Anssers inline. On 12/10/24 6:17 PM, Johnny Hao <qf.hao@hotmail.com> wrote:

Hi Buddies, I am new to ironic, after some study, I have several questions about it, could anyone help to answer? Thanks a lot. 1. According to the workflow chart below, what's the communication method between nova-compute and Ironic API? HTTP API call or RPC call through mq like rabbitmq? So are the communications between Ironic Conductor and other components of Neutron, Glance and Cinder/Swift etc?

Ironic, like all other top-level OpenStack services (e.g. Neutron, Glance, Cinder), exposes a REST API for primary communication between the services. In fact, in Ironic's case, the API is designed to be used with or without Nova fronting it.

2. Where do Ironic services run (Ironic API/Conductor/Drivers and DB)? In the controller node? Or a separate node is also supported?

Ironic should be run as a separate cluster, most deployments have a single, centralized API server cluster, with (depending on scale), multiple conductor groups potentially in different physical locations behind. Typically conductor groups will map to how your datacenter looks physically; for instance, a common configuration would be using a separate conductor group for each computer room in a datacenter. The key with conductors is that they must have some kind of network connectivity to the BMCs of the nodes they manage, additionally if not using virtual media, it needs to be able to serve up DHCP and TFTP as well. Generally speaking, Ironic clusters scale much larger than a nova-compute instance. Some choose to run a nova-compute per conductor group, or to configure shards to spread the same conductor group across multiple nova-computes.

**

-- Best regards QingFeng Hao

Hope this helps, Jay Faulkner