Proxmox VE: Are we getting too many kernel updates for an enterprise virtualization platform?

Back when we were on VxRails, the patches, which took 1-4 hours to apply for our cluster, would straggle in sometime after VMware released them. Having been burned once by Dell pulling a patch after I'd installed it, I learned to wait at least two weeks after it had been released before applying. Further, they encourage running an elaborate pre-flight check called VxVerify on all the nodes. Its findings occasionally required opening an SR, fixing the issue, and then patching.

So, no, I don't rush to apply patches, though the thought of guests breaking confinement is unsettling.

My Ubuntu servers all run a daily unattended update, but aren't necessarily in sync due to phasing and different OS versions.

My Windows servers get patched within a few days after the patch is available, though I stagger the DCs by days or weeks due to another learning experience.
 
Back when we were on VxRails, the patches, which took 1-4 hours to apply for our cluster, would straggle in sometime after VMware released them. Having been burned once by Dell pulling a patch after I'd installed it, I learned to wait at least two weeks after it had been released before applying. Further, they encourage running an elaborate pre-flight check called VxVerify on all the nodes. Its findings occasionally required opening an SR, fixing the issue, and then patching.

So, no, I don't rush to apply patches, though the thought of guests breaking confinement is unsettling.

Welcome to the new normal. In this age waiting multiple weeks just isn't feasible anymore. And even the idea of "first testing updates in a test environment" (e.g. a virtual PVE cluster inside your production cluster) doesn't really work since your test environment won't have the same production workloads which might get hurt by the patch. Imho you can't really do much than using clustering and then first patch one node, wait for any problems, then the next.
If something breaks, wipe the node and reinstall (cattle, not pets).
But not after multiple weeks, but more days if not hours since most known vulnerabilties get used by bad actors less than 24 hours after they are discovered:
https://zerodayclock.com/ They also have a timeline: https://zerodayclock.com/collapse

Even worse due to the amount of fixed CVEs etc in each kernel most admins don't have the time to triage whether they are actually vulnerable to a specific bug or not. It's way less work to assume that one of the multitude of issues fixed in the latest patch IS an issue on your system and just apply the patch instead of wasting time with triage.


My Windows servers get patched within a few days after the patch is available, though I stagger the DCs by days or weeks due to another learning experience.

Microsoft failing quality assurance is not a reason not to patch systems by other vendors.
 
"I would love to see Proxmox adopt a more conservative kernel release strategy for production environments, while continuing to deliver critical security fixes promptly."

None forces you to apply each and every updates, but if you are too lazy to do it, please at least don't bring others with you !
 
  • Like
Reactions: Johannes S
I'm hoping that Proxmox is looking at the updating issue and perhaps will be releasing some updates to PDM and POM to help address managing updates in large PVE environments. This is one area where proxmox has just not caught up to the competition yet.
 
I love all the patches. Better than sitting around waiting (ala VMware). As many have suggested it would be nice if there was something similar to VMware where you could tell it to patch a cluster and it would shuffle around and reboot each node as required.

But for free software I cannot complain at all.
 
I'm hoping that Proxmox is looking at the updating issue and perhaps will be releasing some updates to PDM and POM to help address managing updates in large PVE environments. This is one area where proxmox has just not caught up to the competition yet.

I can understand the need to be able to upgrade the hypervisor nodes from one single view of glass. But I don't get where the connection to the original post is? OP wished to have less updates due to "enterprise" but still get security patches. This doesn't work in todays state of the world and is in no way connected to the possible methods of updating hypervisor nodes.
 
Last edited:
As many have suggested it would be nice if there was something similar to VMware where you could tell it to patch a cluster and it would shuffle around and reboot each node as required.
There is already a possible way, although a bit limited. If you have HA enabled for a certain guest you can patch and reboot the node and HA will automatically migrate every guest with enabled HA to other nodes before the reboot und move back afterwards. You can also manually enable/disable maintenance mode for a node the same way. It's not the same as patching the whole cluster in one go though and obviouvsly need some shared storage or ZFS storage replication to work.
 
