Thanks everyone for joining today's meeting. Here's my summary of the conversation. We covered a lot, so it's longer than usual. Highlights related to bullseye release: - bullseye images for Vagrant have been updated. - Raw images are implemented and an MR is open. It's assigned to Thomas Goirand for validation. - Policy routing configuration for instances with multiple NICs is ready. MR is ready. Bastian wants a chance to test on Azure. Vagrant box update ================== Emmanuel has a new version of the debian/testing64 image uploaded. There are a few bug reports he's looking into, as well as creating a BTS usertag. He's also working on some new CI features for the vagrant image build process. Live image tests ================ My live tests repo at [1] now has ec2 and gce support. It's currently using gitlab.com since it requires ssh and ssh is blocked outbound on the shared salsa runners. I'll add a group runner for this project and move to salsa once and AWS account is ready to host. I also want to start doing gce daily uploads, though it's not clear if debian-cloud-experiments or debian-infra is a better project to use. [1] - https://gitlab.com/rvandegrift/live-tests/-/tree/initial-features fai updates =========== Thomas Lange is working on a new fai release. The pending changes do not affect fai-diskimage, so cloud build tools should be unaffected. Since the freeze is passed, it may be too late for bullseye though. daily image cleanup =================== Thomas Lange also mentioned that his casulana cleanup script is ready for review. Bastian said he will take a look since he is also working on some other cleanup related work. ec2 === Noah has the most recent possible awscli v1 ready for bullseye. He tried to get v2 working, but ran into bugs with python 3.9. v2 will have to wait for bullseye-backports. Also, he may have another dhclient config MR to enable a new prefix delegation feature for AWS. salsa artifact size limits ========================== We had lots of discussion on the salsa artifact size limits. There were two main threads. A few new possibilities for space savings were discussed: - Some small savings might be possible in features that were over the limit. - Images that need to be published as gzip could be passed around in xz, and then converted to gz during an upload step. - Bastian proposed a plan to first write a directory tree with the data, and then create an image of exactly the correct size. This makes it easy to know the min size (unlike shrinking GPT disks, which is very hard). Some of us were concerned that these measures will only go so far, and a more thorough solution is needed. I proposed that someone wanting to build official images could do builds on casulana without gitlab-ci. Bastian pointed out that this requires a lot of setup, so is a difficult alternative. He's right, but at least this would provide a motivated person some way around the 256M limitation. Better ideas welcome! Next meeting will be March 10. Ross
Attachment:
signature.asc
Description: PGP signature