeouts, refused connections, and resets
RetransmitsTCP packets sent again; an increase can indicate a connection problem
ResetsTCP connection resets
RTTRound-trip time between the endpoints
LatencyAn estimate calculated as RTT divided by two

TCP health metrics apply to TCP traffic. The latency estimate describes network timing, rather than the complete execution time of an application request.

Select a column heading to sort the table. For example:

  • Sort by Volume to identify the connections transferring the most data.
  • Sort by Failures to find connections with failed communication.
  • Review Retransmits and RTT when investigating slow connections.

Use the column selector to display additional fields, such as RTT or closed connections.

Health Indicators#

The table highlights values that need attention:

  • Red row: TCP failure rate is 1% or higher.
  • Highlighted values: Failures, retransmits, resets, or elevated latency, including latency of 20 ms or higher.

These indicators help prioritize investigation. Configure alert rules separately when you need notifications.

Table rows highlighted for TCP failures, retransmits, resets and high latency

Connection Details#

Select a row in the Table view to open its Connection Details panel. You can also open connection details by selecting a band in the Sankey view.

The panel identifies the client and server and shows the amount of data transferred in each direction.

TabContents
OverviewDetected issues, connection-specific charts, and shared tags
FlowsIndividual flows with client and server IP addresses, ports, hosts, and tags
AlertsAlert rules associated with the connection
LogsRelated logs filtered to the client and server
TracesRelated application traces filtered to the client and server
ProcessesProcesses associated with the client and server

For DNS traffic, the Flows tab is labeled Requests.

Review Individual Flows#

Open Flows to inspect the individual network flows behind an aggregated connection.

Use the IP addresses, ports, hosts, and tags to identify which machines are involved. For example, a connection grouped by service may contain flows from several hosts or containers.

Use Logs, Traces, and Processes to investigate the resources involved in the connection.

These views use the client and server context to narrow the data. Related telemetry is available when the corresponding log collection, application tracing, or process monitoring is configured.

Connection Details panel with Overview, Flows, Alerts, Logs, Traces and Processes tabs

Inspect IP Addresses#

When grouping by IP address, such as src.address or dst.address, hover over an IP to inspect the available information.

IP TypeAvailable Context
Private IPMonitored host and containers associated with the address, with links to Infrastructure Monitoring
Public IPNetwork ownership or autonomous system (AS), domain, country, and continent

Select Apply as Filter to narrow the page to that IP address.

Use this information to identify the resource behind a private address or understand which external network a service communicates with.

Private IPPublic IP
Hover card for a private IP showing the host and containers behind it
Hover card for a public IP showing network, domain, country and continent
The host and containers behind the IP, with links to them.Who owns the IP: network (AS), domain, country and continent.

Sankey View#

The Sankey view displays communication as flows between clients on the left and servers on the right.

Each band connects a client to a server. Its width represents the selected Measure.

Available measures include:

  • Bytes
  • Packets
  • Retransmits
  • Latency
  • Failures

When Bytes is selected, wider bands represent connections transferring more data. When Failures is selected, wider bands represent connections with more failures.

Hover over a band or resource to inspect its value. Use Limit to control how many connections are drawn, and select a band to open its connection details.

Use this view to compare traffic distribution or identify which connections contribute most to a selected network metric.

Sankey view of client to server traffic

Map View#

The Map view displays resources and the connections between them.

  • Nodes represent services, hosts, containers, or domains.
  • Lines represent communication between resources.
  • Arrows indicate the direction from client to server.
  • Node size and line thickness reflect the selected metric.

Use Highlight Top Traffic to emphasize the busiest connections. Hover over a node to highlight its connections, or select it to focus on that resource and the resources it communicates with.

The View control changes what each node represents. Enable Show Unresolved to include traffic whose resource identity could not be determined.

Use the map to inspect service dependencies, communication with external domains, and resources connected to an affected service.

Group Resources into Clusters#

Use Cluster to group map nodes by attributes such as host, availability zone, or region.

Select a cluster bubble to expand it and inspect the resources inside.

The number on a bubble is the larger of its client count and server count. Hover over the bubble to see both counts. The number of distinct resources inside can be higher because the two groups may contain different resources.

Map view of services, hosts and the connections between them

DNS Monitoring#

In Analytics, switch the traffic type to DNS to inspect name resolution activity.

The DNS table provides information such as:

  • Client making the lookup.
  • Domain being queried.
  • DNS server associated with the request.
  • Number of lookups.
  • Response timing.
  • Percentage of failed or timed-out lookups.

Common error types include:

ErrorMeaning
NXDOMAINThe queried domain name does not exist
SERVFAILThe DNS server could not complete the query
TimeoutNo response was received within the expected time

Sort by % SERVFAIL or % Timeouts to identify affected clients, domains, and DNS servers.

Select a row to open its details, then use Requests to inspect individual lookups and review available logs or traces.

Network Health#

Open Cloud Network → Health to review detected network issues and monitoring coverage.

