How do I configure and troubleshoot Streamlit and Gradio applications in my SageMaker AI notebook instances?

6 minute read

I want to configure and troubleshoot Streamlit and Gradio applications in my Amazon SageMaker AI notebook instances.

Resolution

Make sure that you meet the following requirements:

You have an active SageMaker AI notebook instance.
Your notebook instance has internet access to download and install Python packages and dependencies. If you turned off internet access, then update the scripts and make sure that your packages are available offline in the notebook instance.
The user that runs the notebook has sudo permissions to install the system dependencies. By default, the SageMaker AI notebook instance environment provides sudo permissions.
Your notebook instance has Python 3.6 or later. By default, SageMaker AI notebook instances have Python preinstalled.
Other processes don't use Ports 8501 for Streamlit and 7861 for Gradio in your notebook instance. You can update your default ports in the scripts based on your use case.
The security group that's associated with your notebook instance allows inbound traffic on ports 8501 and 7861.

Note: The following scripts are for a Jupyter notebook code cell.

Install and configure your Streamlit application

Complete the following steps:

Run the following one-time setup script:

%%bash
#!/bin/bash

CURRENTDATE=$(date +"%Y-%m-%d %T")
CYAN='\033[1;36m'
NC='\033[0m'

# Create the app.py file for Streamlit
echo -e "${CYAN}${CURRENTDATE}: [INFO]:${NC}Creating app_streamlit.py for streamlit\n"
cat << EOF > app_streamlit.py
import streamlit as st

st.title("Hello World")
EOF

echo -e "app_streamlit.py has been created successfully \n"
echo -e "${CYAN}${CURRENTDATE}: [INFO]:${NC}Setup completed\n"

The preceding script uses pip and the appropriate package manager to install Streamlit. Also, the script creates a basic app_streamlit.py Streamlit application file with a "Hello World" title.

Run the following script:

%%bash
#!/bin/bash

CURRENTDATE=$(date +"%Y-%m-%d %T")
RED='\033[0;31m'
CYAN='\033[1;36m'
GREEN='\033[1;32m'
NC='\033[0m'

echo -e "${CYAN}${CURRENTDATE}: [INFO]:${NC}Getting the URL to view your Streamlit app in the browser\n"

PORT=8501
echo -e "${CYAN}${CURRENTDATE}: [INFO]:${NC} Port Number ${PORT}\n" 

RESOURCE_NAME=$(jq .ResourceName /opt/ml/metadata/resource-metadata.json || exit 1)
RESOURCE_ARN=$(jq .ResourceArn /opt/ml/metadata/resource-metadata.json || exit 1)

RESOURCE_NAME=$(sed -e 's/^"//' -e 's/"$//' <<< "$RESOURCE_NAME")
RESOURCE_ARN=$(sed -e 's/^"//' -e 's/"$//' <<< "$RESOURCE_ARN")
RESOURCE_ARN_ARRAY=($(echo "$RESOURCE_ARN" | tr ':' '\n'))
REGION=$(echo "${RESOURCE_ARN_ARRAY[3]}")

NOTEBOOK_URL="https://${RESOURCE_NAME}.notebook.${REGION}.sagemaker.aws"
link="${NOTEBOOK_URL}/proxy/${PORT}/"

echo -e "${CYAN}${CURRENTDATE}: [INFO]:${NC} Starting Streamlit App"
echo -e "${CYAN}${CURRENTDATE}: [INFO]: ${GREEN}${link}${NC}"

# Run the Streamlit app and save the output to "streamlit.txt"
echo -e "${CYAN}${CURRENTDATE}: [INFO]:${NC}Allow upto 10 seconds before clicking the link for app launch to be successful"
streamlit run app_streamlit.py --browser.serverPort=${PORT} --server.port=${PORT} >streamlit.txt

The preceding script runs the Streamlit application on port 8501 and saves the output to a temporary file. The script uses SageMaker notebook instance information from the metadata to display the Streamlit URL.

For more information about how to install Streamlit, see Install Streamlit on the Streamlit website.

Troubleshoot issues with your Streamlit application

Take the following actions:

If the Streamlit application doesn't start, then run the following command to check whether you correctly installed the application:
```
pip list | grep streamlit
```
Run the following command to make sure that other processes don't use port 8501:
```
lsof -i :8501
```
Make sure that your SageMaker AI notebook instance's security group allows inbound traffic on the port 8501. If needed, you can update the PORT number in the script to update the port that Streamlit uses.

To debug Streamlit applications, add the --logger.level=debug argument to the last line of the script, such as in the following example:

streamlit run app_streamlit.py --browser.serverPort=${PORT} --logger.level=debug --server.port=${PORT} >streamlit_logs.txt

Install and configure your Gradio application

Complete the following steps:

Run the following one-time setup script:

%%bash
#!/bin/bash
CURRENTDATE=`date +"%Y-%m-%d %T"`
RED='\033[0;31m'
CYAN='\033[1;36m'
GREEN='\033[1;32m'
NC='\033[0m'

PORT=7861

# Install python dependencies
echo -e "${CYAN}${CURRENTDATE}: [INFO]:${NC}Running pip install gradio\n"
pip install --no-cache-dir -q gradio

# Create the app.py file for Gradio
echo -e "${CYAN}${CURRENTDATE}: [INFO]:${NC}Creating app_gradio.py for Gradio\n"
cat << EOF > app_gradio.py
import gradio as gr
def greet(name):
    return "Hello " + name + "!"
demo = gr.Interface(fn=greet, inputs= "text", outputs = "text")
demo.launch(inline=False, debug=True, server_port=${PORT}, root_path="/proxy/${PORT}")
EOF

echo -e "${CYAN}${CURRENTDATE}: [INFO]:${NC}Setup completed. "

The preceding script uses pip and the appropriate package manager to install Gradio and other operating system (OS) dependencies. Also, the script creates a basic app_gradio.py Gradio application file with a greeting function.

