Availability and disaster avoidance are extremely important aspects of any application platform. Although Red Hat OpenShift Service on AWS (ROSA) provides many protections against failures at several levels, customer-deployed applications must be appropriately configured for high availability. To account for outages that might occur with cloud providers, additional options are available such as deploying a cluster across multiple availability zones and maintaining multiple clusters with failover mechanisms.
Red Hat OpenShift Service on AWS (ROSA) provides many features and options for protecting your workloads against downtime, but applications must be architected appropriately to take advantage of these features.
ROSA can help further protect you against many common Kubernetes issues by adding Red Hat site reliability engineering (SRE) support and the option to deploy a multiple availability zone cluster, but there are a number of ways in which a container or infrastructure can still fail. By understanding potential points of failure, you can understand risks and appropriately architect both your applications and your clusters to be as resilient as necessary at each specific level.
An outage can occur at several different levels of infrastructure and cluster components.
By design, pods are meant to exist for a short time. Appropriately scaling services so that multiple instances of your application pods are running can protect against issues with any individual pod or container. The OpenShift node scheduler can also make sure these workloads are distributed across different worker nodes to further improve resiliency.
When accounting for possible pod failures, it is also important to understand how storage is attached to your applications. Single persistent volumes attached to single pods cannot leverage the full benefits of pod scaling, whereas replicated databases, database services, or shared storage can.
To avoid disruption to your applications during planned maintenance, such as upgrades, it is important to define a Pod Disruption Budget. These are part of the Kubernetes API and can be managed with
oc commands such as other object types. They allow for the specification of safety constraints on pods during operations, such as draining a node for maintenance.
Worker nodes are the virtual machines that contain your application pods. By default, a ROSA cluster has a minimum of two worker nodes for a single availability-zone cluster. In the event of a worker node failure, pods are relocated to functioning worker nodes, as long as there is enough capacity, until any issue with an existing node is resolved or the node is replaced. More worker nodes means more protection against single-node outages, and ensures proper cluster capacity for rescheduled pods in the event of a node failure.
When accounting for possible node failures, it is also important to understand how storage is affected. EFS volumes are not affected by node failure. However, EBS volumes are not accessible if they are connected to a node that fails.
ROSA clusters have at least three control plane nodes and three infrastructure nodes that are preconfigured for high availability, either in a single zone or across multiple zones, depending on the type of cluster you have selected. Control plane and infrastructure nodes have the same resiliency as worker nodes, with the added benefit of being managed completely by Red Hat.
In the event of a complete control plane outage, the OpenShift APIs will not function, and existing worker node pods are unaffected. However, if there is also a pod or node outage at the same time, the control planes must recover before new pods or nodes can be added or scheduled.
All services running on infrastructure nodes are configured by Red Hat to be highly available and distributed across infrastructure nodes. In the event of a complete infrastructure outage, these services are unavailable until these nodes have been recovered.
A zone failure from AWS affects all virtual components, such as worker nodes, block or shared storage, and load balancers that are specific to a single availability zone. To protect against a zone failure, ROSA provides the option for clusters that are distributed across three availability zones, known as multiple availability zone clusters. Existing stateless workloads are redistributed to unaffected zones in the event of an outage, as long as there is enough capacity.
If you have deployed a stateful application, then storage is a critical component and must be accounted for when thinking about high availability. A single block storage PV is unable to withstand outages even at the pod level. The best ways to maintain availability of storage are to use replicated storage solutions, shared storage that is unaffected by outages, or a database service that is independent of the cluster.