Conversation
|
I'd like to point out that I have already submitted a version of buildtasks that integrates cloud-init for the Openstack build #23. It's already at version 3 and quite well tested in our cloud. Interaction between cloud-init and inithooks is managed by Comparison:
Any other characteristics that I've forgotten to test? Bugs:On OpenStack, I've got to wait for a timeout until it goes to the right datasource. Sudo is not installed, so the users created by cloud-init cannot do much. Opinion:I like the cloud-init first approach, because IMHO it works as users would expect. From our OpenStack catalog, you can run it as any other image, log in with a key and get the customization menus. |
|
Thank you for the swift response and detailed comparison @vondrt4. Actually, the login name is admin and the root device is resized with my changes. This is done by inithooks though, as usual; if your table only relates to cloud-init then that's all correct. Re datasource delay, you can specify datasources per build to avoid querying ones that won't be present. |
|
Sorry for stepping in so late into the discussion but I finally noticed the email Jeremy sent me to take a look and compare the pull requests, which I missed earlier. As far as I can tell it looks like Tomas's pull request is:
On the other hand, Anton changes the minimum needed to add support to cloud-init with a minimal amount of extra complexity. Unless I'm missing something that sounds like what we want. |
|
OK, I should really read my mail in the opposite order. I see now that we already settled on a hybrid approach for moving forward. |

If inithooks does not recognize the usual shebang (
#!), it passes the data to cloud-init. This allows us to parse all the other formats. If the userdata is the usual format with a shebang, cloud-init never gets called.This has not been tested for cloud providers other than EC2.