Troubleshooting Guidelines for Customer Edge Site


The purpose of this guide is to provide troubleshooting guidelines for App Stack Site and Secure Mesh Site customer edge (CE) sites in various areas. This includes starting with the provisioning phase, data plane, services, and layer 7 (L7) user traffic.

F5 Distributed Cloud Services Components and Architecture

This section introduces the main components of a CE and the interactions between them. It will help you understand how particular components work along with their dependencies.


Volterra Platform Manager (VPM) is an agent on the node providing configuration of services (for example, Kubernetes and Distributed Cloud Services) and the operating system (OS). VPM is a Docker image that, at first start of the node, configures and activates all services for the node role. After deployment, the VPM is responsible for life-cycle management of the OS, F5 Distributed Cloud Services which include ves-system, kube-system, Webhook, and monitoring namespace workloads.

Note the following information for the namespaces:

  • kube-system namespace: Kubernetes components responsible for managing the k8s cluster.
  • ves-system namespace: Distributed Cloud Services components responsible for CE work.
  • Webhook namespace: Webhook pod control for k8s object creation.


The following prerequisites apply:

Troubleshooting Tools

There are several tools you can use to troubleshoot your CE site.

UI Tools

In the Console UI, the main place for CE troubleshooting is the Dashboard, which is under the Tools tab.

Figure: UI Tools


SiteCLI is a command-line interface (CLI) tool, where you log in using SSH or a terminal with username admin with the default password of Volterra123 (which you must change after you log in for the first time).

Note that for public cloud CE sites, you must use the cloud-user account username with your private key after registration is completed.

Figure: SiteCLI Tools

Site Web UI

You can directly access each node via a web browser by navigating to https://[node IP]:65500. For example, For more information, see the Access Site Local User Interface guide.

Figure: Site Web UI Tools

Collect Diagnostic Information

The Distributed Cloud Support Team can request diagnostic information for feature debugging.

  • Directly access using a web browser at https://[node IP]:65500.

  • Navigate to the Debug info page and click Collect Debug Info.

  • After some time, download the debugging information file locally to analyze or share with the Distributed Cloud Support Team. To download, click Browse debugging info files.

Figure: Diagnostic Information Tools

Troubleshoot Provisioning States

This section provides the right starting point during the provisioning of CE sites based on node state:


  • Symptom: No pending registration seen on Console for a new CE site.
  • Action: Check CE connectivity and VPM logs.


  • Symptom: Pending registration available on Console, but either accepting it fails or CE site never reaches online state and likely can no longer be reached via network.
  • Action: Check CE connectivity, VPM logs, and cluster health if formed.


  • Symptom: CE site is online, but some L3 to L7 services do not work as expected.
  • Action: Check cluster health and control plane described in next sections.

Registration Process Workflow

The diagram below shows how VPM interacts with Global Controllers during installation.

Note: Maurice and Charmander are Global controller components responsible for CE configuration management.

Figure: Registration Process Workflow

Troubleshoot VPM

Volterra Platform Manager (VPM) is an agent on the node providing configuration of services (like Kubernetes and services) and the operating system. As the main tool responsible for node provisioning, log output analysis is critical to identify the root cause of an issue.

VPM Status Check

  • Directly access using a web browser at https://[node IP]:65500.
Figure: VPM Status Check
  • For SiteCLI, enter the following: execcli systemctl-status-vpm.
Figure: VPM Status Check
  • For SiteCLI detailed view, enter the following: status vpm.
Figure: VPM Status Check

The status vpm command provides:

  • Software upgrade and OS upgrade status and current version.
  • Registry connectivity status.
  • K8 cluster status.
  • Service responsible for certificates management status.
  • Node certificate validation.
  • Installation logs and status.
  • All node ves-system pod statuses (all remaining Distributed Cloud Services software managed by Kubernetes).

VPM Logs Check

As the VPM is responsible for provisioning of the node/cluster, the first place to start debugging are the VPM logs. Before getting logs, confirm that the VPM works by using the following command: execcli systemctl-status-vpm.

You can access the logs using any of the three methods.

  • Using the SiteCLI method: execcli journalctl -fu vpm.
Figure: VPM Logs Check
  • Using the Console method: navigate to your site's dashboard and then click the Tools tab. From the Select tool menu, select Log.
Figure: VPM Logs Check
  • Directly on the node. Navigate to https://[node IP]:65500. Select Tools. From the Select tool menu, select Show Log.
Figure: VPM Logs Check

Common VPM Log Errors

Common error log messages for the VPM during site deployment. In this section, some of those common log messages and corresponding actions are presented.


A node is ready to be admitted pending registration.

Current state: PENDING, registration c6d2a9a6-b834-466c-87ff-fc9d1e02c0e0 must be manually APPROVED and then automatically ADMITTED.

