Server freezes up when exceeding RAM useage

[sunny_raven]

Hi everyone! 

I've been noticing an issue with my HexOS server whereby the system will freeze up. The system stops showing up in the Deck dashboard, and I can't access any local apps. However, the server hardware is still running. The server never recovers, so I have to reset it via the button on the case.

From what I've been able to deduce, this seems to happen when the server comes close to or reaches the RAM usage maximum.

I noticed that TrueNAS uses a big chunk of RAM for cache, which is understandable. However, when running Ollama (when models spill over to RAM) or when running other RAM heavy tasks (like metadata refresh in Immich) it seems the cache usage doesn't go down to free up extra space and the system runs into the hardware limits.  

Now, onto the question -- is there a way to set a hard limit on the TrueNAS cache RAM usage or allow the server to manage RAM a bit more intelligently?

If there is any way to collect any necessary logs, please let me know how I may be able to do that and let me know if I can share any other extra info to troubleshoot this.

 

[PsychoWards]

Hey, can you please post your complete HW specs ? 

[sunny_raven]

Absolutely!

AMD Ryzen 9 7950X with 128 GB of RAM

5x4TB HDDs for NAS data storage

3x2TB SSDs for apps and related app data

 

I noticed that very often the RAM  cache usage sits at 80-90% with very little overhead for the apps. Last time the server froze-up was when I started metadata refresh in Immich. I hadn't monitored the server during that time, but almost certain that it froze due to the app reaching the total RAM capacity.

I also read up on TrueNAS caching and it sounded like there is no way to set up an upper limit for how much it could use.

 

PS

Also, I realised that before I signed in I hadn't seen all of the possible threads to ask the question so this thread may not be the most appropriate to ask. For any mods, please feel free to move this to the appropriate section or delete this/mark as solved and I can re-post this accordingly.