Search results

You need to benchmark this. When disaster strikes then Ceph will start recovery (refilling the recovered node) and may max out bandwidth.

The symlink is needed (and automatically created) by Ceph itself, since all tooling is looking for it. The distutils is a missing dependency, only recently introduced by upstream.

So, do you have something plugged in or not?

Just to clarify, these settings can be made in the ceph.conf (see the first part of the article). Hopefully then they can start. The probably the RTC is not up to the current time (hwclock). I suppose, since only the OSDs are affected, you could just destroy and re-create them. You can start...

You can try lm-sensors to start with measuring. Run something like atop and look at linux perf.

Sorry, I corrected my sentence above. OFC, the clients need to be able to connect to 8007 (default) or a port of choosing. The remote sync job will also pull as well. https://pbs.proxmox.com/docs/backup-client.html?highlight=port As long as the PBS doesn't remove the backups, eg...

While we didn't observe any big impact on our production and test clusters. They are less then 50x OSDs. ;) But I can see your point. The description might make it to easy to just fire and forget. I updated the upgrade guide with some more cautious wording.

Tape support is on our list. But don't ask me about an ETA. ;) If an attacker can access the PBS server, and is able to encrypt the data on the filesystem. Then the PBS may not recognize the junks at all. Or if that's still possible the remote will only sync the delta and not all junks. If the...

It's not only the NVMe's that need IO, with all the VM/CT, networking and other services/hardware the threads will be used up quickly. It will be quite a trail and error till you come close to the optimum. Probably the memory bandwidth will be the deciding factor, first. And then, in regards to...

What do you mean by that?

Bei default the snapshot on the GUI includes the memory of the VM. That can take a while to be written to the storage.

Hm... that is a very subjective feeling. Do you have any monitoring that measures the CPU or other temperatures? That a lot of time and many things can have changed. Starting with more load from the Guests or the CPU thermal compound is not as effective anymore.

Thanks for the feedback. I changed the autoscale section to be more explicit about the implication.

No that I know of. And it greatly depends on how much metadata will be accumulated. But the special device will spillover to the other zvol in the pool if full.

Slow ops are not a good sign. How is the hardware setup? The OSDs seem not to come up, because it seems the time out on reading from the disk. https://ceph.io/planet/dealing-with-some-osd-timeouts/ I don't know if that was just around the upgrade, but do you have a local NTP where all the...

Well, then it can't work. Either you can use a split DNS (local is different from public) or set the domin/ip in the hosts file. https://pve.proxmox.com/pve-docs/chapter-pct.html#_guest_operating_system_configuration

Is your container using the public IP?

I can't say. But it is not there on a default installation.

PIng may use a different order to resolve domain names. Try dig/nslookup. For the curl issue, you will need to look at your WP (or whatever) to see what it tries to do.

I suppose you are running pillbox? https://hhs.github.io/pillbox/ And it could be the User-Mode Instruction Prevention that is active here. https://lwn.net/Articles/705877/