Lack of it before registration can indicate some issues. For more information, see the next log message description for Can’t reach the state to accept registration.

Action: Accept registration, if it is a multi-node cluster. Be sure that in the registration form, number of nodes equals three (3). During the acceptance of registration, you can provide a specific version of F5 Distributed Cloud Services software (CRT) and operating system (OS) version.

Figure: Pending State
Can’t reach the state to accept registration

There is no …PENDING… log during provisioning and node not seen as pending registration in Console: Multi-Cloud Network Connect > Manage > Site Management > Registrations.

Action: Check communication of the node via execcli command. Check IP address and ping using execcli ip address and execcli ping

SiteCLI diagnostic command provides more complex connectivity tests but cannot be provided as it requires k8s already deployed to be available after deployment diagnosis.

Figure: Cannot Reach State

If communication works, check that the provided cluster and node names match the configuration provided in the UI.

Figure: Matching Names
Invalid token

Invalid token provided; the node never reaches the state to be ready for registration.

Registration failed: Registration request: Request Register failed: Response with non-OK status code: 500, content: "Unknown token: Token doesn't exist.

Alternatively, the wrong API endpoint is provided. If you use F5 Distributed Cloud Services dedicated Global Controllers infrastructure, the correct API endpoint should be provided.

Registration failed: Registration request: Request Register failed: Request error: Post "https://[URL] /registerBootstrap": Internal Server Error.

Action: Check the token value using the SiteCLI get-config command to print VPM configuration.

Figure: get-config Command

In the Vpm: section: look for the Token value and compare with the token generated for this tenant.

Figure: Check Token Value

If the token is invalid, provide a valid one. Be careful with copy/paste actions as not to provide any whitespace characters. To generate a new token, follow the instructions at Create Site Token.

If the token is valid, it can be at the wrong registration API endpoint. Check if the registration URL is correct. The default is Look for the VPM section for MauriceEndpoint.

Figure: Check API Endpoint

If during provisioning registration the API URL was wrongly provided, the only way to fix it is start to from initial configuration via SiteCLI again (after a factory reset operation is performed).

Missing token

Upgrade of the node or redeployment of one of the cluster nodes fails due lack of token.

In the VPM logs following error can be seen: register.go:805: Config request #6 failed with: Request Config failed: Response with non-OK status code: 500, content: 0Can't find tenant for token: Token doesn't exist;2024-03-14 19:07:54.173328263 +0000 UTC m=+129052.531390838,, waiting 16.233694717s before next try.

Action: Re-deploy cluster or contact the Distributed Cloud Support Team.

Important: If token is deleted from the Distributed Cloud Console after installation, the upgrade of the node and update certificates for the node will fail. Be sure that the token exists before upgrading. Good practice is not deleting any tokens used for deployment at all.

Wrong certified hardware profile

In the process of registration acceptance, the certified hardware mismatch error occurs.

Figure: Check Certified Hardware

Action: Check if the provided certified hardware matches the criteria for the given deployment. In the UI, which represents the Global Controller view, the configuration must match to the profile provided during node configuration with SiteCLI. Check the certified hardware profile using the SiteCLI get-config command to print VPM configuration.

Figure: Check Certified Hardware Menu

To check a list of available hardware profiles for a tenant, navigate to Multi-Cloud Network Connect > Manage > Site Management > Certified Hardware. Each detailed specification can be checked.

If the profile is wrong, as in wrong name or interface list, the node must be reset to factory settings using SiteCLI and provisioned again. To execute factory reset, run command: factory-reset.

Figure: Factory Reset Command

Note: The factory reset operation only works on bare metal/KVM cases. If you are faced with resetting configuration, redeployment is an alternate method to move forward.

Certified hardware profiles editing

In some cases, you can find dedicated per-customer-certified hardware profiles that are not presented on the ISO. These profiles require manual edit in the ISO before registration.

To edit the certified hardware profile, run command: execcli edit-certified-hardware.

          active: [name of the profile] # this profile will be used for provisioning
outsideNic: - name of NIC that be used as outside
primaryOutsideNic: in majority of cases the same interface as outsideNIC
PrivateNIC: eth0. – NIC name used by VPM to communicate outside of CE, e.g. download images etc.

Example of full certified hardware profile:

          active: hp-proliant-dl360-voltstack
primaryOutsideNic: ens2f0
      PrivateNIC: ens2f0
    - ens2f0

To locate the full list of Certified Hardware available for a tenant, navigate to Multi-Cloud Network Connect > Site Management > Certified Hardware.

