Frequently Asked Questions for AKO

Overview

This document answers some of the frequently asked questions for AKO:

  1. How do I clean up all my configs?

  2. How is the shared virtual service lifecycle controlled?

  3. How are virtual services sharded?

  4. How do I decide the size of the shard virtual service?

  5. Can I change the shard VS number?

  6. How do I alter the shard VS number?

  7. What happens if the number of DNS records exceed a Shard VS?

  8. What is the use of static routes?

  9. What happens if I have the same SNI host across multiple namespaces?

  10. What out-of-band operations can I do on the objects created by AKO?

  11. What is the expected behaviour for the same host/path combination across different secure/insecure ingresses?

  12. What happens to static routes if the Kubernetes nodes are rebooted/shutdown?

  13. Can I point my ingress objects to a service of type Loadbalancer?

  14. What happens when AKO fails to connect to the Avi Controller while booting up?

  15. What happens if we create ingress objects in an OpenShift environment?

  16. What are the virtual services for passthrough routes or ingresses?

  17. What happens if insecureEdgeTerminationPolicy is set to redirect for a passthrough route?

  18. How to debug Invalid input detected errors?

  19. Some of the pool servers in the NodePort mode of AKO are down. How is this fixed?

  20. Can we create a secure route with edge/reencrypt termination without key or certificate?

  21. What happens if a route is created with multiple backends having the same service name?

  22. Is the NodePortLocal feature available for all CNIs?

  23. Can we use Kubernetes Service of type NodePort as backend of an Ingress in NodePortLocal mode?

  24. Can we use serviceType NodePort or ClusterIP in AKO, if the CNI type is Antrea and NodePortLocal feature is enabled in Antrea?

  25. What are the steps to change ServiceType?

  26. Can the service Type in AKO be changed dynamically?

  27. If service Type is changed from NodePortLocal, will AKO remove NPL annotation from the Services automatically?

  28. How can I see the marker labels associated with an object on the Avi Controller?

  29. What is the difference between AKOSetttings.namespaceSelector and AKOSettings.blockedNamespaceList? What are the scenarios to use one of them?

  30. What is the minimum stable Kubernetes version which supports SCTP protocol?

  31. What is the minimum Avi Controller version which supports the SCTP protocol?

  32. What is the minimum AKO version which supports the SCTP protocol?

  33. How to manually override the Active AKO when HA is enabled?

  34. Can the user scale beyond two instances of AKO in HA?

  35. What happens during a Split brain scenario in HA?

How do I clean up all my configs?

The key deleteConfig in the data section of AKO configmap can be used to clean up the setup. Edit AKO configmap and set deleteConfig: true to delete ako created objects in Avi. After the flag is set in configmap, annotation AviObjectDeletionStatus is added in the AKO statefulset with the value as Started.


  annotations:
    AviObjectDeletionStatus: Started
 

After all relevant objects gets deleted from Avi, the value of the annotation is changed to Done.


  annotations:
    AviObjectDeletionStatus: Done

To re-create the objects in Avi, the configmap has to be edited to set deleteConfig: false.

How is the shared virtual service lifecycle controlled?

AKO follows hostname based sharding to sync multiple ingresses with same hostname to a single virtual service. When an Ingress object is created with multiple hostnames, AKO generates an md5 hash using the hostname and the Shard VS number. This uniquely maps an FQDN to a given Shared virtual service and avoids DNS conflicts. During initial clean bootup, if the Shared virtual service does not exist in Avi, AKO creates the same and then patches the ingress FQDN to it either in the form of a pool (for insecure routes) or in the form of an SNI child virtual service (in case of secure routes).

The Shared virtual services are not deleted if all the FQDNs mapped to it are removed from Kubernetes. However, if the user wants AKO to delete unused shared VSes - a pod restart is required that would evaluate the VS and delete it appropriately.

