Broadly speaking, hosts running nova can be broken into control nodes and compute nodes Compute nodes are where VMs (“instances”) run and interact with other services for things like networking or storage Control nodes handle everything else, including API requests, choosing a compute node, managing access, … Anything more than a toy deployment will have multiple control nodes and many compute nodes
nova-api Receive HTTP requests from users and other services nova-scheduler Determine which compute node to schedule an instance on; talks to placement nova-conductor Manages DB access on behalf of compute nodes and “conducts” long running tasks like live migration nova-compute Host and manage the lifecycle of instances
nova-api-metadata Expose metadata that can be used by e.g. cloud-init; part of nova-api nova-novncproxy Provide a proxy for VNC consoles nova-serialproxy (unsupported in OSP) Provide a proxy for serial console nova-spicehtml5proxy (unsupported in OSP) Provide a proxy for SPICE console
nova relies on INI-based configuration files to configure individual services and flavors and image metadata to configure instances Most operations can be achieved via the RESTful API, but some management operations require a separate tool.
nova-manage A collection of various admin-specific admin commands (admin only) nova-policy Inspect policy configuration (admin only) nova (novaclient) CLI for the nova API with some admin-only APIs openstack (openstackclient) Unified CLI for all OpenStack APIs; mostly end-user focused
➡ API ➡ Conductor API validates request and passes it to the conductor Conductor ➡ Scheduler (➡ Placement) The scheduler requests allocation candidates (here, a compute node) from placement and then filters and weighs these to select one Conductor ⬅➡ Compute Instance is scheduled to host and compute takes over, building the instance and talking to other services (e.g. neutron) where necessary Fin.
RESTful API Well, mostly. There are a couple of RPC’y APIs. Microversions Each change requires a new version (2.1, 2.2, …, 2.87). Support all microversions but some APIs will 404 No (more) proxy APIs or extensions Want networking? Talk to neutron. Ditto for storage etc. Documentation! docs.openstack.org/api-ref/compute
Filters and weighers Filters eliminate hosts and weighers set preferences Configurable Additional filters and weighers ✅ (but different schedulers ) Placement is king An authoritative resource tracker; VCPU, DISK_GB, MEMORY_MB, …
Cells (v2) Multiple conductors to help scale to 100s of nodes SR-IOV, PCI Passthrough, FPGAs, vGPUs, ... Yes, we support them. Cyborg is cool “I want this specific guest configuration” Use RHEV/oVirt
stephenfin
moriyuya
grimoh
yayoi_dd
whywaita
hiashisan
ymotongpoo
yrhorita
colopl
.png){kind=icon}
pharma_x_tech
linedevth
non97
yamasatimi
rmw
cassininazir
zakiwarfel
palkan
trallard
eileencodes
lauravandoore
jonyablonski
brianwarren
destraynor
marktimemedia
philhawksworth