I noted this item in the 7.0.0 changelog, have never thought about it nor looked at it previously.  Fortunately/unfortunately I'm now curious so I guess my first question is...What is classed as "excessive"?

 

I decided to manually capture the below to create a baseline of sorts and I also did a little forum surfing.  So my second question... Are the results that I've captured today concerning?

 

For the 1 hour up to 13:07 I ran the below...

# inotifywatch -r -t 3600 /boot
Establishing watches...
Finished establishing watches, now collecting statistics.
total  access  close_nowrite  open  filename
84684  28230   28227          28227  /boot/config/shares/
60702  22978   18870          18854  /boot/config/
43866  35444   4211           4211  /boot/
37377  12459   12459          12459  /boot/config/plugins/dynamix/
4796   2398    1199           1199  /boot/config/plugins/dynamix/notifications/agents/
4796   2398    1199           1199  /boot/config/plugins/dynamix/notifications/
2464   1120    672            672   /boot/config/plugins/dockerMan/templates/limetech/
2464   1568    448            448   /boot/config/plugins/dockerMan/
2240   1344    448            448   /boot/config/plugins/dockerMan/templates/
1659   553     553            553   /boot/config/plugins/ipmi/
1344   896     224            224   /boot/config/plugins/dockerMan/templates-user/
342    114     114            114   /boot/config/plugins/gpustat/
100    88      6              6     /boot/config/plugins/
27     9       9              9     /boot/config/plugins/community.applications/
18     6       6              6     /boot/config/plugins/zfs.master/
15     5       5              5     /boot/config/plugins/compose.manager/
4      2       1              1     /boot/config/wireguard/

 

As of typing my dashboard reports the below, this is largely typical from my observations...

The only change I've implemented recently other than the upgrade to 7 is to increase my ZFS Arc Max limit and to add a L2 Arc to one of my pools.  I've not implemented the fix mentioned in the change log, from my read it sound like a worthwhile activity as 500MB RAM is nothing really, I'm just interested to get a proper view on the above.

 

Thanks,

K

Since 6.12 /usr and /lib are mounted from the squashfs on the flash drive, so any reads from its contents means reading from the flash drive, that's normal. The posted method will load /usr in RAM and reduce these, /lib is still read from the flash drive. 

Edited January 24, 20251 yr by Kilrah

So addional question if I may.  I realised there is something else I am/was messing with in addition to the ZFS Max Arc limit.  I'd temporarily removed a boot time modification I'd previously made to nginx's rate limiting so this was back to stock during the time period I was seeing the high flash usage.

 

I forgot about this when I added the /boot/config/fastusr file and rebooted about 30 mins ago but I'm now seeing a much improved as in reduced read amount on the flash storage.  I know I've muddied the water with the 2 changes at the same time but could the rate limit change be involved here, the below is in my /boot/config/go file...

 

/usr/bin/sed -i "s;zone=authlimit:1m rate=30r/m;zone=authlimit:1m rate=60r/s;g" /etc/nginx/nginx.conf

 

I don't think so. Fastusr will definitely make a difference though.

Concluding this thread now and answering a couple of my own questions hoping others might find it useful.

1. What is classed as excessive?  In my testing anything over 1 read per minute of the flash drive would be excessive.
2. Where my results above a concern?  Yes, as they exceeded the above and at worst were over 200 reads per minute.

The resolution was to apply the /boot/config/fastusr config which @Kilrah kindly confirmed.

 

I did retest my scenarios today by again removing the nginx rate limit alteration and re-ran with and without the fastusr fix.  I my usecase I pushed the ZFS Arc to the max, kept the server busy with various activities and managed to replicate the issue by then browsing the unraid UI which by then slow which then started to notch up the read counter on the flash drive.

 

16 hours ago, KillerK said:

What is classed as excessive?  In my testing anything over 1 read per minute of the flash drive would be excessive.

Question is why. It's writes that need to be limited to the flash drive, reads don't matter.

I only started looking at flash usage because of system performance or more specifically the unraid UI being laggy. It's an assumption but I believe this was triggered as I had overridden unraids default zfs max arc limit to more fully utilise the available ram. Then once the system was being loaded that's when I noted the increasing read counts.

 

The way the fastusr note in the changelog reads about flash usage its not a simple if x do y and infers their is an amount read which is acceptable but doesn't clarify this. Hence my question.