Even though the unused shared VSes are deleted, the shared virtual service VIPs are retained to regain the same FQDN to virtual service VIP mapping. To delete the retained shared virtual service VIPs, the user has to manually delete them from the Controller UI or shell.

How are virtual services sharded?

If you create ingress with an insecure host/path combination then AKO creates a corresponding Avi pool object and patches the pool on one of the existing shard virtual services. The shard virtual service has a DataScript associated with it that reads the host/path of the incoming requests and appropriately selects a pool by matching it with the priority label specified for each pool member (corresponding to a host/path combination).

For secure ingresses, an SNI virtual service is created which although is a dedicated virtual service, does not have any IP addresses associated with it. The SNI virtual service is a child to a parent virtual service and is created based on the secret object specified in the ingress file against the host/path that is meant to be accessed securely.

How do I decide the size of the shard virtual service?

In the current AKO model, the Shard VS size is an enum. It allows 3 pre-fixed sets of values:

  • LARGE (8 virtual services)
  • MEDIUM (4 virtual services)
  • SMALL (1 virtual service)
  • DEDICATED (Shard Mode is disabled, dedicated virtual services would be created per hostname)

The decision of selecting one of these sizes for Shard virtual service depends on the size of the Kubernetes cluster’s ingress requirements. It is recommended to always go with the highest possible Shard virtual service number that is(LARGE) to account for expansion in the future.

Can I change the Shard VS number?

To Shard to virtual services, AKO uses a sharding mechanism that is driven by the hostname of each rule within an ingress object. This ensures that a unique hostname is always sharded consistently to the same virtual service.

Since the sharding logic is determined by the number of Shard virtual services, changing the Shard VS number has the potential hazard of disrupting an existing cluster’s already synced objects. Hence it’s recommended that the Shard VS numbers are not changed once fixed.

How do I alter the Shard VS number?

Altering the shard VS number is considered as disruptive. This is because dynamic re-adjustment of shard numbers may re-balance the ingress to virtual service mapping. Hence, if you want to alter the shard VS number, first delete the older configmap and trigger a complete cleanup of the virtual services in the Controller. Then, edit the configmap and restart of AKO.

What happens if the number of DNS records exceed a Shard VS?

Currently, the number of A records allowed per virtual service is 1000. If a shard size of SMALL is selected and the number of A records via the Ingress objects exceed 1000, then a greater shardSize has to be configured via the shardSize knob. Alternatively, one can create a separate IngressClass for a set of Ingress objects and specify a shardSize in the AviInfraSettings CRD which would allow AKO to place the A records scoped to the VS that is mapped to the IngressClass.

What is the use of static routes?

Static routes are created with cluster name as the label. While deploying AKO the admin or the operator decides a Service Engine Group for a given Kubernetes cluster. The same labels are tagged on the routes of this AKO cluster. These routes are pushed to the Service Engine’s created on the Service Engine Group. The static routes map each pod CIDR with the Kubernetes node’s IP address. However, for static routes to work, the Service Engines must be L2 adjacent to your Kubernetes nodes.

What happens if I have the same SNI host across multiple namespaces?

The ingress API does not prohibit the user from creating the same SNI hostname across multiple namespaces. AKO will create 1 SNI virtual service and gather all paths associated with it across namespaces to create corresponding switching rules. However, the user needs to denote each ingress with the TLS secret for a given hostname to qualify the host for the SNI virtual service.

Consider the below example:


Ingress 1 (default namespace) --> SNI hostname --> foo.com path: /foo, Secret: foo

Ingress 1 (foo namespace) --> SNI hostname --> foo.com path: /bar, Secret: foo

In this case, only one SNI virtual service will be created with sslkeyandcertificate as foo

However if the following happens, the behaviour of the SNI virtual service would be indeterministic since the secrets for the same SNI are different:


Ingress 1 (default namespace) --> SNI hostname --> foo.com path: /foo, Secret: foo
Ingress 1 (foo namespace) --> SNI hostname --> foo.com path: /bar, Secret: bar

This is not supported.

