Reason PVE host keep write data to disk?

[alexc]

I have single PVE server (6.4, community repo) and I noticed It has constant iodelay > 0 even when host is almost idle. The host is very low loaded, so not expected to see that.

There is single m2 NVMe disk on board (Samsung 970 EVO Plus) installed mostly as a test disk, and couple of Intel 545 SSD SATA disks which are pit in ZFS mirror.

Can’t figure out the only thing: what can produce constant iodelay to disks?

By the way, both Intel SSDs has 27% of wearout even that they are online for maybe 3-4 months. So I suspect constant write steam is definitely there. But the source of it is unknown.

Please advice, how can I find the source of that activity!

 

[oguz]

hi,

Please advice, how can I find the source of that activity!

you can check the processes with iotop and investigate further from there

 

[alexc]

hi,


you can check the processes with iotop and investigate further from there

I did that, no clear picture. VMs are very low intense on I/O, and the only thing I see is zfs-related processes. But I suspect 20+% of wearout for 2-3 months is high a bit for zfs scrub mostly.