Last edited:
I've been following this discussion with interest, and I'd like to add some numbers and a practical perspective.

As of November 18, we've had ten kernel releases in just 43 days, from 7.0.14-21 to 7.0.14-30.

I operate a three-node Proxmox cluster with GlusterFS. Updates are installed automatically, and rolling reboots are already partially automated. Nevertheless, every kernel change requires some attention, particularly regarding storage consistency, quorum and VM availability.

I agree with those arguing that security fixes must not be delayed. But I think we're overlooking another aspect: the quality of the upstream kernel and the operational consequences of this release frequency.

For example, the recent dm-thin data corruption fix addresses a bug introduced in March 2023. It remained undiscovered for more than three years and was eventually fixed by adding a single line of code.

This raises questions about upstream test coverage, concurrency testing, regression detection and the potential of AI-assisted code reviews.

Of course, more frequent updates can also indicate that more bugs are being discovered, which is a good thing.

My concern isn't that Proxmox publishes fixes too quickly. It's that administrators are increasingly forced to choose between frequent hypervisor reboots and running kernels with known vulnerabilities.

Clearer information about the security relevance and urgency of each kernel release would help enormously.

I don't want fewer security fixes. I want fewer bugs reaching production in the first place, and better information to make informed maintenance decisions.
 
I don't want fewer security fixes. I want fewer bugs reaching production in the first place, and better information to make informed maintenance decisions.
Who wouldn't? Unfortunately, it just doesn't work that way (TM).

Your example of a longstanding kernel bug from 2023 actually illustrates my point. Back then, AI-assisted code review was nowhere near as capable as it is today. Interestingly, the recent one-line fix even credits Claude for assistance.

While newly submitted code should nowadays benefit from AI-assisted code reviews, the real challenge is the enormous backlog of existing code.

AI has dramatically accelerated vulnerability discovery, and we're already seeing the consequences in the current flood of CVEs and security patches. Many of these vulnerabilities have been lurking in production code for years, sometimes decades.

I therefore expect this wave of discoveries and subsequent fixes to continue for several years, particularly in software as complex as the Linux kernel. Expecting fewer kernel updates anytime soon seems like wishful thinking to me.

Challenging times, I know. In fact, I'm currently fighting back with my local AI: I'm using it to inventory my infrastructure, correlate installed software with newly published CVEs, improve network segmentation to mitigate lateral movement, and automate upgrades of my OPNsense firewalls and Proxmox VE and Backup Server hosts.

It won't make the bugs disappear, but it should help me stay ahead of them despite limited available time.
 
Last edited:
  • Like
Reactions: Johannes S
You make a very good point, and I agree that AI-assisted code reviews may initially lead to more fixes as previously undetected bugs are discovered.

That's actually a positive development, even if it means more work for administrators in the short term.

However, I think we need to distinguish between discovering and fixing bugs and deciding how urgently those fixes need to be deployed.

A bug that has existed for three years doesn't necessarily require an immediate reboot just because it has now been discovered and fixed.

My main concern is therefore not the number of kernel releases itself, but the lack of clear information about their operational urgency.

If Proxmox provided more detailed security impact assessments, CVE references and guidance on affected configurations, administrators could make much better decisions about when to schedule their next reboot.

And I agree with your approach of using AI to help manage infrastructure. That's certainly part of the solution, too.
 
  • Like
Reactions: meyergru
However, I think we need to distinguish between discovering and fixing bugs and deciding how urgently those fixes need to be deployed.

A bug that has existed for three years doesn't necessarily require an immediate reboot just because it has now been discovered and fixed.
Keep in mind that time-to-exploit has dropped dramatically. And time-to-exploit does not care whether the underlying bug was introduced three days or three years before its discovery.

According to the German Federal Office for Information Security (BSI), in its latest annual report published on November 19, 2026, the average time between vulnerability disclosure and first known exploitation fell from 53 days in the previous reporting period to just 25 days in the current one — and to a mere 8 days in Q2 2026.

