This message was deleted.
# generala
adamant-kite-43734
11/15/2024, 8:30 PMThis message was deleted.
c
creamy-pencil-82913
11/15/2024, 8:36 PMsee the discussion on https://github.com/longhorn/longhorn/issues/5910
creamy-pencil-82913
11/15/2024, 8:37 PMpod disruption budgets are the responsibility of the application, if you can’t drain nodes because it would violate PDB that’s not a SUC problem.
creamy-pencil-82913
11/15/2024, 8:37 PMeither don’t drain, or fix the pdb
a
abundant-hair-58573
11/15/2024, 8:40 PMInteresting thanks. I'm using the same plan I used for the last couple of upgrades and I don't remember running into this. Although that was with Longhorn 1.5.x, so maybe something changed in 1.6.x.
Is it safe to upgrade without draining? If not, do you know where that setting would be in Longhorn? I didn't explicitly set a pdb
abundant-hair-58573
11/15/2024, 8:41 PMreading through that bug report now
c
creamy-pencil-82913
11/15/2024, 8:47 PMgenerally it is safe to upgrade without draining. There is just a momentary outage of the apiserver while things restart, and pods usually tolerate that fine.
a
abundant-hair-58573
11/15/2024, 8:49 PMWould you recommend deleting the pdb or just not worry about draining?
abundant-hair-58573
11/15/2024, 8:50 PMit's just test (for now) so I won't blame you if something goes wrong 🙂
abundant-hair-58573
11/15/2024, 8:54 PMI tried it without the drain and it seemed to work fine
abundant-hair-58573
11/15/2024, 9:03 PMActually, now that it's hitting the nodes that have running longhorn replicas, it's causing them to rebuild. If it hits all 3 nodes that have replicas of an attached volume, I'm guessing it'll corrupt that volume.
In that case this will require a full outage with all volumes being detaching when upgrading production. Last time it worked fine even with attached volumes, because it wouldn't drain the node with the last healthy replica.
7 Views