Skip to content

← Back to Index

🔬 Remote Ollama Testing Guide

This guide shows you how to test IQToolkit Analyzer against a remote Ollama server for distributed deployment scenarios.

🎯 Overview

You can run Ollama on a dedicated server (e.g., a GPU-equipped machine) and connect to it from your development or analysis machines. This is ideal for:

  • Centralized AI infrastructure: One powerful GPU server serving multiple analysts
  • Development/production separation: Dev machines connect to staging/prod Ollama servers
  • Privacy-compliant deployments: Keep AI processing within your network perimeter
  • Resource optimization: Share expensive GPU resources across teams

🖥️ Server Setup

1. Install Ollama on Your Server

# SSH to your server (e.g., 192.168.0.30)
ssh user@192.168.0.30

# Install Ollama
curl -LsSf https://ollama.com/install.sh | sh

# Start Ollama service
ollama serve

# Pull the SQL-specialized model
ollama pull a-kore/Arctic-Text2SQL-R1-7B

2. Verify Server is Running

# Check if Ollama is listening
curl http://localhost:11434/api/version

# List available models
ollama list

3. Configure Firewall (if needed)

Ensure port 11434 is accessible from your client machines:

# Example for Ubuntu/Debian with ufw
sudo ufw allow 11434/tcp

# Example for firewalld (CentOS/RHEL)
sudo firewall-cmd --add-port=11434/tcp --permanent
sudo firewall-cmd --reload

💻 Client Setup

1. Install Dependencies

# On your local machine
cd /path/to/iqtoolkit-analyzer

# Create/Select a Poetry environment (Python 3.13 recommended)
poetry env use python3.13

# Install dependencies
poetry install --with dev,test

2. Test Connection

# Set the remote Ollama host
export OLLAMA_HOST=http://192.168.0.30:11434

# Quick connection test
poetry run python -c "import ollama; print(ollama.list())"

Expected output: 

models=[Model(model='a-kore/Arctic-Text2SQL-R1-7B:latest', ...)]

🧪 Running Integration Tests

Using the Test Script

The repository includes a comprehensive test script that validates the remote connection:

# Run the integration test
export OLLAMA_HOST=http://192.168.0.30:11434
python test_remote_ollama.py

Expected output:

Remote Ollama Test Success

The test script validates: - ✅ Direct connection to remote Ollama server - ✅ Model availability and version information - ✅ Chat functionality with the SQL-specialized model - ✅ Full IQToolkit Analyzer LLMClient integration - ✅ AI-generated query optimization recommendations

Running Unit Tests

Use the provided pytest configuration to run tests without coverage (for faster testing):

# Run all LLM client tests
OLLAMA_HOST=http://192.168.0.30:11434 pytest -c pytest-remote.ini tests/test_llm_client.py -v

# Run only Ollama-specific tests
OLLAMA_HOST=http://192.168.0.30:11434 pytest -c pytest-remote.ini tests/test_llm_client.py::TestLLMClientOllama -v

# Run a specific test
OLLAMA_HOST=http://192.168.0.30:11434 pytest -c pytest-remote.ini \
  tests/test_llm_client.py::TestLLMClientOllama::test_ollama_custom_host -v

Example output: 

tests/test_llm_client.py::TestLLMClientOllama::test_ollama_initialization PASSED
tests/test_llm_client.py::TestLLMClientOllama::test_ollama_generate_recommendations PASSED
tests/test_llm_client.py::TestLLMClientOllama::test_ollama_custom_host PASSED

⚙️ Configuration

Option 1: Environment Variable (Temporary)

# Set for current session
export OLLAMA_HOST=http://192.168.0.30:11434

# Run analysis
python -m iqtoolkit_analyzer your_log_file.log

Option 2: Configuration File (Persistent)

Create or edit .iqtoolkit-analyzer.yml:

# AI Provider Configuration
llm_provider: ollama
ollama_model: a-kore/Arctic-Text2SQL-R1-7B
ollama_host: http://192.168.0.30:11434

# LLM Settings
llm_temperature: 0.3
max_tokens: 300
llm_timeout: 30

# Analysis Options
top_n: 10
min_duration: 1000
output: reports/analysis.md

Then run without environment variables:

python -m iqtoolkit_analyzer your_log_file.log

Option 3: Shell Profile (Permanent)

