Failed to initialize database

[Chuckame]

I'm getting this error when starting the comfyui container: Failed to initialize database. Please ensure you have installed the latest requirements. If the error persists, please report this as in future the database will be required: (sqlite3.OperationalError) unable to open database file

There is no issue when using webui/ollama for classical textual chat, so the gpu passthrough seems to work.

I never used comfyui, so I'm sorry if I failed on the usage of it!

Then when trying to execute a workflow:

got prompt
Warning, This is not a checkpoint file, trying to load it as a diffusion model only.
model weight dtype torch.bfloat16, manual cast: None
model_type FLOW
/comfyui-nvidia_init.bash: line 732:   175 Killed                  ${COMFY_CMDLINE_BASE} ${COMFY_CMDLINE_EXTRA}
!! ERROR: ComfyUI failed or exited with an error
!! Exiting script (ID: 33)
!! ERROR: subscript failed
!! Exiting script (ID: 1)

I'm currently executing this stack in an LXC with gpu passthrough in proxmox.

LXC versions

# nvidia-smi --version
NVIDIA-SMI version  : 580.95.05
NVML version        : 580.95
DRIVER version      : 580.95.05
CUDA Version        : 13.0

# docker --version
Docker version 28.5.0, build 887030f

# uname -a
Linux docker 6.14.8-3-bpo12-pve #1 SMP PREEMPT_DYNAMIC PMX 6.14.8-3~bpo12+1 (2025-09-12T11:29Z) x86_64 GNU/Linux

Proxmox versions

# nvidia-smi --version
NVIDIA-SMI version  : 580.95.05
NVML version        : 580.95
DRIVER version      : 580.95.05
CUDA Version        : 13.0

# uname -a
Linux pve2 6.14.8-3-bpo12-pve #1 SMP PREEMPT_DYNAMIC PMX 6.14.8-3~bpo12+1 (2025-09-12T11:29Z) x86_64 GNU/Linux

docker compose content

services:
  ollama:
    image: ollama/ollama:latest
    volumes:
      - ollama_data:/root/.ollama
    restart: always
    ports:
      - 11434:11434
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

  ollama-webui:
    image: ghcr.io/open-webui/open-webui:main
    volumes:
      - ollama_webui:/app/backend/data
    depends_on:
      - ollama
      - comfyui
    ports:
      - 8080:8080
    environment:
      OLLAMA_BASE_URL: http://ollama:11434
      RAG_EMBEDDING_ENGINE: ollama
      RAG_EMBEDDING_MODEL_AUTO_UPDATE: True

      WHISPER_LANGUAGE: fr

      ENABLE_IMAGE_GENERATION: True
      IMAGE_GENERATION_ENGINE: comfyui
      COMFYUI_BASE_URL: http://comfyui:8188
      IMAGE_GENERATION_MODEL: flux1-schnell.safetensors
      # COMFYUI_API_KEY: TBD in ui
    restart: always

  comfyui:
    image: mmartial/comfyui-nvidia-docker:ubuntu24_cuda13.0-latest
    ports:
      - 8188:8188
    volumes:
      - comfyui_data:/app/basedir
      - comfyui_run:/comfy/mnt
    restart: always
    environment:
      WANTED_UID: 0
      WANTED_GID: 0
      BASE_DIRECTORY: /app/basedir
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu, compute, utility]

  filebrowser:
    image: filebrowser/filebrowser:s6
    ports:
      - 8081:80
    restart: always
    volumes:
      - comfyui_data:/srv/comfyui_data
      - filebrowser_db:/database
      - filebrowser_config:/config
    environment:
      PUID: 0
      PGID: 0

volumes:
  ollama_data:
  ollama_webui:
  comfyui_data:
  comfyui_run:
  filebrowser_config:
  filebrowser_db:

[mmartial]

I see the database error too. So far, I have not seen any problems with running the container despite this error.

The /comfyui-nvidia_init.bash: line 732: 175 Killed means your workflow crashed Comfy. It is not related to the container.
I can not help debug it, unfortunately.
Maybe the stored logs (see the "Comfy crash" section in the README.md) might help you determine the issue.

I like the idea of adding filebrowser in the compose.yaml file ;)

[mmartial]

When I look further into your error log, that

Warning, This is not a checkpoint file, trying to load it as a diffusion model only.
model weight dtype torch.bfloat16, manual cast: None

might be why it is crashing. What model are you trying to use, and is it a valid .safetensors file?

