Flowise on Hugging Face. Persistent storage issues.

[SleepingSmel]

Hi everyone! Well, Hugging Face's server for Flowise is fantastic, it offers more free space and RAM than other paid options. However, there's no documentation on maintaining persistent storage across restarts. Even with the option to purchase a paid storage tier on HF, it doesn't work. Achieving this would make Flowise more accessible by preventing data loss during server updates. Thanks by advance!

[bitsnaps]

There are many Spaces on huggingface which you can try, the following code support persisting storage on data directory:

FROM node:18-alpine
USER root

# Arguments that can be passed at build time
ARG FLOWISE_PATH=/usr/local/lib/node_modules/flowise
ARG BASE_PATH=/data/.flowise
ARG DATABASE_PATH=$BASE_PATH
ARG APIKEY_PATH=$BASE_PATH
ARG SECRETKEY_PATH=$BASE_PATH
ARG LOG_PATH=$BASE_PATH/logs

# Set the ARGs as ENV
ENV FLOWISE_PATH=$FLOWISE_PATH
ENV BASE_PATH=$BASE_PATH
ENV DATABASE_PATH=$DATABASE_PATH
ENV APIKEY_PATH=$APIKEY_PATH
ENV SECRETKEY_PATH=$SECRETKEY_PATH
ENV LOG_PATH=$LOG_PATH

# Install dependencies
RUN apk add --no-cache git python3 py3-pip make g++ build-base cairo-dev pango-dev chromium

ENV PUPPETEER_SKIP_DOWNLOAD=true
ENV PUPPETEER_EXECUTABLE_PATH=/usr/bin/chromium-browser

# Install Flowise globally
RUN npm install -g flowise

# Configure Flowise directories using the ARG
RUN mkdir -p $FLOWISE_PATH/uploads /data && chmod -R 777 $FLOWISE_PATH /data

WORKDIR /data

CMD ["sh", "-c", "mkdir -p $LOG_PATH && chmod -R 777 $LOG_PATH && npx flowise start"]

You just need to make sure securing access (if you want to deploy it for prod by using lower user permission, e.g add these lines at the end before running flowise):

...
# Switch to the non-root user
USER 1000

# Expose the Flowise port
EXPOSE 3000
....

[SleepingSmel]

Hello, thanks for the response! But the link of the space you provided doesn't work (error 404)

Edit: Sorry, It was on my one of the private repo.

Hello @bitsnaps , I tried with runtime error, I modified the docker like this, did I do it right? Cause maybe I'm missing something. Thanks for your help again.

[bitsnaps]

I can see your Space is working fine, you may had to wait until the process finish the work:

P.S. The Space may go to sleeping due to inactivity mode, if you don't use it for hours/days...

[SleepingSmel]

Hi, that's another space, so I couldn't make it work with persistent storage. I created a new space, added variables like user and password, and pasted the code from the screenshot in the last message. Am I doing something wrong? Thanks again for your help 🙏

[bitsnaps]

@SleepingSmel

Huggingface Spaces supports enable data persistence, to do so, you'll need to configure a persistent storage volume using Spaces capabilities and ensure Flowise is set up to use this volume for its database and other necessary files, ot use that service you'll need to enable Persistent Storage

Keep in mind, Persistent Storage is a paid service which comes with its cost:

Here is how to do so:

1. Go to your Space's Settings in Huggingface.
2. Enable Persistent Storage.
3. Configure the persistent volume path to /app/data (as specified in the Dockerfile).

You'll also need to configure environment variables DATABASE_PATH (e.g., /app/data/flowise.db).

You can use the following docker-compose.yml example:

version: '3.1'

services:
  flowise:
    build: .
    ports:
      - "7860:7860"
    environment:
      - FLOWISE_USERNAME=admin
      - FLOWISE_PASSWORD=password
      - DATABASE_PATH=/app/data/flowise.db
    volumes:
      - ./data:/app/data

If you want more control or persist data on restart, you must have a VPS with volume storage.

[SleepingSmel]

Hello @bitsnaps, huge thanks for all your help, it helped me understand how the server and databases work. I think I'm close to solving the persistent storage issue with the solution of @amansoni7477030 as he is using HF for production purposes and therefore aligns with my purpose. Happy new year and wish you the best!

