grub 2.04 net does not like deployments with kvm maas pods
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
grub2 (Ubuntu) |
Fix Released
|
Undecided
|
Unassigned | ||
Focal |
Fix Released
|
Undecided
|
Unassigned | ||
Groovy |
Fix Released
|
Undecided
|
Unassigned |
Bug Description
[Impact]
* MAAS KVM LXD pods fail to deploy, in very beefy servers.
1) cherrypick upstream fix that caused crash upon completion of http networking
2) revert patches that add support for TCP window scaling and non-ethernet cards
With above changes one can deploy 100 out of 100 MAAS KVM LXD pods using the patched grubnetx86.efi
[Test Case]
* Configure maas
* Configure networking for lxd kvm pods
* Deploy ubuntu focal on a node
* Manually init lxd, allow netowrking, add remote password
* Add kvm lxd host in the kvm page of maas
* Disable image syncing in maas
* replace /var/snap/
* compose & commision hundred nodes with cli api:
for i in `seq 100`; do maas ps5 vmhost compose 8; done
(where 8 is the id of the kvm host, see url for the kvm node in question)
Without these patches failure rate is up to 30%. With these patches 100 out of 100 pods deploy fine.
Thus one should be ok testing with less number of pods too.
[Where problems could occur]
* These patches mean that we are using smaller (same ones we did in bionic) TCP window, thus for very large kernels/initrds we may hit http timeouts on the server. Also it means that throughput speed of deployments is lower again.
* However it is better to deploy where we used to deploy before, than failing to deploy small things on small networks.
description: | updated |
Changed in grub2 (Ubuntu): | |
status: | New → Fix Committed |
https:/ /paste. ubuntu. com/p/8wSwG4PSp g/