Add to your ~/.bashrc, ~/.zshrc, or ~/.profile:

# Remote Ollama Configuration
export OLLAMA_HOST=http://192.168.0.30:11434

Reload your shell: 

source ~/.bashrc  # or ~/.zshrc

🔍 Analyzing PostgreSQL Logs

With the remote server configured, analyze logs normally:

# Basic analysis
python -m iqtoolkit_analyzer postgresql /path/to/postgresql.log --output report.md

# Advanced analysis
python -m iqtoolkit_analyzer postgresql /path/to/postgresql.log \
  --output report.md \
  --top-n 15 \
  --min-duration 500 \
  --max-tokens 200 \
  --verbose

The AI recommendations will be generated using the remote Ollama server.

🐛 Troubleshooting

Connection Refused

# Test if server is reachable
ping 192.168.0.30

# Test if port is open
nc -zv 192.168.0.30 11434
# or
telnet 192.168.0.30 11434

Solution: Check firewall rules on the server.

Model Not Found

# List models on remote server
curl http://192.168.0.30:11434/api/tags

# Pull the model on the server
ssh user@192.168.0.30
ollama pull a-kore/Arctic-Text2SQL-R1-7B

Timeout Errors

Increase timeout in .iqtoolkit-analyzer.yml:

llm_timeout: 60  # Increase from default 30 seconds

Or check network latency: 

ping -c 10 192.168.0.30

SSL/TLS Errors

If using HTTPS, ensure valid certificates:

ollama_host: https://192.168.0.30:11434  # Use https://

📊 Performance Considerations

Network Latency

  • Local Ollama: <1ms latency
  • LAN (same subnet): 1-5ms latency (minimal impact)
  • WAN/VPN: 20-100ms+ latency (may affect interactive use)

Recommendation: For best performance, deploy Ollama on the same network segment as your analysis machines.

Bandwidth Requirements

  • Model loading: One-time per model (7-8GB for Arctic-Text2SQL-R1-7B)
  • Inference: ~10-50KB per query analysis
  • Concurrent users: Server can handle multiple clients simultaneously

Server Resources

  • Minimum: 16GB RAM, 4 CPU cores
  • Recommended: 32GB+ RAM, 8+ CPU cores, GPU (NVIDIA)
  • Storage: 50GB+ for models and cache

🔒 Security Best Practices

Network Security

  • Use private networks: Keep Ollama on internal networks only
  • VPN access: Require VPN for remote access
  • Firewall rules: Restrict port 11434 to authorized IPs only

Authentication (Optional)

For production deployments, consider: - Reverse proxy with authentication (nginx, Apache) - API gateway with token-based auth - Network segmentation and access controls

Data Privacy

  • Advantage: All query data stays within your infrastructure
  • No external APIs: Data never leaves your network
  • Compliance-ready: Meets data residency requirements

💡 Tips & Tricks

Multiple Environments

Manage different environments with shell aliases:

# Add to ~/.bashrc or ~/.zshrc
alias ollama-dev='export OLLAMA_HOST=http://192.168.0.30:11434'
alias ollama-staging='export OLLAMA_HOST=http://192.168.0.31:11434'
alias ollama-prod='export OLLAMA_HOST=http://192.168.0.32:11434'
alias ollama-local='export OLLAMA_HOST=http://localhost:11434'

Usage: 

ollama-dev
python -m iqtoolkit_analyzer your_log.log

Health Check Script

Create a health check script (check_ollama.sh):

#!/bin/bash
OLLAMA_HOST=${OLLAMA_HOST:-http://localhost:11434}

echo "Checking Ollama at $OLLAMA_HOST..."
curl -s $OLLAMA_HOST/api/version && echo "✅ Server is healthy" || echo "❌ Server is down"

Make it executable and run: 

chmod +x check_ollama.sh
./check_ollama.sh

Monitoring

Monitor Ollama server performance:

# On the server
htop  # CPU/Memory usage
nvidia-smi  # GPU usage (if applicable)
journalctl -u ollama -f  # Service logs

Made with ❤️ for distributed AI-powered database analysis