Hello,
I have installed Immich with the ROCM machine learning image. The app has been configured to use host paths located on SSD backed storage (tank_fast) for Data, Machine Learning Cache, and Postgres.
After images have been imported into immich, it seems that any ML related task that runs causes major activity on my raid-z2 pool (tank), which should not even be visible to immich. Shortly after this activity begins the following alert is triggered:
Quota exceeded on dataset tank/.system/cores. Used 99.32% (1017.06 MiB of 1 GiB).
If I change the ML image to ‘Default Machine Learning Image’ and disable GPU passthrough (I have not tried these changes independently), the disk activity on tank is non-existent.
Any advice on how to resolve the issue with the above quota issue, and disk thrashing my pool is experiencing is greatly appreciated.