Host Coverage#

The coverage bar shows how many hosts are sending network data.

Select Enable Hosts to configure collection on additional hosts. Review coverage when checking whether a service or host is represented in the network views.

OpsAI Findings#

OpsAI Findings lists connections with detected problems.

Findings include:

  • Critical: TCP failures have crossed the applicable failure threshold.
  • Warning: Connections show retransmissions, resets, or elevated round-trip times.
  • Impact: Metrics describing the detected issue.

Select a finding to inspect the affected connection, or use Investigate to start an OpsAI analysis.

If no findings are listed, no issues have been detected within the current monitoring scope. Review host coverage and the selected time range when assessing the result.

DNS Failures#

The DNS Failures section highlights domains with failing lookups.

Each entry provides the failure reason, such as SERVFAIL or timeout, together with the client and DNS server involved.

Select Details to inspect the requests, or View DNS Analytics to explore the wider lookup activity.

Cloud Network Health tab with host coverage, OpsAI findings and DNS failures

Investigate Network Issues with OpsAI#

Use Investigate from a connection or Health finding to analyze a network issue.

  1. Open the affected connection or finding.
  2. Select Investigate.
  3. OpsAI opens a chat panel with the connection and time range included as context.
  4. Review the investigation as OpsAI retrieves and analyzes available metrics, logs, and traces.
  5. Read the root cause analysis and ask follow-up questions to clarify the findings.

For example:

  • What could be causing failures between these services?
  • Do the related logs show errors during the same period?
  • Does increased network latency coincide with slower application requests?
  • Which hosts or processes should I inspect next?

Investigations are saved under OpsAI → Chats. Use the expand control to open the conversation in full view or return to it later.

OpsAI chat panel investigating a network connection

Network Dashboards#

Cloud Network Monitoring includes four predefined dashboards.

Two are available under Health → Dashboards. The other two open from the Dashboards menu at the top of Cloud Network.

DashboardLocationWhat It Shows
Intro to Cloud NetworkHealth → DashboardsTraffic volume, connections, failures, services, external domains, and host coverage
Connectivity & PerformanceHealth → DashboardsService-to-service flows, traffic sent and received, RTT, connection failures, retransmits, dependencies, and DNS
CNM Metrics OverviewDashboards menuNetwork load, TCP and DNS health, traffic by availability zone, and APM errors and latency
CNM Use Case OverviewDashboards menuCross-zone traffic, external destinations, service dependencies, highest-volume connections, and failing DNS lookups

Use CNM Metrics Overview to compare network behavior with application errors and latency. Use CNM Use Case Overview to inspect traffic patterns, including traffic crossing availability zones that may contribute to data-transfer costs.

The two CNM dashboards are also available in the main Dashboards list by searching for CNM.

To customize a dashboard, open it and select Save as a Copy.

Investigation Workflows#

Investigate a Slow or Failing Connection#

  1. Open Cloud Network → Health and select a finding, or use Analytics → Table to locate the affected client and server.
  2. Review failures, retransmits, RTT, and traffic volume for the relevant time range.
  3. Open Connection Details → Overview to inspect the connection's metric trends.
  4. Check Flows to identify the hosts, IP addresses, and ports involved.
  5. Review related Logs, Traces, and Processes for activity during the same period.
  6. Select Investigate to use OpsAI with the connection context.

Investigate DNS Resolution Failures#

  1. Open Analytics and select DNS.
  2. Filter to the affected service, host, or domain.
  3. Sort by % SERVFAIL or % Timeouts.
  4. Open an affected entry and review its Requests.
  5. Compare the failure reason, client, DNS server, and related telemetry to narrow the investigation.

Troubleshooting#

The Setup Page Still Appears After Installation#

Allow approximately 1–2 minutes for data to arrive after network collection starts.

If the setup page remains:

  • Confirm that the agent version supports Network Monitoring.
  • Verify that the host uses Linux with an eBPF-compatible kernel.
  • Confirm that Network Monitoring is enabled and the agent was restarted after configuration changes.
  • Check the API key, target URL, and agent connectivity.

Some Connections Show N/A#

N/A means Middleware could not resolve one side of the connection to a named resource.

Enable Show Unresolved to include this traffic. Try grouping by IP address or domain, or enable collection on additional hosts to provide more resource context.

The Map Cluster Count Differs from the Resources Inside#

The cluster badge displays the larger of the client and server counts. Hover over it to review both values; the distinct resources inside may include members of either group.

NXDOMAIN Is High in Kubernetes#

Kubernetes DNS search domains can cause a lookup to try several names before resolving the intended one. Some of those attempts can return NXDOMAIN.

Inspect the queried names and whether the intended lookup eventually succeeds before treating the count as an application failure. See Kubernetes DNS for Services and Pods for search-domain behavior.

Are Application Code Changes Required?#

No. Network traffic is collected through eBPF on the Linux kernel. Application traces and logs require their corresponding instrumentation or collection to appear in related-data views.

What did you think of this content?