Search results

hmm. this is where a feature request for "last state" is probably your next step :)

yes. For your test vms (and other non critical) just set the HA request state to ignored. you can still live migrate them just fine.

You realize that HA request state of started means, you know, that it should be started. If you dont want that, set a different request state.

This implies that the operator has both the skill, experience, and the wherewithal (not having a dozen other responsibilities) to understand the docs and apply them. Ceph is not attractive at the low end precisely because it requires engineer level admin, which is neither common or cost...

If by redundancy you mean disk fault tolerance, the higher the value after "raidz" the higher the fault tolerance. In practice, raidz2+ (never use single parity raidz unless prepared to lose the pool at any time) performance= striped mirrors. full stop. if you wish to sacrifice some performance...

Good plan. I hope you understand that this experiment will not ever be worth any money you put into it. You already have a working solution- and if the "politics" of the money prevent you from putting together a sane configuration you're just throwing money away for no good reason.

I'm sure there would be readers grateful for any such information. any reason you're not just posting it?

@RodolfoRibeiro If you want more direct assistance, post the content of (from both hosts) lsblk multipath -ll -v2 if you have system logs available from the point in time when your vm became corrupted, would be good to look at what happened.

For the generations of hardware where iscsi and SAS were offered as available SKUs there was no meaningful performance difference- 16G FC simply had more headroom to fill cache. When 25GB iscsi product started shipping, THOSE were faster (even for fc16.) it is theoretically be SASg4 host...

K=6,M=2 results in 6 data strips per 8 total. 6/8=0.75 in replication you have 1 data strips per 3 total. 1/3=0.33 its not exactly the "same" availability because survivability in a replication group is much higher; you need one living osd per pg to recover, whereas with an EC 6+2 you need 6...

"lower" and "higher" are subjective. Ceph achieves HA using raw capacity. suit yourself. this is not a recommended deployment. You are far better served by just having two SEPERATE VMs each serving all those functions without any ceph at all- you'll have better fault tolerance, better...

The number of OSDs isn't relevant to a pool as long as it is larger then the minimum required by the crush rule. For example, If you have an EC profile of K=8,N=2 rule, you need a minimum of 10 OSDs DISTRIBUTED ACROSS 10 NODES. so 1 OSD per node. you can have more OSDs and more nodes and data...

from understanding failure domains. damn @UdoB beat me to the punch. I wont "professor" you on this. You can either read and understand, or deploy your preconcieved notions and learn on your flesh and blood. I would also note that if your expectations that this 4 node ceph+EC+HDDs will be faster...

ahhhh that makes so much sense... I so rarely set up windows guests so I didnt even think of that.

Ok. lets touch on this. From my perspective, there are two types of storage (there are more but in scope.) There is payload (think OS and application) storage and bulk storage. Bulk storage can most efficiently be served by a single device such as your 36 bay with slow spinning drives...

Caching occurs in multiple layers of presentation. By the time a virtual disk is presented to a guest, the multiple caching layers can conflict and actually SLOW the guest storage performance. see https://forum.proxmox.com/threads/disk-cache-wiki-documentation.125775/

you dont need pci passthrough for lxc- just would need to install the proper nvidia driver based on hardware and kernel deployed. You are better off creating an installation script, especially if you intend on having multiple nodes with GPUs. FYI, that 4x node solution is VERY old and has...

I think you need to carefully consider what your end goal is. PCIe passthrough is not a good citizen in a PVE cluster, since VMs with PCIe pins not only cannot move anywhere, but also liable to hang the host. if you MUST use PCIe passthrough, consider leaving that node outside the cluster. I...

in the many years I've been using PVE, I havent had much call for using Windows guests, and when I did it was usually Windows 2016 (and older before) and had reasonably good results. In the last few weeks, I had need of a Windows guest for a specific purpose and I figured its a good time to set...

In a cluster you dont need or even want to backup a host. everything important lives in /etc/pve which exists on all nodes. If you DID back up a host(s), you'd open the possibility of restoring a node that has been removed from the cluster and causing untold damage when turning it on. This is...