What out-of-band operations can I do on the objects created by AKO?

AKO runs a refresh cycle that currently just refreshes the cloud object parameters. However, if some out-of-band operations are performed on objects created by AKO via directly interacting with the Avi APIs, AKO may not always be able to mitigate an error caused due to this.

AKO has the best effort, retry layer implementation that would try to detect a problem (For example, an SNI VS deleted from the Avi UI), but it is not guaranteed to work for all such manual operations.

Upon reboot of AKO, a full reconciliation loop is run and most of the out-of-band changes are overwritten with AKO’s view of the intended model. This does not happen in every full sync cycle.

What is the expected behaviour for the same host/path combination across different secure/insecure ingresses?

The ingress API allows users to add duplicate hostpaths bound to separate backend services. For example,

Ingress1 (default namespace) --> foo.com path: /foo, Service: svc1

Ingress2 (default namespace) --> foo.com path: /foo, Service: svc2
Also, ingress allows you to have a mix of secure and insecure hostpath bound to the same backend services like so:

Ingress1 (default namespace) --> SNI hostname --> foo.com path: /foo, Secret: secret1

Ingress2 (default namespace) --> foo.com path: /foo, Service: svc2

AKO does not explicitly handle these conditions and would continue syncing these objects on the Avi controller, but this may lead to traffic issues. AKO does the best effort of detecting some of these conditions by printing them in logs. A sample log statement looks like this:

key: Ingress/default/ingress2, msg: Duplicate entries found for hostpath default/ingress2: foo.com/foo in ingresses: ["default/ingress1"]

What happens to static routes if the Kubernetes nodes are rebooted/shutdown?

AKO programs a static route for every node IP and the pod CIDR associated with it. Even though node state changes to NotReady in Kubernetes this configuration is stored in the node object and does not change when the node rebooted/shutdown.

Hence, AKO will not remove the static routes until the Kubernetes node is completely removed from the cluster.

Can I point my ingress objects to a service of type Loadbalancer?

No. The ingress objects should point to the service of type clusterIP.
Loadbalancer services either point to an ingress Controller pod if one is using an in cluster ingress controller or they can directly point to application PODs that need layer 4 load-balancing.

In a configuration where the ingress objects are pointing to services of the type load balancer, AKO’s behaviour would be indeterministic.

What happens when AKO fails to connect to the Avi Controller while booting up?

AKO would stop processing the Kubernetes objects and no update would be made to the Avi Controller. After the connection to the Avi Controller is restored, AKO pod has to be rebooted. This can be done by deleting the exiting POD and ako deployment would bring up a new pod, which would start processing Kubernetes objects after verifying connectivity to the Avi Controller.

What happens if we create ingress objects in an OpenShift environment?

AKO does not process ingress objects in an OpenShift environment. If any route corresponding to the ingress object is found, AKO would process that route.

What are the virtual services for passthrough routes or ingresses?

A set of shared virtual services are created for passthrough hosts to listen on port 443 to handle secure traffic using L4 DataScript. These virtual services have names of the format ‘cluster-name’-Shared-Passthrough-‘shard-number’. The number of shards can be configured using the flag passthroughShardSize while installation using Helm.

These virtual services would listen on port 443 and have one L4 SSL DataScript each. Name of the virtual service would be of the format clustername–‘Shared-Passthrough’-shardnumber. The number of shards can be configured using the flag passthroughShardSize while installation using Helm.

What happens if insecureEdgeTerminationPolicy is set to redirect for a passthrough route?

A set of shared virtual services are created for pass through hosts to listen on port 443 to handle secure traffic using L4 datascript. These virtual services have names of the format ‘cluster-name’-Shared-Passthrough-‘shard-number’. Number of shards can be configured using the flag passthroughShardSize while installation using Helm.