Figure: Full List of Certified Hardware

In the list, the most important details are the profile name, interface allocation, and naming. Profile naming conventions include the following:

  • voltstack: Dedicated for App Stack Site
  • voltmesh: Dedicated for Secure Mesh Site
  • kvm: Kernel-based Virtual Machine
  • gcp: Google Cloud Provider
  • aws: Amazon Web Services
Duplicated key

Re-using the same name for new sites can cause issues, and during node deployment, no pending registration notice appears in UI. In VPM logs, the following error occurs:

          register.go:707: Registration failed: Registration request: Request Register failed: Response with non-OK status code: 409, content: 'Create: Adding with usage transaction: STM ErrorValue playground-wtppvaog/pg-test1
/pg-test1-m0, Key: 1198fad4-0c8c-4c88-b334-4b03b82dc23f, Duplicate Keys: 854cffb5-f316-4ab5-89e3-3b340f931e429

Action: Check if the node in the cluster is already created or if some failed deployment with the same name exist. If yes, delete them and start a new node deployment. Alternatively, you can change the name of the cluster for a new deployment to avoid conflicts. If there is no configured cluster with the same name, create a support ticket to fix the issue.

Check Cluster Health

This section provides guidance on how to check cluster health on a different level. Clusters have a minimum of three master nodes, to make sure that all components are running. Each component requires a number of instances/PODs. Typically, it is three for master nodes (etcd) and equal to number of nodes for others.

For this guidance, a three-node CE site (pg-sms-1) will be used with the following nodes: pg-sms-1-m0, pg-sms-1-m1, and pg-sms-1-m2. All SiteCLI commands are executed on node pg-sms-1-m0.

Check Health with Console UI

For each site, navigate to its general view dashboard. The important metrics for health of a cluster are sections: System Metrics (expected 100%) and Connectivity (expected two REs connected).

Figure: Site Dashboard

The system health percentage score is calculated using System Health = (tunnel_status * reachability * ares_status * % of nodes that are UP) * multiplier:

  • tunnel_status: This is the average of a site's tunnel status to both REs.
  • ares_status: This is the average of a site's ares' connectivity status.
  • The multiplier depends on RE connectivity status. If both are up, multiplier is 1. If only 1 is up and 1 is down, then multiplier is 0.75. If both are down, multiplier is 0.

In the Nodes tab, it displays a per node overall condition that points out which of the nodes is connected to an RE as primary (Controller-DP). Note that workers can also be a primary connection to an RE having direct IPsec/SSL tunnels to REs. The whole site will also have two tunnels toward the REs. It can be master or worker types of nodes selected.

Figure: Site Dashboard Nodes Tab

Check Health with SiteCLI

There several methods available to check the cluster's health using SiteCLI. One method is to use the health command.

The health command provides cluster-related information (and many others):

  • Provisioning state, public IP, software version, and OS version
Figure: Cluster Health
  • Kubernetes (k8s) cluster state
Figure: Cluster Health
  • General workload (ves-system) PODs health
Figure: Cluster Health

To check connectivity to external systems or interface IP addresses, the diagnosis command is a general tool for networking communication validation. This command only works on a deployed CE site (healthy or almost healthy node). The command will not work before a node is properly deployed or for a seriously broken cluster state.

The diagnosis command checks the following information per node:

  • External URL service connectivity (repositories, APIs, and more).
  • RE registration for each node
  • ARP table status
  • DNS query to certain domains
  • ifconfig command output
Figure: Cluster Health with Diagnosis Command

Use the volterra-status command to check all PODs/components and their health state. This command prints the PODs status on each node across a cluster. If something is wrong with the POD, it will display a short message. For more information, the POD logs need to be collected. The column name indicates the POD name and can be used to get logs from the POD directly.

Figure: Cluster Health with volterra-status Command

Use the execcli crictl-ps command to check the POD logs. From the list displayed, find the container ID and then use it to check specific logs. For example, execcli crictl-logs 09c12bb1a5d9a.

Figure: Cluster Health with execcli crictl-ps Command

For App Stack sites, use the kubectl command to check directly cluster health. For this command, you can mainly check the ves-system namespace. To check general node state, use the simple kubectl get nodes command.

Figure: Cluster Health with kubectl Command

Major Distributed Cloud Services components are executed as Daemon Set PODs, so the quickest way to see cluster component states across nodes is to use command: kubectl get ds -n ves-system.

Figure: Cluster Health with kubectl Command

Troubleshoot Control Plane

Control plane issues are mainly caused by a lack of communication with the Global Controllers/REs or inconsistency of their states. Some symptoms of control plane issues include not configuring components (zero configuration) or lack of configuration updates (as in no new configuration applied).

The following sections describe tools to diagnose control plane issues.

Check Vega Objects

Vega (VER) is a control plane component of a CE. Using SiteCLI, you can list all Vega objects and detail values for them. It is raw data of the configuration and in most cases, valuable only for the F5 Distributed Cloud Support Team. Regular users can use this tool to verify if the configuration applied on the Global Controller programs the local CE correctly, and if the CE (etcd communication) works to store the configuration locally.

