deniz.in

Markets

Weather

Loading weather

· via dev.to (home feed)

Portworx Enterprise 3.7.2 reaches GA with Kube Datastore pool migration

Portworx Enterprise 3.7.2 makes Kube Datastore generally available, letting Kubernetes storage pools move between nodes so replica-1 volumes keep running when a node fails.

Portworx Enterprise 3.7.2 reaches GA with Kube Datastore pool migration

Kube Datastore reaches general availability

Portworx Enterprise 3.7.2 has reached general availability with Kube Datastore (KDS), a feature that moves a Kubernetes storage pool from one node to another while the volumes inside it stay accessible. According to a write-up on dev.to, Portworx announced the GA on 9 November 2026, pairing it with Operator 26.5.0 and Stork 26.5.0 as companion releases.

Why Portworx built it

Portworx organises storage into pools, each grouping disks of a single class and type, and replicates volumes between pools. That model fits local disks, which die with their node and therefore need a copy elsewhere. On a shared array such as a Pure FlashArray, which already carries its own redundancy, the software copies bring no extra durability. Portworx points out that data reduction keeps the copies from consuming much space, but they still add writes and acknowledgement latency — a cost the dev.to post describes as a replica tax, and one that weighs most heavily on virtual machines.

KDS takes a different route: a dynamic pool holds only replica-1 volumes, and the pool itself moves when a node goes away. Durability stays with the array; mobility comes from Portworx.

How the handoff works

The post describes three triggers for a pool move: a detected node failure, an administrator-initiated non-disruptive relocation, or a detected imbalance. On failure the pool migrates to another node and the virtual machine restarts there; during maintenance the same move happens without disruption.

Two properties make this possible. Dynamic pools combine ReadWriteMany access modes with block-level intelligence, so the source and destination nodes can both reach a volume mid-transition. And because Portworx sits in the I/O path, it can steer traffic for a virtual machine that is no longer co-located with its pool. Portworx frames the outcome as single-replica efficiency paired with multi-replica resilience, borrowed from the SAN underneath.

Measured performance

Portworx ran its tests on a six-node OpenShift cluster using the 3.6.0 Directed Availability build, backed by a FlashArray over iSCSI, with workloads on the px-csi-vm StorageClass at replica count 1. A live migration of a running CentOS Stream 9 virtual machine finished in under 10 seconds with no observed I/O interruption. Portworx credits the destination node pre-attaching and sharing the volume path, which removes the detach-and-reattach step that usually interrupts a containerised VM move.

A second test cut power to the node hosting a critical VM, forcing the cluster to re-orchestrate the pool rather than merely restart a pod. Recovery took about 51 seconds. The mechanism is a fastpath failover: when the incoming node claims the volume while the old coordinator is still listed as active, KDS registers a dual-attach and forces the volume over to the new coordinator.

Both figures come from a KubeVirt environment. KDS supports containers as well, but every published measurement covers virtual machines.

What the GA release adds

Version 3.7.2 turns KDS from a Directed Availability feature into a supported one and expands its operational reach:

  • The per-node ceiling rises to 12 dynamic pools, up from 4 in the Directed Availability release, with headroom for four more so a node can absorb pools during maintenance or an outage. Cluster scale is stated at 5,000 volumes and 10,000 snapshots.
  • WaitForFirstConsumer binding places the virtual machine before provisioning the volume, so workload and data start on the same node and traffic stays north-south to the SAN. Fresh 3.7.2 installations get a replica-1 plus WaitForFirstConsumer storage class as the default.
  • Local and remote FastPath both work on dynamic pools; remote FastPath exports replicas over NVMe or TCP with automatic promotion and demotion.
  • Affinity groups keep related pools on one node and pull them back together after a failover, limiting how much a rebalance shuffles running VMs.
  • Upgrades become non-disruptive: the operator moves eligible pools off a node before restarting it.

Requirements stay narrow

The stated prerequisites are specific: Red Hat OpenShift with a Pure FlashArray as the shared backend, replica-1 volumes in every dynamic pool, and an opt-in flag in the StorageCluster spec.

Why it matters

Teams running stateful workloads on shared storage keep hitting the same trade-off: pay for software replication they may not need, or give up mobility and fast failover. For OpenShift shops on Pure arrays — particularly those running KubeVirt virtual machines, where write amplification hurts most — KDS offers a third option that offloads durability to the array and keeps node-level agility. The catch is scope: it requires that exact OpenShift-plus-FlashArray pairing, moves durability entirely onto the array, and every published number is vendor-measured and VM-only. Containers are claimed to work, but no measurement has been published for them yet.

  • #kubernetes
  • #cloud-storage
  • #openshift
  • #portworx
  • #kubevirt

Related posts