Torus: Focusing Storage for Kubernetes
summary
A prototype distributed storage layer for Kubernetes that uses etcd for metadata and a Torus server to present immutable inodes made of 512k blocks (default) replicated and sharded with a configurable ring function stored in etcd. Integration uses FlexVolumes and nodes run as a daemonset (data directory mints node identity). Rebalance happens by replacing the hash function and peer-to-peer copying until three replicas exist; raw throughput saturates gigabit in tests while small IOPS sit ~100–200. Not yet production stable; repo: github.com/torus.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:18Agenda
- 0:50Who is this for
- 1:19Computer Networking Storage
- 2:11Storage
- 2:35What do we need
- 3:11Right block storage
- 4:26Replication controller deployment
- 5:06What we want
- 5:45Why we created Taurus
- 6:40General overview
- 8:40File pointer
- 10:01Updating
- 11:33Distribute
- 13:04Rebalance
- 18:21Wrapping Up
related talks
28mPersistent Storage with Kubernetes in Production - Which Solution and Why?
29mReliable and Cost-Effective: Open Storage Strategies for Kubernetes
33mPiraeus Datastore - Persistent Volumes for Kubernetes
36mPersistent Storage with Kubernetes in Production - Which Solution and Why?
37mBuild a Kubernetes Based Cloud Native Storage Solution From Scratch
25mContainer Attached Storage: Open Source Extensions: Kubernetes + Data
19mPersistent Storage for Containers using Gluster
48mIntro to Longhorn: Open Source Cloud-Native Distributed Block Storage Built On and For K8s
