This browser is no longer supported.
Upgrade to Microsoft Edge to take advantage of the latest features, security updates, and technical support.
Hi,
I have created a AKS cluster with Availability Zones enabled (all three Zones). In node pool, availability zones are enabled (all three) and scale method is "autoscale" with a minimum node count of 1. Assume that the cluster is running with just 1 node (no scaling was required till this point). Now if the zone hosting the node (VM) fails, would the node and the pods be automatically created in next available zone. If yes, how long does it take - minutes or seconds. Any help would be appreciated.
Regards,
Rajiv
Hi @Rajiv Bansal,
By default, the auto scaler will scale new node but no guarantee to be the next zone since Azure platform automatically distributes the underlying VMs across the specified availability zones, and scaling new node time depends on the scan internal that you configured when you enabled the auto scaler, See the details here.
For your scenario it is recommended to enable auto scaler with this --balance-similar-node-groups to balance the zones.
Hope this helps, and please "Accept as Answer" if it helped, so that it can help others in the community looking for help on similar topics.
Thanks @mutaz-msft
I would like to confirm one thing. I have 5 microservices which don't need much resources, so one node is sufficient. But I want to provide Zone redundancy for the scenario where a Zone fails. Cold start is acceptable. So if I go with enabling zone redundancy on AKS and node pool. Keep minimum node as 1 and enable auto scaler, I will get zone redundancy although there will be some time in detecting the failure, provisioning the node and scheduling pods on it.
Hi @Rajiv Bansal,
No, one node is not sufficient, you need at least 3 nodes since there are other system components, and kubernetes needs time to detect the lost node (default 5 mins) in case the zone failed and the node was not reachable, during this time the pods status on the lost node will be Running state and cluster auto scaler will work only if there is any pending pod, so this need the Node controller to evict the pods from the lost node and once they are in pending state the auto scaler will scale a new node, but there will be a downtime for your workloads.
It is recommended to have multiple nodes and multiple replicas for your workloads and disturbe your workloads on the nodes to avoid the downtime on any zone failure.