Source: https://www.publicnow.com/view/769045BBAEE10ADD3994A6E068D9C7F168068FE0

Not every kernel update requires an immediate reboot, of course. But given how rapidly the window between disclosure and exploitation is shrinking, delaying security fixes may carry considerably more risk than it used to.

However, only you can assess your own attack surface and exposure, and therefore how urgently you need to react. Deciso simply cannot make that decision for you in advance.

P.S.: With proper redundancy and live migration in place, the operational impact of rolling kernel reboots can be kept relatively low anyway. Anyone operating an environment with substantial availability requirements (and exposure) should already have such mechanisms.

P.P.S.: Aren't the CVEs in the Proxmox kernel release notes?
 
Last edited:
A bug that has existed for three years doesn't necessarily require an immediate reboot just because it has now been discovered and fixed.
Well thanks to AI, exploits for the corresponding vulnerabilities are also being developed much faster. So, critical vulnerabilities should be patched as soon as possible.
If Proxmox provided more detailed security impact assessments, CVE references and guidance on affected configurations, administrators could make much better decisions about when to schedule their next reboot.
That's would be a massive amount of work if they wanted to do it properly, and would costs a lot of money. If at all it would probably have to be a paid add-on subscription.

At the end of the day, all you can really do is patch when in doubt and, yes, reboot.

And honestly, my home servers (1-liter mini PCs) are back up and running within two minutes. And yes, I realize that enterprise hardware takes longer to boot. But I'd say that even in an SMB environment, scheduling 5–10 minutes of maintenance in the evening or early the next morning should be perfectly doable if you really want to.

And if not – i.e. if even five minutes of planned downtime are already an issue – you should have a proper HA setup anyway. Then you can reboot one host at a time, even in the middle of the day, without users noticing anything.
 
Last edited:
That's would be a massive amount of work if they wanted to do it properly, and would costs a lot of money. If at all it would probably have to be a paid add-on subscription.

At the end of the day, all you can really do is patch when in doubt and, yes, reboot.

And honestly, my home servers (1-liter mini PCs) are back up and running within two minutes. And yes, I realize that enterprise hardware takes longer to boot. But I'd say that even in an SMB environment, scheduling 5–10 minutes of maintenance in the evening or early the next morning should be perfectly doable if you really want to.

And if not – i.e. if even five minutes of planned downtime are already an issue – you should have a proper HA setup anyway. Then you can reboot one host at a time, even in the middle of the day, without users noticing anything.


I agree that critical vulnerabilities should be patched as quickly as possible. I never argued otherwise.

My point is that administrators need reliable information to distinguish critical vulnerabilities from less urgent fixes. Without that information, "patch and reboot whenever in doubt" effectively becomes the default for every kernel release.

Regarding maintenance: the time a server needs to boot is not the same as the time required for a safe cluster maintenance operation.

My three-node Proxmox cluster already supports live migration, and rolling reboots are partially automated. Nevertheless, I need to consider GlusterFS consistency, cluster quorum, VM migration and verification after each reboot.

HA does not eliminate these operational considerations.

I'm also not asking Proxmox to conduct a comprehensive security audit for every release. Even better CVE references, links to upstream security advisories and clearer descriptions of affected components would be valuable.

I think we fundamentally agree on the importance of security updates. My concern is how to make informed operational decisions without treating every kernel release as equally urgent.
 
Last edited:
  • Like
Reactions: meyergru
Keep in mind that time-to-exploit has dropped dramatically. And time-to-exploit does not care whether the underlying bug was introduced three days or three years before its discovery.

According to the German Federal Office for Information Security (BSI), in its latest annual report published on November 19, 2026, the average time between vulnerability disclosure and first known exploitation fell from 53 days in the previous reporting period to just 25 days in the current one — and to a mere 8 days in Q2 2026.

Source: https://www.publicnow.com/view/769045BBAEE10ADD3994A6E068D9C7F168068FE0

Not every kernel update requires an immediate reboot, of course. But given how rapidly the window between disclosure and exploitation is shrinking, delaying security fixes may carry considerably more risk than it used to.