[amansoni7477030]

@SleepingSmel , as @bitsnaps mentioned, Hugging Face provides persistent storage, but you have to pay for it. If you don't want to pay and still want to use persistent storage, you can consider Aiven Cloud. They offer free PostgreSQL or MySQL databases. Simply create a free account on Aiven Cloud and select the SQL option. Afterward, enter the credentials (like host, port, and other details) in your configuration, and it should work fine.

Please note that free services like Aiven Cloud are completely free, but if there is no activity for weeks or months, they may close your service. To address this, the best approach is to deploy a cron job that makes a single request to Hosted Flowise every 1-2 days. By doing this, your Hugging Face Space and persistent storage will run continuously without any issues.

In fact, my Flowise setup has been working perfectly fine in production for the past six months without any problems. If you want, you can directly duplicate my space and use it for your own purposes by simply entering the necessary credentials.

Here is my space: FlowiseAI on Hugging Face

[SleepingSmel]

Hi @amansoni7477030, your solution is exactly what I was looking for, so big thanks! I actually duplicated your space and created a database in Aiven Cloud. I have a few questions to ensure I understood correctly, I'd appreciate it if you could answer them:

• If I understood correctly, with this solution, it is not necessary to purchase persistent storage in HF, so storage will not be erased (flows, credentials...) when updating the server to use new Flowise versions, right?
• The cron job works by sending a request to one of the CURL URLs of the chatflows, not the server URL itself, right?
  Thanks again, I'm really grateful if this solution works for the production level we are seeking for 🙏

[amansoni7477030]

@SleepingSmel
Yes, both points are correct:

1. Persistent Storage: With this solution, you don't need to purchase persistent storage in HF. Using external storage like a database in Aiven Cloud ensures that your data (flows, credentials, etc.) remains intact even when updating the Flowise server to newer versions.
2. Cron Job: The cron job operates by sending requests specifically to the CURL URLs of individual chatflows, not to the server URL itself.

Let me know if you have further questions. 😊

[bitsnaps]

@amansoni7477030 Thank you for all these details, @SleepingSmel you should also consider deactivate the Sleep mode when creating the paid Space service:

[mikevallesteros]

@SleepingSmel , as @bitsnaps mentioned, Hugging Face provides persistent storage, but you have to pay for it. If you don't want to pay and still want to use persistent storage, you can consider Aiven Cloud. They offer free PostgreSQL or MySQL databases. Simply create a free account on Aiven Cloud and select the SQL option. Afterward, enter the credentials (like host, port, and other details) in your configuration, and it should work fine.

Please note that free services like Aiven Cloud are completely free, but if there is no activity for weeks or months, they may close your service. To address this, the best approach is to deploy a cron job that makes a single request to Hosted Flowise every 1-2 days. By doing this, your Hugging Face Space and persistent storage will run continuously without any issues.

In fact, my Flowise setup has been working perfectly fine in production for the past six months without any problems. If you want, you can directly duplicate my space and use it for your own purposes by simply entering the necessary credentials.

Here is my space: FlowiseAI on Hugging Face

Hey, thanks for sharing, really helpful.

Can you guide me related to cron job? Shall I create a new chatflow and add any credentials to LLM or is it just fine to add a random flow and create a cron job with the CURL url?

[amansoni7477030]

@mikevallesteros You can either create a free Hugging Face Space separately or use a service like Make to integrate with Hugging Face Space. Here's a Dockerfile setup for the Hugging Face Space:

# Use an official lightweight base image
FROM debian:bullseye-slim

# Install curl, bash, and netcat (to run a simple HTTP server)
RUN apt-get update && apt-get install -y \
    curl \
    netcat \
    && rm -rf /var/lib/apt/lists/*

# Create a script to run the curl requests in an infinite loop
RUN echo '#!/bin/bash\n\
while true; do\n\
    # First request to the Flowise API (response and response code)\n\
    response1=$(curl -s -w "%{http_code}" -o /tmp/response1.txt https://amansoni7477030-flowiseai.hf.space/api/v1/prediction/7b59afaa-c200-491a-b428-c99bfe1be4fa \\\n\
        -X POST \\\n\
        -d "{\\"question\\": \\"Hey, how are you?\\"}" \\\n\
        -H "Content-Type: application/json")\n\
    response_code1="${response1: -3}"\n\
    response_body1=$(cat /tmp/response1.txt)\n\
    echo "Response from Flowise API (status code $response_code1): $response_body1"\n\
    \n\
    # Sleep for 24 hours (86400 seconds) to run the job once per day\n\
    sleep 86400\n\
done' > /usr/local/bin/run-curl.sh

