Allow admin to configure required votes for quorum OR out of quorum behaviour.

[tomtom13]

Hi,
It would be nice to allow admins of a cluster to define either:
- a minimum votes required for a quorum (yes, admins might be aware of pitfalls and are prepared to deal with consequences)
OR
- define behaviour when quorum is not achieved - current "stop everything and just sit there" is a bit suboptimal for certain cluster setups (example: a cluster with compile and render nodes which are dynamically powered up and down for energy preservation)

 

[leesteken]

It would be nice to allow admins of a cluster to define either:
- a minimum votes required for a quorum (yes, admins might be aware of pitfalls and are prepared to deal with consequences)

I believe this is already possible, see the manual for an example on setting the expected/required number of votes.

 

[tomtom13]

I believe this is already possible, see the manual for an example on setting the expected/required number of votes.

Forgot to add:
“through GUI”
Also when I tried manually setting it in config corosync would often ignore / remove this part. At the moment, only reliable was is to assign large amount of votes to one node that is always on, but when this one goes for down for updates, the cluster goes down.

 

[bbgeek17]

yes, admins might be aware of pitfalls and are prepared to deal with consequences)

There are weekly posts in the forum from admins not aware of pitfalls or consequences of the "design" they put together that "worked" and then broke.

define behaviour when quorum is not achieved - current "stop everything and just sit there"

If official PVE tools would allow you to setup weird one-off configs, that would mean that PVE team must be testing these one-off configs.

Forgot to add:
“through GUI”

Some things should not nor will ever be in GUI.
In general, if its in PVE CLI/API/GUI - its supported and tested. Is there a demand for such functionality from the users who are willing to pay (subscription or otherwise) for such functionality? If not, then how do you see these "features" being funded and prioritized for development and testing?
Besides, forum is not the place to submit feature requests : https://forum.proxmox.com/threads/where-to-post-feature-requests.46317/


If you want a stable cluster, then you need to put the cluster vote/witness as much outside of daily use as possible:
- minimal required stable hardware
- no VMs running there
- no hyper-converged storage
- reliable network

---

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

 

[tomtom13]

There are weekly posts in the forum from admins not aware of pitfalls or consequences of the "design" they put together that "worked" and then broke.

Better and continuously maintained documentation.

If official PVE tools would allow you to setup weird one-off configs, that would mean that PVE team must be testing these one-off configs.

Continuous automated testing. Also you make it sound like somebody is suggesting implementing orbital guidance.

Some things should not nor will ever be in GUI.
In general, if its in PVE CLI/API/GUI - its supported and tested.

Is that's why vm vs ct shutdown varies greatly in reliability in web gui vs cli ? is that why some have scripts walking the nodes looking for orphaned lock files so at least the migrations (with shutdown) don't blow up ?

Is there a demand for such functionality from the users who are willing to pay (subscription or otherwise) for such functionality? If not, then how do you see these "features" being funded and prioritized for development and testing?

There was no demand for finger triggered touch screen phone - just a stylus + keyboard from business users, yet we're all using iphone clones now. Customers evaluate you on your features and stability, and if feature set doesn't comply with your needs than they simply go somewhere else rather than coming in asking for that feature. So please don't shoot down simple life improvements suggestion, unless you work for proxmox and are in decision making chain.

Besides, forum is not the place to submit feature requests : https://forum.proxmox.com/threads/where-to-post-feature-requests.46317/

Thank you very much for pointing this one out. I've followed the https://forum.proxmox.com/tags/suggestion/ path but obviously I was in error.

If you want a stable cluster, then you need to put the cluster vote/witness as much outside of daily use as possible:
- minimal required stable hardware
- no VMs running there
- no hyper-converged storage
- reliable network

And as I've mentioned - if those are assigned extreme vote count (due to minimum vote count continuously being disregarded) and those go down due to for example kernel update, whole cluster halts.