Search results

Hi! How would I go about moving encrypted OSD's between nodes? I've moved some around and not only is ceph not detecting the drives it's not even showing the new hosts (they all have ceph installed) in the osd tree.

What does proxmox have to do with this? It's ceph. Configure your cluster correctly and it will be fine.

Based on the guide above its enabled via pool setting, which clear every reboot. You should be able to re-enable it by running those settings again.

Can you add VDO integration as an option when creating an OSD? Thanks!

Is it possible to use WDT with ceph and if so what kind of performance improvement is there (if it's been benchmarked)? https://github.com/facebook/wdt

TIL rebooting the node reverts pool settings to default. How do I make ceph save pool settings short of setting up an init script? root@cl-01:~# rados bench -p primary_volatile 10 write --no-cleanup hints = 1 Maintaining 16 concurrent writes of 4194304 bytes to objects of size 4194304 for up to...

Does your build also fix the monitor ram issue with RDMA? IIRC it hits 100% ram and crashes after a day or two. Thanks!

So is the RDMA patch included with your build of ceph @wolfgang or would ceph have to be manually compiled with it? Got any notes for that? Or will IPOIB work without that on ceph.

Will they both support Infiniband (IPOIB) without VPI?

Incidentally the reason the dedup tool is in testing is because it fails to build in rpm packages otherwise. Iirc it's stable otherwise.

Using the instructions in the source code I was able to get dedup to work correctly, so I'm happy about that! root@cl-01:~# rados -p dedup_base put foo ./test_lol root@cl-01:~# rados -p dedup_base put foo ./test_lol2 root@cl-01:~# rados ls -p dedup_chunk bar-chunk foo-chunk...

I have reviewed the source code of 14.2.4. Deduplication is supported. The ceph-dedup-tool is in 14.2.4, but has to be built with debug/testing flags. Well, it looks supported to me anyway. Lots of stuff in there regarding deduplication - and this is a production build, not developer. Edit...

iirc the dedup tool is just for finding the optimal offsets and fixing the reference count. The actual commands for enabling deduplication worked for me so the base functionality is there.

It's a turing 2000 series nvidia GPU. Works fine on Windows. XML is more or less identical to this, without qxl/spice. args: -machine q35,kernel_irqchip=on balloon: 0 bios: ovmf boot: dcn bootdisk: sata3 cores: 10 cpu: host efidisk0: local:102/vm-144442-disk-0.raw,size=128K hostpci0...

Using Q35 I am able to pass through my GPU to Linux successfully. And native games run flawlessly. I am however running into issues with wine/dxvk games. The GPU on some/most wine games with DXVK stays under 30% with low framerates. Not too sure why - does not happen if running on bare hardware.

Can we get a guide made for deduplication on 14.2.4? Following the examples here I was able to get a basic setup going, which wasn't possible on 14.2.2. https://docs.ceph.com/docs/master/dev/deduplication/ Thing is, I don't really know what they want me to put in for offsets. And I believe...

It refused to delete itself in the OSD list on proxmox. It's still "loaded" on the host node even after reboot. I've also wiped/zapped the entire disk with gdisk and restarted. Can't remove the OSD from /var/lib/ceph/osd because it's active. ...this is not the first time I've had this issue...

What is it? Because when I manually define XFS with AES256 in the ceph.conf and recreate the cluster my rand performance drops from 1200MB/s to ~500MB/s compared to the native encryption proxmox comes with.

Yeah, it actually worked itself out once I added a third IP block.

I can't configure the monitors to use the private network. Ceph is configured private on both pub/private in the ceph.conf. But if the monitors aren't on the public network (10.0.0.0/12) ceph no longer responds. Ceph traffic does flow over the private network (172.16.2.0/12) so not sure why...