Run the following script:

%%bash
CYAN='\033[1;36m'
GREEN='\033[1;32m'
RED='\033[0;31m'
NC='\033[0m'

PORT=7861
MAX_RETRIES=15
RETRY_INTERVAL=3
GRADIO_PID=""

cleanup() {
    echo -e "\n${CYAN}[INFO]: Cleaning up processes...${NC}"
    if [ ! -z "$GRADIO_PID" ]; then
        pkill -P $GRADIO_PID
        kill -9 $GRADIO_PID 2>/dev/null
    fi
    pkill -f "python app_gradio.py"
    exit 0
}


check_gradio() {
    netstat -tuln | grep ":${PORT}" >/dev/null
}


trap cleanup EXIT INT TERM

echo -e "${CYAN}[INFO]: Starting Gradio App${NC}"
python app_gradio.py &
GRADIO_PID=$!

echo -e "${CYAN}[INFO]: Waiting for Gradio to start on port ${PORT}...${NC}"
retries=0
while ! check_gradio; do
    if [ $retries -ge $MAX_RETRIES ]; then
        echo -e "${RED}[ERROR]: Gradio failed to start after ${MAX_RETRIES} attempts${NC}"
        cleanup
        exit 1
    fi
    sleep $RETRY_INTERVAL
    ((retries++))
done

RESOURCE_NAME=$(jq -r .ResourceName /opt/ml/metadata/resource-metadata.json)
RESOURCE_ARN=$(jq -r .ResourceArn /opt/ml/metadata/resource-metadata.json)
REGION=$(echo "$RESOURCE_ARN" | cut -d':' -f4)

echo -e "${GREEN}Gradio is running at:${NC}"
echo -e "${GREEN}https://${RESOURCE_NAME}.notebook.${REGION}.sagemaker.aws/proxy/${PORT}/${NC}"

while true; do
    if ! check_gradio; then
        echo -e "${RED}[ERROR]: Gradio process is no longer running${NC}"
        cleanup
        exit 1
    fi
    sleep 10
done

The preceding script sets up a cleanup function to remove the Gradio process on exit. The script runs the Gradio app in the background, and then retrieves the SageMaker AI notebook instance information from the metadata. Then, it constructs and displays the URL to access the Gradio application in the browser.

For more information about how to install Gradio, see Installation on the Gradio website.

Troubleshoot your Gradio application

Take the following actions:

If the Gradio app doesn't start, then run the following command to check that you correctly installed the application:
```
pip list | grep gradio
```
Run the following command to make sure that other processes don't use port 7861:
```
lsof -i :7861
```
If you can't generate a URL, then check the file permissions on the metadata file.
Make sure that the SageMaker AI notebook instance's security group allows inbound traffic on the required port.
If you can't access the application, then make sure that the Gradio launch configuration root_path matches the URL proxy path.

If you receive the "URL 500: Internal Error" message, then stop and rerun the script from step 2 in the Configure your Gradio application section. Or, run the following command to reinstall your application:

pip install --no-cache-dir -q gradiohttps://www.gradio.app/guides/quickstart

Topics: Machine Learning & AI
Tags: Amazon SageMaker
Language: English

AWS OFFICIALUpdated a year ago

No comments

Relevant content

sagemaker AI notebook is broken
Sandy1234
asked 8 months ago
How to Migrate Sagemaker Notebook with Platform Identifier notebook-al2-v2 to notebook-al2023-v1?
Marcos Ramos
asked 8 months ago
Jupyternaut (jupyter-ai) in Sagemaker Studio notebook. How to get it working?
Lauren
asked 3 years ago
SageMaker Notebook Instance: cannot import name 'get_execution_role' from 'sagemaker'
Accepted Answer
Samuel
asked 3 years ago
How do I run a python script on a Sagemaker notebook instance?
JohnB
asked 3 years ago
How do I troubleshoot SageMaker AI notebook instance connectivity issues?
AWS OFFICIALUpdated 9 months ago
How do I troubleshoot a SageMaker AI notebook instance that can't open Jupyter?
AWS OFFICIALUpdated a year ago
How do I configure KMS encryption in a SageMaker AI Notebook instance?
AWS OFFICIALUpdated a year ago
How do I troubleshoot notebook scheduling errors in SageMaker AI Studio?
AWS OFFICIALUpdated 9 months ago
How do I resolve the Amazon SageMaker AI endpoint error “Unable to locate at least 2 availability zone(s) with the requested instance type that overlap with SageMaker subnets”?
EXPERT
Avinash
published 9 months ago