# Make the script executable
RUN chmod +x /usr/local/bin/run-curl.sh

# Expose port 7860 for Hugging Face
EXPOSE 7860

# Run the curl script in the background and start a simple HTTP server to keep the container running
CMD /usr/local/bin/run-curl.sh & nc -lk -p 7860

Instructions:

1. Hugging Face Space: You can use this Dockerfile for Hugging Face Space. Input your Flowise API URL (e.g., https://amansoni7477030-flowiseai.hf.space/api/v1/prediction/7b59afaa-c200-491a-b428-c99bfe1be4fa) and replace the URL in the curl request.
2. Cron Job Timing:

• In the script, I've set sleep 86400 to make the job run once per day (86400 seconds).
• This interval can be adjusted based on your requirement, e.g., for running the job every 5 minutes, you can use sleep 300.

3. Random Flow: It's fine to use any random flow for testing purposes, but if you're looking to integrate it with a real use case, you should ideally point it to a specific flow.
4. Using Make: Another alternative is to use Make.com, which is a more advanced integration tool. It allows for more flexible scheduling, but it can be a bit more complex. However, I find using Hugging Face Space simple for setting up cron jobs. If you prefer something more powerful, you might want to research Make further.

[SleepingSmel]

Hello @amansoni7477030 . Actually, I've had a problem with the credentials, this error pops up every day, when I discover all the credentials are either lost or don't work. However, if I create a new one, it works temporarily until the same error pops up. Do you have any idea why this happens? Everything else work fine, which makes it a weird problem. Thanks in advance.

PS: I just found this discussion in Github forum, and the problem is not yet solved apparently: #2467

[amansoni7477030]

@SleepingSmel To use Flowise with MySQL, the required variables are quite straightforward and are already listed in this space: Flowise Space, which belongs to one of my clients. Regarding the FLOWISE_SECRETKEY_OVERWRITE, I don’t believe it’s causing the issue, but I recommend generating a proper 32-bit encryption key for it.

I also suggest reviewing all the environment variables in the Flowise Documentation to ensure everything is configured correctly.

If you're using MySQL, you can duplicate the setup provided in the above link for reference.

As for PostgreSQL, I haven’t encountered any errors as mentioned in the screenshot you shared. It’s possible there might be a misconfiguration or an oversight causing the issue. To investigate further, please share the logs from Hugging Face when you experience any errors with Flowise. That will help identify the root cause more effectively.

[bitsnaps]

@SleepingSmel Try to set a minimal amount of variable environments, as most of the other variables have a good set of default values. For example, FLOWISE_SECRETKEY_OVERWRITE seems to be optional, like most of the other variables, since it's only required when you don't use SECRETKEY_PATH, which is set by default. I had no issues deploying with PostgreSQL, and I don't think using MySQL is a better option, but it should work anyway.

[SleepingSmel]

Hello all! After some weeks, this method definitely works, in my case with MySQL, so huuuuuge thanks, cause this is a game changer for me and hopefully for more people that follow this thread! With a cronjob every 24 hours to one of the CURL URL of the flows, the server will remain active automatically. I used this web to do the cronjob, it works for free, and I have 3 cronjobs: https://cron-job.org/en/

Hi @amansoni7477030 , on the other hand, I wonder how updating the server works. I've restarted the server, but the Flowise version stays the same. Yesterday, a new update released, but I couldn't update the flowise version in my server. Do I have to wait for the space that I duplicated (yours) to be updated? Or maybe your branch in GitHub? Thanks 🙏

[amansoni7477030]

@SleepingSmel You're welcome! 😊 Regarding the Flowise version update, you can resolve this by going to the Settings of your Hugging Face Space and searching for the Factory Rebuild option.

This will clear all the cache from your Hugging Face Space and reinstall Flowise with the latest version.

No need to worry about your database, as it is stored separately and will not be affected during this process. It's a great feature that keeps your data intact while updating the application.

[HenryHengZJ]

Huge thanks to @amansoni7477030 and others for helping out!

