docs: document safer procedures for updates#435
docs: document safer procedures for updates#435psafont wants to merge 1 commit intoxcp-ng:masterfrom
Conversation
|
We probably need to double check with the XO team about our RPU algorithm, if it's taking this into account (I think it does, but it won't hurt to make sure it does). |
psafont
force-pushed
the
dev-pau-update
branch
from
52afc12 to
bb3c86e
Compare
stormi
left a comment
stormi
left a comment
There was a problem hiding this comment.
Thanks for the improvement. I was sure that we already mentioned host-disable and host-evacuate, yet apparently we didn't (I'm puzzled. I did compare the update process in our docs with that applied by Xen Orchestra a few months or years ago...).
I also think it does, however comparing to make sure we follow the same steps in a manual update vs an XO RPU would be good. |
|
Adding @Wescoeur to ensure that the updated steps are fully compatible with an linstor upgrade (I think they are, but better double check). |
Previous procedure recommended updating the packages in all hosts, and after that was complete, restart all hosts. This is dangerous sometimes because the processes running in a host and the filesystem are out of sync. Instead recommend updating and restarting one host at a time. The previous procedure did not mention how to avoid running tasks in a host before updating, now this is done using host-disable. Now the choice of rebooting or restarting xapi is put before they are explained, because the former needs to shut down the VMs, or evacuate them onto other hosts, while the latter doesn't, so they must choose before starting the procedure to be consistent. Signed-off-by: Pau Ruiz Safont <pau.safont@vates.tech>
psafont
force-pushed
the
dev-pau-update
branch
from
bb3c86e to
79ce986
Compare
Wescoeur
left a comment
Wescoeur
left a comment
There was a problem hiding this comment.
Regarding LINSTOR, I think that's ok.
5 participants
Previous procedure recommended updating the packages in all hosts, and after that was complete, restart all hosts. This is dangerous sometimes because the processes running in a host and the filesystem are out of sync. Instead recommend updating and restarting one host at a time.
The previous procedure did not mention how to avoid running tasks in a host before updating, now this is done using host-disable.
Now the choice of rebooting or restarting xapi is put before they are explained, because the former needs to shut down the VMs, or evacuate them onto other hosts, while the latter doesn't, so they must choose before starting the procedure to be consistent.