For passthrough routes, the supported values for insecureEdgeTerminationPolicy are None and Redirect. To handle insecure traffic for passthrough routes a set of shared virtual services are created with names of the format 'cluster-name'-Shared-Passthrough-'shard-number'-insecure. These virtual services listen on port 80. If for any passthrough route, the insecureEdgeTerminationPolicy is found to be Redirect, then an HTTP Policy is configured in the insecure passthrough shared virtual service to send appropriate response to an incoming insecure traffic.

How to debug Invalid input detected errors?

AKO reboots and retries some of the invalid input errors. Look out for the following Below cases to in the logs:

  • If an invalid cloud name is given in values.yaml or if ipam_provider_ref is not set in the vCenter and No Access clouds.
  • If the same Service Engine group is used for multiple clusters for vCenter and No Access clouds in Cluster IP mode. This happens as AKO expects unique SE group per cluster if routes are configured by AKO for POD reachability. Look for the labels does not match with cluster name message in the logs which points to two clusters using the same Service Engine Group.

Some of the pool servers in the NodePort mode of AKO are down. How is this fixed?

The default behaviour for AKO is to populate all the Node IP as pool server. If master node is not schedulable then, it will be marked down. nodePortSelector can be used to specify the labels for the node. In that case, all the node with that label will be picked for the pool server. If the master node is not schedulable then, remove the nodePortSelector label for the master node.

Can we create a secure route with edge/reencrypt termination without key or certificate?

For secure routes having termination type edge/reencrypt, key and certificate must be specified in the spec of the route. AKO would not handle routes of these types without key and certificate.

What happens if a route is created with multiple backends having the same service name?

AKO rejects those routes, as each backend should be unique ith it’s own weight. Multiple backends having same service would make the weight calculation indeterministic.

Is the NodePortLocal feature available for all CNIs?

No. The feature NodePortLocal can be used only with Antrea CNI and the feature must be enabled in Antrea feature gates.

Can we use kubernetes Service of type NodePort as backend of an Ingress in NodePortLocal mode?

No. Users can only use service of type ClusterIP as backend of Ingresses in this mode.

Can we use serviceType NodePort or ClusterIP in AKO, if the CNI type is Antrea and NodePortLocal feature is enabled in Antrea ?

Yes. AKO would create the Avi objects based on the relevant service Type set in AKO.

What are the steps for ServiceType change?

The serviceType in AKO can be changed from ClusterIP to NodePortLocal or NodePort. The serviceType change is considered disruptive. Hence before the serviceType change, all the existing AKO configuration must be deleted. To do this, follow the steps given below:

  1. Set the deleteConfig flag to True.
  2. Wait for AKO to delete all the relevant Avi configuration and update the deletion status in AKO’s statefulset status.
  3. Change the serviceType
  4. Set the deleteConfig flag to False
  5. Reboot AKO

For example, during the change of serviceType from ClusterIP to NodePortLocal, the deleteConfig flag will:

  • Delete the static routes.
  • Delete the SE group labels.

Can the service Type in AKO be changed dynamically?

No. After changing the service type, AKO has to be rebooted and all objects which are not required, would be deleted as part of the reboot process.

If service Type is changed from NodePortLocal, will AKO remove NPL annotation from the Services automatically?

No. After changing the service type, the users have to remove NPL annotation from the Services themselves.

How can I see the marker labels associated with an object on the Avi Controller?

Markers, associated with an Avi object, are visible on the Avi Vantage UI and the Avi shell.

To view the markers via the Avi UI,

  1. Edit an object like a virtual service or pool or pool group.

  2. Navigate to the Advanced tab.

The Role-Based Access Control (RBAC) shows the markers associated with the object.

Markers

To view markers for an object via the Avi Shell, use the following command.


