Enhance master images caching for PXE driver
To speed up large deployment, Ironic Conductor requires local caching of master images downloaded from Glance. Currently, it stores image while it is used by any instance, and any newly deployed instance may also reuse this image.
This blueprint proposes the following enhancements:
* Store images for some time after they're not in use by any instance, limit cache in size and TTL (bug #1199665)
* Delete instance images (but not TFTP images) right after deployment, so that they can be deleted from cache eventually (bug #1231351)
* Consider also Glance checksum when caching. For example, add it to file name of master image in local cache, so that when image is updated in Glance, it is downloaded again and stored in cache under a different file name. This image will only be used for subsequent deployments, old deployments will still use the old image.
NOTE(dtantsur): I have doubts now that Glance does support updating active image, so I suggest withdrawing the third action item.
NOTE2: As per dicussion on IRC, the 3rd point is dropped, so there's no more point in this blueprint.
Blueprint information
- Status:
- Complete
- Approver:
- None
- Priority:
- Undefined
- Drafter:
- Dmitry Tantsur
- Direction:
- Needs approval
- Assignee:
- Dmitry Tantsur
- Definition:
- Obsolete
- Series goal:
- None
- Implementation:
- Unknown
- Milestone target:
- None
- Started by
- Completed by
- Dmitry Tantsur
Related branches
Related bugs
Bug #1199665: download images only once per conductor in pxe driver | Fix Released |
Bug #1231351: nova-bm keeps a copy of baremetal image after deployment | Fix Released |
Sprints
Whiteboard
Gerrit topic: https:/
Addressed by: https:/
Implement more robust caching for master images
Addressed by: https:/
Consider Glance checksum when caching master images