Persistent storage and S3

[euh2]

I run Open-WebUI from this Helm Chart with S3 configured for storage. On every restart of the pods, there are several files pulled to the pod. Startup time is often more than 20s. Can those files be persisted somewhere?

...
v0.6.13 - building the best AI user interface.

https://github.com/open-webui/open-webui

data_config.json: 100% 39.3k/39.3k [00:00<00:00, 34.6MB/s]
.gitattributes: 100% 1.23k/1.23k [00:00<00:00, 6.54MB/s]
onnx/model_O1.onnx: 100% 90.4M/90.4M [00:08<00:00, 10.6MB/s]
onnx/model_O3.onnx: 100% 90.3M/90.3M [00:08<00:00, 10.6MB/s]
onnx/model_quint8_avx2.onnx: 100% 23.0M/23.0M [00:03<00:00, 6.31MB/s]
openvino_model.xml: 100% 211k/211k [00:00<00:00, 1.55MB/s]
onnx/model_O4.onnx: 100% 45.2M/45.2M [00:12<00:00, 3.53MB/s]
openvino_model_qint8_quantized.xml: 100% 368k/368k [00:00<00:00, 1.80MB/s]
openvino/openvino_model.bin: 100% 90.3M/90.3M [00:05<00:00, 17.2MB/s]
openvino/openvino_model_qint8_quantized.(…): 100% 22.9M/22.9M [00:06<00:00, 3.76MB/s]
rust_model.ot: 100% 90.9M/90.9M [00:06<00:00, 14.8MB/s]
onnx/model_qint8_arm64.onnx: 100% 23.0M/23.0M [00:20<00:00, 1.11MB/s]MB/s]
train_script.py: 100% 13.2k/13.2k [00:00<00:00, 20.2MB/s]2, 13.3MB/s]
onnx/model_O2.onnx: 100% 90.3M/90.3M [00:22<00:00, 4.03MB/s]20.1MB/s], ?B/s]
pytorch_model.bin: 100% 90.9M/90.9M [00:09<00:00, 9.93MB/s] 1.11MB/s]
tf_model.h5: 100% 91.0M/91.0M [00:05<00:00, 16.4MB/s]19MB/s]M [00:06<00:00, 3.76MB/s]
onnx/model.onnx: 100% 90.4M/90.4M [00:25<00:00, 3.51MB/s]
...

If I set persistent storage to local, those files are persistent and not pulled on every restart. But I prefer the pods to be deployments, not statefulsets.

[MannerMan]

Seconded, also - these files does not appear to even be stored in the S3 bucket configured, but rather downloaded directly from the internet each time. I use a local minio instance and it would be much faster to store & later pull these files from the configured S3 bucket.

I also found a bug with the STT option, for instance if i enter Systran/faster-whisper-large-v3 in STT Model and click the download button, it appears the model is only downloaded locally to the container - gone if restarted - and not automatically downloaded on startup either.

I'm guessing both these issues are more relevant as a bug directly to openwebui than the chart repo though?

[peem5210]

Same issue here

[westbrook-ai]

Sorry for the delay on this one. When I get the chance, I'll set up a test Open WebUI instance with S3 storage to see if this is a bug on Open WebUI's S3 integration or if it's an issue with the chart itself. If anyone has the time to perform that test and post the results, that would be helpful!

[goldyfruit]

Same issue for me with Minio.

[Gufderald]

Indeed, it looks like setting persistence.provider: s3 does not actually provide persistence, as /app/backend/data/* files (including webui.db) are not stored on s3.

[this-is-tobi]

Persistence only affects files uploaded by users (chat files, knowledge files, etc.), it is not really explicit but seems to match the OpenWebui documentation.