Redis Enterprise on Kubernetes: How the Operator Simplifies Deployment
In a session at the Data Science Innovators Conference 2023, Redis' Talon Miller explained why running Redis on Kubernetes is hard and how the Redis Enterprise operator automates it. He also described claimed availability and speed gains and previewed Redis on Flash and active-active deployment.
By The AI Enterprise Magazine Team4 min read
Key takeaways
- Redis on Kubernetes is hard. Miller said open-source Redis needs init scripts, firewall rules, service discovery and manual upgrades.
- The operator automates the lifecycle. It deploys and manages Redis Enterprise clusters from a YAML file.
- Availability claim. Miller cited five nines, or under 30 seconds of downtime per month.
- Speed claim. He said writes can be up to 100 times faster and reads up to 500 times faster than disk-based databases.
- Roadmap. Redis on Flash and active-active deployments on Kubernetes were previewed.
In a session at the Data Science Innovators Conference 2023, Talon Miller of Redis explained how the Redis Enterprise operator is meant to simplify running Redis on Kubernetes, and why that matters for data science and AI workloads. Miller, listed as Senior Product Marketing Manager at Redis, presented a vendor tech talk titled "Make Kubernetes Even More Efficient," covering the operator, how it is configured and what Redis had on its roadmap at the time.
The session was a product presentation, and the performance and availability figures below are Miller's own claims, not independent benchmarks.
Why is Redis hard to run on Kubernetes?
Miller described Redis as the one data management system that works in memory, so it handles transactions faster than a traditional database. He said it is a go-to option anywhere a response is needed in under a second, and that ChatGPT's application used it to serve data queries and features.
Kubernetes, he said, is a container technology for deploying applications and infrastructure quickly. Pairing the two is the difficult part. Miller said open-source Redis on Kubernetes needs advanced init scripts, firewall rules and service discovery, and still requires manual work for version upgrades and resizing clusters.
What does the Redis Enterprise operator do?
An operator, Miller explained, is built to handle stateful applications, which can include an AI model that stores and feeds features. He called it an "expert in a box" that deploys and manages the lifecycle of a Redis Enterprise cluster, filling the gap between what Kubernetes automates and what Redis needs.
Without it, he said, Redis on Kubernetes can fail to react correctly to internal edge cases and failure conditions. Kubernetes orchestration also runs outside the Redis cluster, so it could trigger failover events in split-network scenarios. The operator divides responsibilities between what Kubernetes does well and what Redis Enterprise does well.
He said the operator gives one deployment approach across Kubernetes environments including GKE, AKS, Red Hat OpenShift, VMware Tanzu and EKS, which supports moving workloads between multi-cloud, hybrid and on-premises setups.
How does it address availability and performance?
On durability, Miller said Redis Enterprise uses network-attached storage. When a pod fails, Kubernetes starts another and reconnects it to the same storage, and Redis Enterprise loads the data back. He said this gives "five nines" of availability, which he described as less than 30 seconds of downtime per month for models and feature serving.
He also claimed that, because Redis Enterprise needs a single I/O operation for a write and none for a read, write requests can be up to 100 times faster and reads up to 500 times faster in typical Kubernetes environments than with disk-based databases such as MySQL or disk-based NoSQL stores.
Miller added that Redis deploys multiple database instances on a single pod instead of the usual one service per pod. In his account, this avoids a bottleneck in the Kubernetes network and makes better use of CPU, memory and network resources while keeping isolation.
How does the operator work in practice?
The operator supports two custom resource definitions: the Redis Enterprise cluster, or REC, and the Redis Enterprise database, or REDB. Miller said a single YAML file states how much memory and how many nodes are needed and where, and the operator creates the cluster, stores database credentials in a Kubernetes secret and exposes a service endpoint.
In his example, changing the node count from three to five makes the operator update the stateful set. Kubernetes then bootstraps new nodes one at a time, and each joins the cluster. Miller said that doing this with open-source Redis "takes months and months of work."
Miller said the setup also provides elastic scaling that can scale down when usage drops, which he said helps control cost, along with built-in security controls and encryption.
What was on the Redis roadmap?
Miller pointed to two items. Redis on Flash for Kubernetes lets large datasets, common in data science and AI, live in flash instead of RAM at lower cost. Active-active deployment, then in public preview on Kubernetes, replicates data across clusters in several regions so each user gets sub-second responses. He said configuring it through YAML would save customers significant time.
Where to find more data and AI sessions
For practitioners comparing infrastructure for models and feature stores, the sessions AI Connect hosts are a place to hear from vendors and users directly. The 2027 schedule includes 24 free live webinars in four series: the AI × Enterprise Panel Series, AI Demo Days, Agentic AI Showcases and the Plug & Play Series. Details are at /webinars/, and registration is coming soon.
Frequently asked questions
What is the Redis Enterprise operator?
Miller described it as an expert in a box that deploys and manages the lifecycle of Redis Enterprise clusters on Kubernetes.
Why is open-source Redis difficult on Kubernetes?
Miller said it needs advanced init scripts, firewall rules and service discovery, and manual work for upgrades and resizing.
Which Kubernetes platforms does the operator support?
Miller named GKE, AKS, Red Hat OpenShift, VMware Tanzu and EKS, among others.
What is Redis on Flash?
Miller said it stores large datasets in flash instead of RAM to cut cost while keeping Redis performance.
This article is based on remarks by Talon Miller, Senior Product Marketing Manager, Redis at Data Science Innovators Conference 2023 on April 12, 2023. More coverage: More AI Connect Events.
Filed under: Data & AI Infrastructure
