Skip to main content

​
Overview

Polystack Compute follows a distributed, service-oriented architecture. API requests enter through a central API tier, are routed through the Scheduler and Placement services to select the optimal host, and are executed by Compute Agents running on every hypervisor node. Each agent communicates directly with the local hypervisor to manage instance lifecycle operations — creation, power state, migration, and console access.
Administrator Access Required — This operation requires the admin role. Contact your Polystack administrator if you do not have sufficient permissions.

​
Service Topology

The diagram below illustrates how requests flow from the end user through the Polystack Compute control plane to the hypervisor layer.

​
Service Components

The Conductor service decouples compute agents from direct database access. All database writes from hypervisor nodes flow through the Conductor, which enforces access control and serializes state transitions.

​
Service Lifecycle

The following table describes the expected runtime state of each service on a healthy cluster.
In the deployment console, navigate to Operations and run Prechecks to validate all Compute service components across the cluster. The output shows:
  • Per-service status on every node
  • Last heartbeat timestamp
  • Service version and host assignment
Use Reconfigure from Operations to recover a failed component without a full node reboot.

​
Request Flow: Instance Creation

Understanding the creation flow helps diagnose failures at each stage.

API receives the request

The Compute API validates the request, checks quota, and creates an instance record in the database with status BUILD.

Placement selects resources

The Placement API queries resource inventories to find hosts with sufficient vCPU, RAM, and disk. It returns a list of allocation candidates.

Scheduler filters and ranks

The Scheduler applies the configured filter chain to eliminate ineligible hosts, then ranks remaining candidates using weighers. The top-ranked host is selected.

Conductor orchestrates the build

The Conductor sends a build_instance RPC call to the Compute Agent on the selected host. It monitors progress and updates the database with state transitions.

Compute Agent creates the instance

The Compute Agent on the target hypervisor node provisions the instance — downloading the image, allocating network interfaces, attaching volumes, and starting the virtual machine. The instance transitions from BUILD to ACTIVE.

​
Next Steps

Compute Hosts

Manage hypervisor nodes — list, inspect, enable, and disable hosts.

Scheduling

Configure filters, weighers, host aggregates, and availability zones.

Admin Guide

Return to the Compute Administration Guide index.