I also looked a little bit more into the database issue, and according to the issues I am seeing on Comfy's GH, it might be related to data that needs to be in the user folder.
The strange thing being that my own installation runs fine.

comfyui-nvidia  | Context impl SQLiteImpl.
comfyui-nvidia  | Will assume non-transactional DDL.
comfyui-nvidia  | No target revision found.

[mmartial]

Looking at https://github.com/comfyanonymous/ComfyUI/blob/72c2071972d3207ed92bc20535299c5f39622818/comfy/cli_args.py#L215 it seems that the comfyui.db is in the run folder: ./run/ComfyUI/user/comfyui.db

I see it does not follow the --base-directory option and might need to be set using the --database-url method.

[mmartial]

I opened a feature request, as this appears to be an issue of a new feature not using an existing option.
comfyanonymous/ComfyUI#10264

When it is implemented, most users of the --base-directory option will no longer see the error message.

For the time being, it should be possible to use --database-url="sqlite:///basedir/user/comfyui.db" (untested) to resolve this error.

An alternate solution might be (as per the details of that CLI) to use --database-url="sqlite:///:memory:" as part of the COMFY_CMDLINE_XTRA variable.

[Chuckame]

Thanks for your time!

In the meantime, today I tried comfyui directly on my Macbook, and it appeared that it worked (able to generate image, audio, etc). Now I understand much better the basics, so I'll try again from scratch, using a standard workflow at first !

The filebrowser is awesome, but I hate the fact that you cannot execute a command to add user or change password on the running container (yes...) and there is no env vars to pre-parameterize the container (I, DON'T, WANT, AUTH!!)

Looking at

ComfyUI-Nvidia-Docker/init.bash

Line 259 in ca4007e

if [ ! -z ${COMFY_CMDLINE_XTRA+x} ]; then COMFY_CMDLINE_EXTRA="${COMFY_CMDLINE_XTRA}"; fi # support previous variable

When I set COMFY_CMDLINE_EXTRA (it feels like a typo without the E 😛) as --database-url=sqlite:///app/basedir/comfyui.db, it still doesn't find the db as the file still not exist... I'm searching how to create it 👀

For sure, the in-memory db works, but isn't persistent :/

[Chuckame]

Now I'm hitting the limits of my crappy SSD, but when it goes to the next step, it shuts down my server because of tight RAM 😆

I have 16GB RAM and 12GB VRAM, I'm going to buy 32GB RAM instead, hoping it's enough!

So it's supposed to be "fixed" with your workaround, but I cannot ensure it to you 😄

[realnikolaj]

@Chuckame Until you get more resources, I recommend you work on smaller images or downscale to like 512x512 before generating. Also look for smaller diffusion models and text_encoders.

• svdq-int4_r32-flux.1-kontext-dev (nunchaku FLUX1 kontext)
• svdq-int4_r32-flux.1-dev (nunchaku FLUX1 dev)

Those nunchaku FLUX1 models are both 6.8 GB in size

• t5xxl_fp8_e4m3fn_scaled 5.2 GB
• awq-int4-flux.1-t5xxl 3.0 GB

Text encoders are also a good place to look for savings.

Try looking into some of the server args found here ComfyUI server args.

It might be worth to look into if ComfyUI determines your VRAM size is too small and thus enables "--lowvram". I thought of this, because just now I noticed the server logs on my instance had this entry:

VRAM > 14GiB, disabling CPU offload

"--lowvram" would force models to get aggresively offloaded to CPU system memory, but with the model suggestions above and some tweaking on the args/image input sizes, you might be fine.
The docs (link above) even suggests this as well ..

vram_group.add_argument("--normalvram", action="store_true", help="Used to force normal vram use if lowvram gets automatically enabled.")

Try e.g., "--normalvram --reserve-vram 0.5" in COMFY_CMDLINE_EXTRA_ARGS and see if this helps with the crashes.

Please note that you'd want as much to stay in GPU but if your laptop is also running windows the above suggestion is still pushing it too far and you should go for GGUF models.

[ImagineerNL]

Another workaround is to recreate the ComfyUI/user directory again after the initial run that moves the userfolder to /basedir/user
This also removes the error and since then i have an empty (0 bytes) comfyui.db file in that directory.
I have no idea where the actual comfyui.db file is now stored, as it isn't in the /basedir/user directory.

I see you also created a FR at the Comfy github, i added some related existing issues to it for extra traction.