Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Alex Atkin UK]

Well I think I figured out why my RRD was broken:

Code: [Select]

-rw-r--r--  1 root   wheel  147848 Dec 11  2023 /var/db/rrd/?-quality.rrd
-rw-r--r--  1 root   wheel  147848 Nov 26  2023 /var/db/rrd/??-quality.rrd
-rw-r--r--  1 root   wheel  147848 Nov 25  2023 /var/db/rrd/?????9-quality.rrd
-rw-r--r--  1 root   wheel  147848 Nov 14  2023 /var/db/rrd/?-quality.rrd
-rw-r--r--  1 root   wheel  147848 Oct 22  2023 /var/db/rrd/?-quality.rrd
-rw-r--r--  1 root   wheel  147848 Oct 29  2023 /var/db/rrd/Ѓ?6?!?^?9?!???-quality.rrd
I guess something went very wrong when migrating the configuration over at some point, so a clean slate it is.

One thing, could your modification run into a problem attempting to read the file before its been fully written?
I run into that a lot when using temporary files to pull stats into my server home page.  Though its kinda rare on the same machine, usually occurs when writing to a shared folder which obviously is slower.

Usual solution to that is to write to one file, then at the last moment rename that file to the actual file to be read.

[Chrysalis]

The file will always exist, I suppose at worst it might read the previous value but the RRD poll runs later.  I dont think its an issue I have already been running it, but if you want, you could modify the cron to add a rename and then I add it to the wiki.  The direct sysctl command runs in a split second.

Also I forgot to mention you do need to disable graphing and enable it again for the patched RRD script to take effect (no need to wipe just turn off then on), a reboot will also work.  I will edit the wiki in a bit for that.

Also it seems it auto populates the graph based on the results of the poll, I think I may add a further patch to add NVME drive temperature to the graph, I would just need to append it to the text file in the same format and the RRD generation should populate it automatically I think.  So keep an eye out for this here.

[Alex Atkin UK]

The file always exists (maybe not on first run after boot?), but its possible to read the file when its half-written if its timed exactly right.  By doing the rename it prevents that possibility as the new file is always complete before the filesystem is updated to link the old filename to the new file.

I've been hit by this issue before with all the stats I monitor on my server.  Though it happened a lot more often there due to polling the stats much faster than once a minute.

I wiped all the RRD files and indeed had to reboot to get them to generate again, as I can't for the life of me find where you turn it off and on again.

[Chrysalis]

On the graph page click advanced, then a 'disable graphing' button appears, that will turn it off, and when its off it becomes a 'enable graphing' button to turn it back on again.

[Alex Atkin UK]

Hmm, I would never have thought that was what enabled/disabled the internal data collection.  Its so inconsistent with how the rest of LuCI works.

[Chrysalis]

It is in an odd place hidden behind advanced, yep.

[Chrysalis]

I will update this for 2.8.0 over the coming weeks and will also add some more advanced patches, including the option to toggle IPv6 support in Unbound for those who prefer to force single stack either for troubleshooting or operational purposes.

These more advanced patches have multiple parts on my system, but will try to merge to one patch file for the wiki.  Some of them actually add their own extra configuration variables stored in the pfsense xml for additional functionality.

I will also be adding a serve expired/serve stale toggle, I will try to submit this one to pfSense directly, but in the mean time it will be available as a custom patch.  With the help of another kitz community member I did get the serve expired feature added to pfSense officially some years ago, however the next time pfSense does a Unbound update, it will be to a build which changes the default behaviour of serve expired to serve stale.  The patch will add an option in the GUI to toggle between serve expired and serve stale (serve stale is really just a reliability option of using a expired cache entry if a DNS lookup fails, it will offer no performance benefit, whilst serve expired is a huge performance win).

[Chrysalis]

pfSense 2.8.1 is out now with the new serve expired defaults, meaning the serve expire toggle behaviour on it has different behaviour (if not tuned), I will post a patch here later to allow the toggle to choose between the two behaviours.
https://nlnetlabs.nl/news/2025/Apr/24/unbound-1.23.0-released/
https://unbound.docs.nlnetlabs.nl/en/latest/topics/core/serve-stale.html (unbound 2.3.0 is the new version in pfSense 2.8.1)

[Chrysalis]

Patch has been added, I have also updated the main page for updates for 2.8.x on the other patches, the 2.7.0 page is now a linked archive.

I have other patches to add, which I will over the coming week.

https://wiki.kitz.co.uk/index.php/GranularServeExpired

[kitz]

Thank you.