[RFE] Add IMAGE_UPDATE_POLICY that will first "shutdown old server"
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Heat |
New
|
Wishlist
|
Unassigned |
Bug Description
The current workflow when updating an image is:
* Detach interfaces
* Start new server
* Re-attach interfaces
* Delete old server
This workflow fail if the guest instance does not have ACPI capabilities. The "detach interfaces" step return OK, but in fact libvirt/qemu does the interface detach asynchronously and the detach requires guest cooperation via ACPI drivers.
The current workflow works for most guests, but if the guest does not have ACPI capabilities the current workflow will fail with "Device or resource busy" at the re-attach interfaces step if the new guest is scheduled on the same host.
The suggested additional IMAGE_UPDATE_POLICY would have the following workflow:
* Shutdown old server
* Detach interfaces
* Start new server
* Re-attach interfaces
* Delete old server
By shutting down the old server first, the interface detach should succeed even on servers that do not have ACPI capabilities to do PCI hotplug in the guest.
tags: | added: spec-lite |
Changed in heat: | |
importance: | Undecided → Medium |
Changed in heat: | |
importance: | Medium → Wishlist |
Changed in heat: | |
milestone: | none → no-priority-tag-bugs |