[PVE-User] Need some advice on pve writeback caching

[Alwin Antreich]

On Fri, Apr 03, 2020 at 12:54:36PM +0200, Rainer Krienke wrote:
> Hello Alwin,
> 
> thanks for you very much answer.
> 
> Regarding LVM:
> I initially thought the cache size is 25MB for each RBD device. So a LVM
> based on in my case 4 RBD devices could loose 100MB. This would have
> been a difference. But after reading the docu again it seems that this
> cache is for all rbds of one ceph client (pve host).  Then you are right
> and it does not  make a difference.
You can activate io-thread, to get a thread per disk. This will gain
performance, but as you mentioned it will also raise the total rbd cache
for the VM.

In any case, the VM should be gracefully shutdown prior running out of
power. A sudden power loss, will result in any number of issues. Also
for the data loss, memory usage of the VM itself is usually bigger than
the rbd cache.

--
Cheers,
Alwin