HwameiStor is a local storage system for Kubernetes that turns the disks on your worker nodes (HDD, SSD, NVMe) into a pool of persistent volumes exposed via CSI. It is developed by DaoCloud and aimed at workloads that want the performance of direct-attached storage without giving up dynamic provisioning or volume lifecycle management.
Architecturally it runs as a set of controllers and per-node agents that discover local block devices, group them into storage pools by media type, and carve out LVM-backed logical volumes on demand. A scheduler extender ensures pods are scheduled onto the node that actually has the data, and it supports volume replication across nodes for HA, snapshots, expansion, and migration when a node is drained. Volumes are exposed through the standard CSI driver.
It occupies the same niche as OpenEBS LocalPV / Mayastor, Longhorn, and TopoLVM: use local NVMe for databases and AI/ML workloads where the Ceph-style network-replicated approach is too slow. Compared to Longhorn it leans more toward raw LVM performance and less toward distributed block replication. CNCF sandbox.