To list all Vega objects, use the following command: execcli vegactl-configuration-list.

To list all sites in a tenant, use the following command: execcli vegactl-configuration-list

Figure: Check All Sites in Tenant

Use the site unique identifier (UID) to get the site configuration and other object references: execcli vegactl-introspect-get ff1f98f3-045e-4071-b46c-466675ba3b8c. See sampled output (only the configuration part):

Figure: Check Site UID

Check Envoy Raw Configuration

You can print detailed and low-level configuration information for Envoy that is only usable for the F5 Distributed Cloud Support Team. Regular users can check if the configuration provided on the UI is applied via the Global Controller on the CE site.

To see very detailed configuration information for Envoy, including all internal components where among them there is user load balancer configuration, use command execcli envoy-config-dump. This command provides very large output, with all internal and user data.

To print information for load balancer listeners on a CE site, use command execcli envoy-listeners.

Figure: Check Load Balancer Listeners

To print information for load balancer health check for the CE site, use command execcli envoy-hc-config-dump.

Figure: Raw Envoy Configuration

Check CE to RE Tunnel Status

Communication between a CE site and an RE site is critical due to it being a method to download configuration of the CE site from Global Controllers. The CE site (with one or three-node clusters) establishes two tunnels, where one tunnel is primary and marked as Active. RE tunnels are also used to expose services to the public Internet using load balancer virtual IP addresses (VIPs), to the public Internet on an RE. In the Console UI, tunnel status is well presented, including the load of tunnels.

  • To check tunnel status in Console, navigate to your site dashboard and select the Site Status tab.
Figure: Check Site Tunnel Health
  • To check CE connectivity status in Console, navigate to the site dashboard and select the RE (PoP) Connectivity tab. This is a view of CE connectivity to the RE for whole-tenant CEs. Using the filters given, the site can be displayed, including connectivity health, throughput, latency, control plane, and data plane status.

Note: The Console UI displays have some latency to present status and cannot be considered as a live view. To get a more accurate live status, use the SiteCLI terminal.

Figure: Check CE Connectivity Status
  • To use SiteCLI to view the tunnel status, use command status ver.

The example below shows that the RE tunnel is SSL to ny8-nyc RE (state: TUNNEL_UP and URL:). To see both RE tunnels, check verStatus > aresStatus > state: ESTABLISHED. It shows two ESTABLISHED states for ny8-nyc and pa4-par.

Figure: Check Tunnel Status with SiteCLI

Troubleshoot Data Plane

This section describes how to troubleshoot a data plane for a CE site. The main component of the data plane on the CE site is Argo, and there is a place to focus on: capture traffic, examine routing, and examine flow tables.

The following tools can be useful to troubleshoot application connection issues, including:

  • Lack of connectivity;
  • Packets dropping during transmission (by link saturation);
  • Too big of a maximum transmission unit (MTU), and any other related issues.

Capture Packets from a POD (App Stack Site)

Make sure you identify the POD name and node name.

  • In the Console UI, navigate to the Distributed Apps service.

  • Click Managed K8s > Overview.

  • Select the managed K8s cluster, and then click the Pods tab.

Figure: Get POD Name
  • Use the kubectl get pods -n pg command:
          NAME                             READY   STATUS    RESTARTS   AGE    IP             NODE                    NOMINATED NODE   READINESS
nginx-project-6b9d44488b-2rgj4   2/2     Running   2          8d   pg-appstack-single-m0   <none>           <none>
  • To capture network packets in Console:

    • Navigate to the tools on your site dashboard by clicking Multi-Cloud Network Connect > Overview > Sites.

    • Select your site, and the click the Tools tab.

    • Select the Tcpdump tool.

Figure: Tcpdump Tool
  • Select the node name.

  • Select the pod for the Target.

  • Select the pod.

Figure: Tcpdump Tool
  • After traffic is captured, download the file locally and open it using Wireshark.
Figure: Download File
  • Use the SiteCLI execcli command for more options to gather information:

    • Identify POD interface name by checking the CNI interface name in Vega configuration. As an example, use command execcli vegactl-introspect-dump-table In the output, search for the POD/container name. The next line will be the devName, indicating Argo virtual interface to the POD.
Figure: Identify POD Interface
  • Use the interface name list virtual interfaces of Argo to find the interface ID. As an example, use command execcli vif -l. Note that you cannot use the grep command to filter output as it is not supported. Instead, use terminal search to find the interface name.
Figure: Identify POD Virtual Interface List
  • Use the interface ID to capture packets by the vifdump command. As an example, use command execcli vifdump 28 -ne icmp.
Figure: Interface ID to Capture Packets