Need help with ZFS Cache, it is crashing my TrueNAS

I am more and more incline to believe this is the issue. Combination of consumer level mobo with ECC RAM.

I don’t think so. Following this thread, the pattern that makes the most sense would implicate the “shared RAM” with the integrated GPU.

With Plex running, if the crashes stop after doing this, then it’s fair to say that the iGPU was the culprit the entire time.

1 Like

Oki, I will try to disable iGPU this weekend and try again

You can refuse to belive it but fact is, your mobo has no official support for ECC RAM: ASRock > B650 Steel Legend WiFi

I had the ASRock > X570M Pro4 mobo which had official support for ECC RAM and still had issues (probably different chip on RAM module than what was supported).

Uhhh, what you talking about? Just because it is not on List of RAM doesn’t mean its not supporting.

“Support” as used there is doing a bit of work.
The “memory QVL” is the manufacturer’s way of saying “we have tested this memory kit and it works with our motherboard”.

Memory not on the list can still work, it just hasn’t been verified by the manufacturer.
That’s it. Nothing more, nothing less. In order to sell more memory, some memory makers do their own validation with kits and boards. If you trust them you can use this as an extra indicator of what works with what.

There’s also the part that ASRock makes no mention of ECC in the documentation related to that specific motherboard. I am not sure why they choose not to market it, but I suspect it’s due to them not spending resources validating that functionality. Users have tested and found that it can work, but again, the manufacturer is not giving any guarantees, so results are somewhat undefined.

All of this may have something to do with your issues, or absolutely nothing at all.
I’m inclined to agree with the GPU memory theory.


:thinking: seems like it’s “supported” although that may not mean “fully utilizes”

Right you are, surmise I messed up my ctrl-f search…

At this point I just want this dude to get home and get hands on it. This thread is like Days of Our Lives.

I’m gonna hedge my bet with Winnie. Since it survived with Plex down when it normally would have died, there’s some causal relationship between the iGPU and ram usage as we see the ARC balloon up during Plex’s maintenance phase. Borrowing ram during a ram heavy event without telling anyone is bad.

1 Like

3 Likes

Hello All, I am finally home and expecting that my Truenas would be in limbo status. But somehow it survived last night’s maintenance.
Now, I did update Plex app update that was release last night. I really do not know if this was bug on last app version.


I did find disable igpu but not sure if I should give another maintenance to see if last night was fluke or not.

Decide to give one more day without zfs arc size cap.
I will update tomorrow!

I’ve seen similar issues with known-good ram, that was not on the QVL, when memory pressure suddenly ramped, the server crashed.

So yes, the fact its not on the QVL can be quite relevant.

1 Like

Cool. You have one day to get your non ecc ram in there, assuming you’re just there for the weekend. Flaky hardware is so annoying especially if you aren’t there to babysit it every day. If it kept acting bad, you would have addressed it. Now it got lucky and you feel less pressure to fix it.

I guess we will all know next week if it healed itself through that update.

It survived another day but although I deleted script to zfc size cap, it still maxed out at what I setup last time. I am going to manually remove zfc size then run plex maintenance again

1 Like

What are you testing?

The iGPU and shared RAM has been disabled, but everything else is back to the defaults?

Only thing changed now is Plex update that was released two days ago.
Everything else is back to default. iGPU is still enabled. It is middle of 4 hour plex maintenance and still it is up and running without crashing.
So, Plex had some issue with process zfs arc??

Maintenance is running for about 2 hours and its literally stressing testing my cpu and ram. So far stable.

Wondering if there was a bug discovered where if you pass all ‘non-nvidia’ gpus, it decides at random which to use. I think we were on the right track that igpu memory allocation was causing OOM when arc was full & plex was doing something and this something maybe was using igpu instead of your arc card. Possibly… lol or it was just a bad version from Plex & the update plugged a memory leak.

Guess we’ll know if it keeps being stable.

3 Likes

I disabled iGPU and kept everything same. I probably won’t be using iGPU anyways. I no longer see a lot of movement in graph in zfs report after the update, so it was definitely Plex related issue.
I will monitor for next few days and if it happens again, I will update you guys.
Thank you so much for everyone who gave their opinion on this issue!

3 Likes

Oh joy! This means @afrosheen and I won’t lose our homes.

4 Likes