Scrub incredibly slow with 13.0-RC3 (as well as RC1 & 2)

[Mathieu Chouquet-Stringer]

On Mon, Mar 22, 2021 at 04:39:27PM +0100, Mathieu Chouquet-Stringer wrote:
> 	Hello Martin,
> 
> Thanks for taking the time to reply to me.
> 
> On Mon, Mar 22, 2021 at 03:15:17PM +0000, Martin Simmons wrote:
> > You could try:
> > 
> > procstat -kk 0
> > 
> > to get an overview of what the kernel threads are doing.

Exploring some more with top -SH, it seems that's my thread:
   31 root         -8    -     0B  3984K CPU7     7  38:14  99.92% zfskern{dsl_scan_iss_0}

That's the only busy thing on this OS and I know it's the one I see in
htop because I see it ping-ponging on different CPUs.

Seeing dsl_scan_iss led me to there:
https://github.com/openzfs/zfs/blob/64e0fe14fff465604952d576051ca80c2f1be898/module/zfs/dsl_scan.c

And that led me to reading about vfs.zfs.scan_legacy:
https://openzfs.github.io/openzfs-docs/Performance%20and%20Tuning/Module%20Parameters.html?highlight=scan_mem_lim_fact#zfs-scan-legacy

I killed the current scrub, put vfs.zfs.scan_legacy=1 in sysctl.conf,
rebooted and restarted it, it's already telling me it'll finish quicker
than the one which was running and wich had already done 10%. It's been
running for 22 minutes now.

  scan: scrub in progress since Tue Mar 23 01:03:09 2021
        63.0G scanned at 87.5M/s, 63.0G issued at 87.5M/s, 24.5T total
        0B repaired, 0.25% done, 3 days 09:14:15 to go

I'll let it run overnight and I'll check back tomorrow: I mean you
can't trust the progress bar...

-- 
Mathieu Chouquet-Stringer
            The sun itself sees not till heaven clears.
	             -- William Shakespeare --