Please check the 3.5 release notes as there are new requirements https://maas.io/docs/reference-release-notes-maas-3-5 . By heart, I’d say you have 2 regions and one of them is not working properly. From 3.5 the regions must be routable and you can’t upload custom images to a VIP or to HA proxy
Thank you for the rapid response. I’m working on the same system.
We do have a second regiond, which came about because we built our MAAS setup with a single regiond+rackd some 7 years ago, then we built a separate regiond a few years later and disabled regiond on the original. Then we added a second rackd machine. The database is on a dedicated postgresql machine. So the MAAS components are three separate machines:
1x regiond
1x rackd for one set of racks
1x rackd for another set of racks, with disabled regiond daemon
We have not seen any instructions on how to get rid of the old regiond, so we stopped the daemons and it has worked pretty well so far. Should we now look to properly get rid of the original regiond? How would we do that?
Regarding the changes 3.4 → 3.5, we did notice the database shrink, and there is 4.7GB of files in regiond:/var/lib/maas/image-storage. I don’t know if they were there before, and since we are planning to roll back to snapshots running 3.4 tomorrow morning (it’s now late evening here) we will have another look at that.
Also, we noticed at https://maas.io/docs/how-to-manage-controllers#p-9026-move-a-rack-controller that Moving a rack controller between MAAS instances or versions is unsupported and we should delete and re-register with new versions. We went to do that, and the first rackd we tried complained of the subnets it is looking after DHCP for, and did not delete. Should we be continuing to do this, and if so how? I guess conceptually we could:
clone the rackd VM
change its config (network address etc)
enrol the clone as a second rackd controller for those subnets
At the controller page, it is listed as Region and rack controller and when we click into it, it shows
(x) regiond
(!) rackd - 50% connected to region controllers (makes sense)
I can’t see a way to just delete the region controller from that. The delete option seems to delete the whole controller. That was disallowed by the DHCP & subnets I mentioned earlier.