So, you’ve finally decided to use Kubernetes for stateful applications? Congrats! (And good luck.)
Let’s put the Champagne back in the ice bucket and talk about data – the chain that binds your stateful architecture to a single location. If you’re only using a single region, you’re in luck. But what happens when the same application needs to run on multiple regions? Or even worse – multiple clouds?
Stateless applications use service meshes so that the application layer could communicate across clusters. But stateful applications are a different animal. They require you to have available synced data.
Now you are faced with some tough questions. How can you ensure that your application is running consistently if the distance between the application and its data varies? How would you solve that issue, and why venture into it all that mess at all?
Stateful Challenges Come at Scale. Remember the CAP Theorem?
If I’m running a database on a Kubernetes cluster, all the pods require access to a local volume to store and read data. In other words, any entry that was done with one pod should be seen by the rest of the pods.
Consistency, or the requirement that every read is updated with the latest write, sounds simple. But if your goal is to enjoy the true benefit of distributed network availability, limiting yourself to applications that run close to their data with as little room as possible for error is not enough.
Not a problem, you might say. I’ll set up a centralized database to take care of all my pods and clusters stateful requests.
Congrats, you’ve just introduced a single point of failure to unite them all; if something happens, none of your pods will have access to data, a double-edged sword that breaks the partition tolerance.
Balance is key, and the tradeoff between consistency, availability, and partitioning is of paramount importance. Could we solve this by simply adding another cluster?
What is Multi-Cluster, and What to do With it?
Once you’ve designed and coded your application and you’ve built containers, in theory, all that is left is the simple task of running them. But getting from code to a running up is not nearly as simple (as anyone who has ever built a containerized application will attest.)
Before deploying to the production environment, you need to run various dev/test/stage cycles. You also need to think of scale — your production application may need to run in many different places for reasons like horizontal scalability, resiliency, or close proximity to end-users.
Multi-cluster is a deployment strategy that runs multiple Kubernetes clusters. Running multiple clusters is common, but the issues start when you need pods to communicate with one another.
Multi-cluster is a strategy to deploy containerized applications across multiple Kubernetes Clusters.
Multi-Cluster Use Cases:
- Improved application availability – A cluster that does not have another cluster is a single source of failure. Having multiple cloned clusters that can failover in case my main cluster is damaged provide higher regional performance.
- Support for large organizations – I run multiple clusters in different environments. Multi-clustering will consolidate all my clusters to a single management portal, giving me the ability to deploy applications across multiple availability zones and clusters. By standardizing the cluster creation across environments, I can reduce overhead and time to market for features and updates. In addition, multi-cluster deployments are easily scalable.
- Isolation – The ability to un-multicluster by creating individual fault domains. Updating of the clusters can be phased to reduce the impact of faulty versions or malicious code.
- Performance – The closer the application’s proximity to the end-user, the lower is the latency and the risk to data in transit.
- Compliance – There are laws in many countries that govern where you can store users’ data. Depending on the regulation, you might have to store the data from users in China in China, and you might need to store the data from users in Russia in Russia. Having a system that spans multiple regions enables you to do just that. If you only have a data center in the US, then you’re going to have a tough time working with a global user base.
Federating Stateful Applications
The idea behind federation is to provide a single configuration to manage the application across multiple clusters or regions.
Federation use cases:
- Reduced Configuration Management Complexity – A single location to consolidate cluster management. In this use case, the data is not shared across the application, and it works well for stateless applications.
- High Availability (HA) – Add a cluster redundancy for business continuity (BCP), which is also a good solution for stateless applications.
Stateless Applications Enjoy the True Benefits of Multi-Cluster and Federated Kubernetes; Stateful is a Different Story
The portability of stateless applications gives them the ability to run anywhere, but not all applications are stateless; most applications are dependent on data, data that does not act by the same rule book as stateless applications.
Data bind the application to its storage locations. A physical location becomes an app dependency, and every request from the data creates latency according to its distance from the application resulting in service inconsistency.
When it comes to stateful applications, you can solve those problems by treating your state just as you do your containers. What I mean – instead of forcing the application to run where the data happened to be originally provisioned, the data needs to follow the application.