[avi-shell] show <object type> <object name>

For example,


[avi-shell] show virtualservice kubernetes--Shared-L7-EVH-0
+------------------------------------+----------------------------------------------------------------------------------+
| Field                              | Value                                                                            |
+------------------------------------+----------------------------------------------------------------------------------+
| uuid                               | virtualservice-bc4e964a-af79-4b8f-91b2-de0e7ee9388d                              |
| name                               | kubernetes--Shared-L7-EVH-0                                                      |
| enabled                            | True                                                                             |
| vsvip_ref                          | kubernetes--Shared-L7-EVH-0                                                      |
| use_vip_as_snat                    | False                                                                            |
|        .                           | .                                                                                |
|        .                           | .                                                                                |
|        .                           | .                                                                                |
|        .                           | .                                                                                |
|        .                           | .                                                                                |
| markers[1]                         |                                                                                  |
|   key                              | clustername                                                                      |
|   values[1]                        | kubernetes                                                                       |
| allow_invalid_client_cert          | False                                                                            |
| vh_type                            | VS_TYPE_VH_ENHANCED                                                              |
+------------------------------------+----------------------------------------------------------------------------------+

What is the difference between AKOSetttings.namespaceSelector and AKOSettings.blockedNamespaceList? What are the scenarios to use one of them?

AKOSetttings.namespaceSelector allows AKO to process Ingress/Routes, L4 services from the namespaces that have given labels. AKO blocks processing Ingress/Routes and L4 services from the namespaces that does not have given labels, hence prevents Avi object creation. However, AKO also processes other objects (such as secrets, endpoints, and so on) from incorrect or unlabelled namespaces.

AKOSettings.blockedNamespaceList is a list of namespaces specified upfront from which AKO will not process any objects, such as ingress/route, L4 services, secrets, endpoints, and so on. So, AKOSettings.blockedNamespaceList is a broader way of blocking unwanted objects compared to AKOSetttings.namespaceSelector. AKOSettings.blockedNamespaceList currently does not support regex, so the user must specify each namespace.

Following are the few scenarios to consider AKOSettings.blockedNamespaceList:

  • If the list of namespaces to be blocked is known to the user upfront (e.g., system-namespaces), users typically do not deploy any application. Any new addition of namespace to this list will require an AKO reboot.

  • There are many objects (such as secret, pods, endpoint, and so on) in unwanted namespaces that are consuming AKO processing time.

Following are the few scenarios to consider AKOSetttings.namespaceSelector:

  • In a cluster, namespace churn is more frequent. Therefore, using a namespace selector is a better way of processing only wanted namespaces.

One of the ways to use AKOSetttings.namespaceSelector and AKOSettings.blockedNamespaceList effectively are:

  • Use AKOSettings.blockedNamespaceList for system-generated namespaces where users typically do not deploy any application.

  • Use AKOSetttings.namespaceSelector for user-defined namespaces.

What is the minimum stable Kubernetes version which supports SCTP protocol?

The minimum stable Kubernetes version which supports SCTP protocol is 1.20.

What is the minimum Avi Controller version which supports the SCTP protocol?

The minimum Avi Controller version which supports the SCTP protocol is 22.1.3.

What is the minimum AKO version which supports the SCTP protocol?

The minimum AKO version which supports the SCTP protocol is 1.9.1.

How to manually override the Active AKO when HA is enabled?

Trigger a deletion of the Active AKO pod, and the passive AKO pod automatically comes up as the new active AKO.

Can the user scale beyond two instances of AKO in HA?

Currently, AKO beyond two instances is not been tested. Hence, it is not the support.

What happens during a Split brain scenario in HA?

AKO detects this during the periodic refresh of the lease lock object and makes itself a passive AKO.

Document Revision History

Date Change Summary
December 24, 2021 Updated the FAQs for AKO version 1.6.1
August 31, 2021 Updated the FAQs for AKO version 1.5.1
April 28, 2021 Updated the FAQs for AKO version 1.4.1
December 18, 2020 Published the Frequently Asked Questions for AKO