How stable is "subscription" repository?

[agenkin]

I'm evaluating Proxmox for use in our environment. I had a node with version 8.1 installed last week with non-subscription repositories. The node updated to 8.2 yesterday.

The update knocked the node off the network because the new kernel renamed the Ethernet interfaces, and the bridge configuration in /etc/network/interfaces became invalid. Took me about 20 minutes to figure it out, good thing it was not in production.

My question is: if I had "subscription" repositories (which are supposed to be more stable), would I still have received the new kernel yesterday, which would have broken my node? Or did this happen, primarily, because I was using the non-subscription repository?

Is there a repository that would only provide security and stability updates, but no major updates like new kernels?

 

[dietmar]

The update knocked the node off the network because the new kernel renamed the Ethernet interfaces, and the bridge configuration in /etc/network/interfaces became invalid. Took me about 20 minutes to figure it out, good thing it was not in production.

We are working on a fix for that, so that such kind of things can't happen in the future...

 

[bbgeek17]

It's too late for you, this time, but before applying major upgrades, especially those involving Kernel, one should read "Breaking Changes & Known issues" portion of Release Notes.

The interface name change is not limited to upgrades. It can happen with a new PCI device being installed, or removed. Unfortunately, it's a Kernel architectural design that PVE team has no control over.

There is a way to prevent renaming from happening. One needs to make proactive changes, which are discussed in this thread: https://forum.proxmox.com/threads/proxmox-ve-8-2-released.145723/page-2#post-657026

The Enterprise repo would not have saved you here. The device rename will still happen when the new Kernel is rolled out to Enterprise.

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

 

[bbgeek17]

We are working on a fix for that, so that such kind of things can't happen in the future...

Proactive name pinning during initial setup is the way to go! And a procedure to help existing userbase.

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

 

[agenkin]

Thanks for replies, everyone.

Interface names can be dealt with (with custom naming), but I'm a bit concerned about the bigger picture. I may be unlucky, but I've only been playing with Proxmox for about a week, and it already broke from an official update.

I wish I could have a production-safe package repository that would give me only security updates and bug fixes, but no major changes (e.g. major kernel or library upgrades), at least not without an explicit opt in.

 

[leesteken]

I wish I could have a production-safe package repository that would give me only security updates and bug fixes, but no major changes (e.g. major kernel or library upgrades), at least not without an explicit opt in.

Maybe stay on the latest 7.4 (which still supported) until 9.0 comes out and only then go to the latest 8.x, it you're really conservative.
Don't take this minor update from 8.1 to 8.2 on your system (in your short period of using Proxmox) as representative for all minor updates in the future. It went flawless for many other systems (as it almost always did in the past).

 

[zzz09700]

Well, I guess you haven't seen how broken a machine can go with official updates from, other very respective software vendors (cough, VMware, cough, Microsoft, cough)

We are still with VMware at the time being and some of our machines are still running vSphere 6.7.

Oh did I mention vSphere 7 broke a lot of customers VMs with an Ux update? That was sooo broken that VMware had to take down the update. The only way out back then was re-install ESXi through the ISO. (No big deal here since all your VMs can be migrated to other nodes and you just need to re-install ESXi and reconfigure the node to re-join the cluster/link up with network storage, you will be fully back on line in just a few days, I promise you, lol.)

I'm staying put and see how the update rolls out now. Installating updates in the first wave is a glorious expedition and I'm too old to go onboard.

 

[agenkin]

Well, I guess you haven't seen how broken a machine can go with official updates from, other very respective software vendors (cough, VMware, cough, Microsoft, cough)

You're right, this is my first foray into a product like this. We've been using KVM/qemu extensively, but without any kind of UI around it.

 

[zzz09700]

You're right, this is my first foray into a product like this. We've been using KVM/qemu extensively, but without any kind of UI around it.

Rule of thumb: don't trust any update from any software vendor unless you have concrete proof that those updates can be trusted.
So either put together some testing gear and see how the update goes on that, or wait for others to do it.

Enterprise repo basically forces you to wait for others (who are on free repo) to test it/iron out any remaining bugs for ya.

Microsoft once doomed all of our Chrome browsers with a windows update. That was a marvelous Wed.