However, only you can assess your own attack surface and exposure, and therefore how urgently you need to react. Deciso simply cannot make that decision for you in advance.

P.S.: With proper redundancy and live migration in place, the operational impact of rolling kernel reboots can be kept relatively low anyway. Anyone operating an environment with substantial availability requirements (and exposure) should already have such mechanisms.

P.P.S.: Aren't the CVEs in the Proxmox kernel release notes?
Thanks for sharing the BSI figures. The shrinking time-to-exploit is certainly concerning, and I fully agree that critical vulnerabilities need to be addressed promptly.

I also agree that the age of a bug says nothing about how quickly it might be exploited after disclosure.

However, those figures don't mean that every kernel vulnerability carries the same risk or requires the same response time. The actual attack surface, required privileges and affected components still matter.

Regarding your question about CVEs: Proxmox does mention some CVEs in its kernel changelogs, but the lists are not comprehensive. I understand why — some kernel updates incorporate hundreds of upstream commits, many of which receive CVE assignments only later.

I'm not expecting Proxmox to assess every single vulnerability individually. But highlighting particularly critical fixes and their operational relevance would make maintenance decisions considerably easier.

My concern remains the same: finding the right balance between timely security updates and reliable operation of production infrastructure.

And yes, HA and live migration help enormously. I already use them. But they reduce the operational impact of kernel maintenance; they don't eliminate its risks.
 
  • Like
Reactions: meyergru
However, those figures don't mean that every kernel vulnerability carries the same risk or requires the same response time. The actual attack surface, required privileges and affected components still matter.
Yes, sure ... the problem is also that without a working exploit that is known to the reviewer, you cannot say how critical a fix needs to be applied. Some are very easy to exploit, some very complicated; some are just sending a few crafted packages, others are timing relevant, etc. This assessment can even change after time and if you read about it before it was escalated, you are now misinformed and still attackable. This detailed information is not available - also to keep you safe, but I don't know if that argument is still true if AI assisted or AI automated attacks are now a thing.
Another problem is the changelogs ... often problems were fixed and then the CVE released, so that you cannot effectively check if the patch was already applied or not, because a changelog of an already installed kernel is not going to change magically after the CVE has been released. That is all on top of everything you already talked about. Keeping track of CVEs and systems that are effected is already a nightmare and the time between patches as strunken so much it's VERY hard and is not fun. This is a new full time job ...

So, the cross product of everything is sooo huge that it is practically unhandlable, especially if you want to test everything automatically, yet how would we solve this? The only way that sounds feasible is to dramatically reduce the kernel source size to probably make the changes less frequent, more testable, etc. but I don't see that happening.

Wasn't there the EPSS value to get an estimate how critical a bug is with respect to its exploitability? This "just" needs to be cross related to all other information and you're good to go. Of course it's not as easy as that, but maybe a starting point.
 
Youd have to spend an enormous amount of time, tokens or money to research each and every CVE. Does it affect me? Am I vulnerable? Is there an exploit out there? If not now, is there one tomorrow?
It's faster and safer to just update and be done with it. Ill reckon this situation will go on for awhile. OpenAI found 1000 exploits in the kernel recently, idk if they are all patched already or if they are slowly walking through to upstream.
Itll take more initiatives like this to get the whole kernel audited and "fixed". Proxmox can help with this via bug reports, patches and the like. The things they are already doing for years.
 
Hi,

Don't forget that not only Linux kernels are impacted, Microsoft and Windows are also impacted.

When Microsoft patch CVE releases the patch on Tuesday, and then lots of users report issues due to that, it's the same issue.

As of today it's not just a "should I patch or not" it's "Should I want a 'stable' and vulnerable system or should I want a patched and unstable system".

And personally, with my superior, we agreed that we need to patch (server or client) whatever is the 'cost', because we know for a fact that if the system is breached because we have not patched, then the fault is on our company, and guess what insurance will not cover because you are considered 'faulty' for not having deployed the security patch, and if the client really wants to uninstall the security patch, he just has to sign a disclaimer of liability.

Welcome to the world of IT 3.0

Best regards,