[SleepingSmel]

Hi @amansoni7477030 and @bitsnaps , sadly, the factory rebuild crashed an error as in the screenshot, I send a pasted code of the error in case you can guess the problem with the MySQL database... Thanks 🙏 : https://justpaste.it/gjzz7

I created a new space with the same credentials and the server works and retrieves the data from the database, but the flowise version remains the same, without updating to the new 2.2.4 version

[amansoni7477030]

@SleepingSmel , I think you might have a memory issue that's causing the 137 error. I performed a factory rebuild in my space, and it worked perfectly fine. Perhaps you changed some variable or details that are leading to this problem. As you can see in the screenshot:

it is working perfectly fine for me.

Regarding the memory error in Docker, you can visit this link:
GitLab pipeline exiting with error code 137 when running Cypress.

Another suggestion is to try installing Flowise without cloning it—just directly using npm as mentioned here:
Flowise Quick Start

You can add this to your Dockerfile and check if the same error occurs. Also, refer to this documentation for more details:
FlowiseAI Deployment on Hugging Face.

[SleepingSmel]

Hi @amansoni7477030, I've decided to try again with Postgres database, as it appears to be the common option. However, I still have the credentials issue I posted a few weeks ago, everything else work perfectly. You said that I could share the logs. Thanks!

Logs:

2025-01-17 11:49:02 [INFO]: ⬆️ POST /api/v1/internal-prediction/4c0f08aa-2f29-400f-bba8-61dc9592f0a2
2025-01-17 11:49:02 [ERROR]: Error: Malformed UTF-8 data
Error: Error: Malformed UTF-8 data
    at getCredentialData (/home/node/app/packages/components/dist/src/utils.js:604:15)
    at async ChatOpenAI_ChatModels.init (/home/node/app/packages/components/dist/nodes/chatmodels/ChatOpenAI/ChatOpenAI.js:181:32)
    at async buildFlow (/home/node/app/packages/server/dist/utils/index.js:468:36)
    at async buildAgentGraph (/home/node/app/packages/server/dist/utils/buildAgentGraph.js:88:32)
    at async utilBuildAgentResponse (/home/node/app/packages/server/dist/utils/buildChatflow.js:442:31)
    at async utilBuildChatflow (/home/node/app/packages/server/dist/utils/buildChatflow.js:195:20)
    at async createAndStreamInternalPrediction (/home/node/app/packages/server/dist/controllers/internal-predictions/index.js:33:29)
2025-01-17 11:49:02 [ERROR]: [server]: Error: Error: Error: Malformed UTF-8 data
Error: Error: Error: Malformed UTF-8 data
    at buildFlow (/home/node/app/packages/server/dist/utils/index.js:517:19)
    at async buildAgentGraph (/home/node/app/packages/server/dist/utils/buildAgentGraph.js:88:32)
    at async utilBuildAgentResponse (/home/node/app/packages/server/dist/utils/buildChatflow.js:442:31)
    at async utilBuildChatflow (/home/node/app/packages/server/dist/utils/buildChatflow.js:195:20)
    at async createAndStreamInternalPrediction (/home/node/app/packages/server/dist/controllers/internal-predictions/index.js:33:29)
2025-01-17 11:49:02 [ERROR]: [server]: Error: Error buildAgentGraph - Error: Error: Malformed UTF-8 data
Error: Error buildAgentGraph - Error: Error: Malformed UTF-8 data
    at buildAgentGraph (/home/node/app/packages/server/dist/utils/buildAgentGraph.js:409:15)
    at async utilBuildAgentResponse (/home/node/app/packages/server/dist/utils/buildChatflow.js:442:31)
    at async utilBuildChatflow (/home/node/app/packages/server/dist/utils/buildChatflow.js:195:20)
    at async createAndStreamInternalPrediction (/home/node/app/packages/server/dist/controllers/internal-predictions/index.js:33:29)
2025-01-17 11:49:02 [ERROR]: [server]: Error: Error buildAgentGraph - Error: Error: Malformed UTF-8 data
Error: Error buildAgentGraph - Error: Error: Malformed UTF-8 data
    at utilBuildAgentResponse (/home/node/app/packages/server/dist/utils/buildChatflow.js:548:15)
    at async utilBuildChatflow (/home/node/app/packages/server/dist/utils/buildChatflow.js:195:20)
    at async createAndStreamInternalPrediction (/home/node/app/packages/server/dist/controllers/internal-predictions/index.js:33:29)