[SOLVED] Restore from PBS1.0-5 to PVE7.0-11 ground to halt very slow

[JamesT]

Hello. I rebuilt a standalone node with PVE7 and did latest updates. This node runs 5 VMs. I restore 4/5 (small Debian VMs - 2GB-5GB used of 32GB volsize) which each restored in <1 minute. I went to do the big windows fileserver ~3TB and it started off pretty fast but then ground to a halt overnight. It has been running for 17 hours now. There are no other tasks running on the PBS server (PBS v1.0-5). None of the VMs on this node have been started yet and the server is doing nothing else.
Memory, CPU, storage load on the target node are all very free (CPU 99% idle, 50% memory used)

Here is load on the target node:

Here is hourly load on the PBS server: note the sustained 200 WRITE IOPS - why is writing happening when only a READ task is running?


Wondering if it's safe to reboot the PBS server during a restore? Will the restore resume once its back up? Or can I just start another restore, will it see the disks already exist and utilize already copied data, or will it do the entire thing from scratch?

Thanks in advance.

Update: some additional info in case it helps.
1 day load on PBS server:

This is the task status window on the target node:


The realtime stats on the PBS server show a slightly different picture to those on the dashboard:

 

[dcsapak]

do you use zfs on the pbs server? try turning on 'relatime' for the zfs dataset. this may be the reason why you get write iops on read operations (updating the atime)

 

[JamesT]

Yes, I use zfs for the pbs server. After rebooting PBS server and doing apt dist-upgrade I tried doing the restore from the target node and it completed OK over the weekend.
Per your suggestion I've turned on relatime for the pool root@PBSNODE:~# zfs set relatime=on backup-pool. If I get time I'll try another restore and see how it goes.