Kubernetes has launched version 1.31, bringing a crucial upgrade focused on enhancing cluster performance by implementing consistent reads from cache. This feature, now in Beta, solves a key issue for users managing large clusters: providing timely access to the latest cluster state while minimizing any performance impact.
The consistent reads functionality enables Kubernetes components to fetch information directly from a cache, significantly reducing the reliance on the etcd datastore. Historically, retrieving data from etcd could create slowdowns, especially when filtering results based on labels or field selectors. Previously, these operations demanded loading the entire dataset before filtering in-memory, which was resource-heavy.
Now, the watch cache uses progress notifications from etcd to assess data freshness. When a consistent read is initiated, it checks the cache status and, if needed, queries etcd for updates before responding. This innovation lessens the load on etcd while enhancing response times for Kubernetes components, particularly in environments with large datasets.
Performance Improvements
Performance evaluations on clusters with 5,000 nodes show significant gains, including:
- A 30% reduction in CPU usage for the kube-apiserver
- A 25% drop in CPU usage for etcd
- Latency for pod LIST requests decreased from five seconds to just 1.5 seconds at the 99th percentile
To capitalize on these benefits, ensure your clusters are running etcd version 3.4.31 or higher. With consistent reads from cache enabled by default in v1.31, Kubernetes seeks to deliver a seamless performance boost to users. Future updates may introduce pagination support within the watch cache, further enhancing efficiency.
If you want to harness this improvement, upgrading to Kubernetes v1.31 and verifying compatibility with the required etcd version is essential.
For more information, visit the official blog: Kubernetes Blog.
