Is PiVirt a drop-in replacement for VMware ESXi?

PiVirt targets the same workloads as ESXi — hypervisor-level VM management, clustering, live migration, and centralized storage — but is built on Linux/KVM instead of a proprietary hypervisor, so there is no per-socket licensing and configuration is declarative YAML rather than a vCenter GUI.

What hypervisor does PiVirt use?

PiVirt manages QEMU/KVM directly through the QEMU Machine Protocol (QMP) via its pivirtd daemon, giving each node full access to Linux virtualization extensions (Intel VT-x / AMD-V).

Does PiVirt support live migration and Kubernetes nodes?

Yes. PiVirt supports QMP-based live migration between cluster nodes over shared storage, and its Copy-on-Write image pipeline with cloud-init provisioning is tuned specifically for fast, disposable Kubernetes worker node bring-up.

Virt

Virtualization Engine

Drop-in replacement for VMware ESXi

The following demo is demonstrating the capabilities of the Virt Engine. This is an appliance that can be used to manage virtual machines, clusters, high availability, disaster recovery, backup and restore, live migration, and container/Kubernetes engine. The engine is designed to be a drop-in replacement for VMware ESXi, Linux and KVM based virtualization solutions.

The following demo is demonstrating the the Virtual Host Appliance console.

Distributed KVM Virtualization

A cloud-native alternative to VMware ESXi

PiVirt (internally codenamed pibase) is a high-performance, distributed virtualization management platform built on QEMU/KVM. A node-agent model runs a lightweight agent on every hypervisor host, coordinated through a ConnectRPC control plane, so cluster-wide operations — provisioning, live migration, storage orchestration — look the same whether you're running three nodes or thirty.

Node-Agent Architecture & Flow

Every node runs the pivirtd agent which manages KVM guests directly via QMP sockets, orchestrates iSCSI/OCFS2 storage, and exposes a gRPC/ConnectRPC interface to administrators and cluster managers.

Core Engine Modules

Node Agent (`apps/agent`)

Lightweight Go daemon running on each hypervisor host. Handles VM state machine transitions, ConnectRPC streaming, Zerolog logging, and health heartbeats over NATS.

Storage Manager (`apps/iscsiman`)

Orchestrates iSCSI targets and OCFS2 cluster file systems across nodes, providing Copy-on-Write QCOW2 disk allocation for instant guest provisioning.

Management CLI (`pvcli` & `pivirtctl`)

Unified command-line interface with interactive terminal dashboards (Charm Bubble Tea TUI) for real-time VM console access, cluster management, and OAuth2 auth.

OS Image DB (`apps/osinfodb`)

Automated image retriever and libosinfo parser, managing base cloud images (Ubuntu, Rocky Linux, Debian) with automatic seed ISO generation.

What's under the hood

Distributed Node-Agent Architecture

A lightweight agent runs on every hypervisor host and talks to the pivirtd daemon over ConnectRPC (gRPC-compatible), so a single control plane can manage an entire cluster of KVM nodes.

Native QEMU/KVM Management

VMs are controlled directly through the QEMU Machine Protocol (QMP) — no libvirt layer in between — with each VM isolated in its own folder holding config, disks, sockets, and logs.

Clustered Storage

iSCSI and OCFS2 (Oracle Cluster File System) provide shared block storage across nodes, with Copy-on-Write QCOW2 overlays on top of a central base-image library for fast VM provisioning.

Cloud-Init Provisioning

Every VM gets a generated nocloud seed ISO with user-data and meta-data — SSH keys, packages, and hostname are injected automatically on first boot, no manual imaging required.

Open vSwitch Networking

VM NICs attach as TAP devices to programmable OVS bridges with VLAN tagging, a built-in Go DHCP server for automatic IP assignment, and NAT/masquerading for outbound connectivity — no dnsmasq dependency.

Live Migration

A handshake between source and target pivirtd instances prepares a paused VM on the target node, transfers state over QMP, then finalizes the switch — enabling load balancing and zero-downtime maintenance.

Secure Multi-Node Clustering

Inter-daemon RPC and live-migration traffic are secured with mutual TLS against a cluster-wide CA; node identity and peer topology are defined per-node in node.yaml.

Kubernetes-Optimized Guests

CPU passthrough (-cpu host), explicit socket/core/thread topology, the q35 machine type with OVMF UEFI, virtio-balloon memory ballooning, and a qemu-guest-agent channel are all tuned for stable Kubernetes worker nodes.

Built-In Observability

Structured logging (Zerolog) and metrics collection (InfluxDB) run alongside every agent, with serial console output captured per VM for debugging cloud-init and boot failures.

PiVirt vs. VMware ESXi

Same job — hypervisor-level VM management at cluster scale — built on open source instead of proprietary licensing.

Capability	PiVirt	VMware ESXi
Hypervisor	Linux KVM (open source)	Proprietary (VMkernel)
Per-socket / per-core licensing
Cluster management API	ConnectRPC / gRPC	vSphere API
Configuration model	Declarative YAML (GitOps-friendly)	vCenter GUI-driven
Live migration
Clustered shared storage	iSCSI / OCFS2	VMFS / vSAN
Software-defined networking	Open vSwitch (VLAN/OVS flows)	vSwitch / NSX
Kubernetes worker provisioning	Native Copy-on-Write + cloud-init	Third-party (CPI/CSI plugins)
CLI tooling	pvcli / pivirtctl (Cobra + TUI)	esxcli / PowerCLI

Built with

ConnectRPC / gRPC

QEMU / KVM (QMP)

Open vSwitch

iSCSI

OCFS2

SQLite

InfluxDB

NATS

Cobra & Viper

Bubble Tea (TUI)

Read the architecture docs or talk to us

Dig into the full PiVirt architecture, YAML config schema, and cluster provisioning guides, or get in touch to discuss a deployment.

Read the Docs Talk to Us