ThoughtMCP

#!/bin/sh # Ollama Container Entrypoint Script # # This script ensures required models are pulled before the container is considered ready: # 1. Starts Ollama server in background # 2. Waits for Ollama API to be ready # 3. Pulls the embedding model from EMBEDDING_MODEL env var # 4. Pulls the inference model from LLM_MODEL env var # 5. Keeps Ollama running in foreground # # Environment Variables: # EMBEDDING_MODEL - Embedding model to pull (default: nomic-embed-text) # LLM_MODEL - Inference model to pull (default: llama3.2:1b) # OLLAMA_WAIT_TIMEOUT - Max seconds to wait for Ollama startup (default: 60) # SKIP_MODEL_PULL - Skip model pull if set to "true" (default: false) # # Usage: # ./ollama-entrypoint.sh # EMBEDDING_MODEL=nomic-embed-text LLM_MODEL=llama3.2:1b ./ollama-entrypoint.sh set -e # Configuration with defaults EMBEDDING_MODEL="${EMBEDDING_MODEL:-nomic-embed-text}" LLM_MODEL="${LLM_MODEL:-llama3.2:1b}" OLLAMA_WAIT_TIMEOUT="${OLLAMA_WAIT_TIMEOUT:-60}" SKIP_MODEL_PULL="${SKIP_MODEL_PULL:-false}" # Logging functions log_info() { echo "[OLLAMA-INIT] $(date '+%Y-%m-%d %H:%M:%S') - $1" } log_warn() { echo "[OLLAMA-INIT] $(date '+%Y-%m-%d %H:%M:%S') - WARN: $1" } log_error() { echo "[OLLAMA-INIT] $(date '+%Y-%m-%d %H:%M:%S') - ERROR: $1" >&2 } # Wait for Ollama API to be ready using ollama list command wait_for_ollama_ready() { log_info "Waiting for Ollama API to be ready..." elapsed=0 interval=2 while [ $elapsed -lt $OLLAMA_WAIT_TIMEOUT ]; do # Use ollama list to check if server is ready (curl not available in image) if ollama list > /dev/null 2>&1; then log_info "Ollama API is ready!" return 0 fi sleep $interval elapsed=$((elapsed + interval)) done log_error "Timeout waiting for Ollama API after ${OLLAMA_WAIT_TIMEOUT} seconds" return 1 } # Check if model is already present using ollama list is_model_present() { model_name="$1" # Use ollama list and grep for the model name if ollama list 2>/dev/null | grep -q "$model_name"; then return 0 fi return 1 } # Pull a model pull_model() { model_name="$1" model_type="$2" if [ "$SKIP_MODEL_PULL" = "true" ]; then log_info "Skipping ${model_type} model pull (SKIP_MODEL_PULL=true)" return 0 fi log_info "Checking for ${model_type} model: ${model_name}" if is_model_present "$model_name"; then log_info "${model_type} model ${model_name} is already present" return 0 fi log_info "Pulling ${model_type} model: ${model_name}..." log_info "This may take several minutes on first run..." # Use ollama pull command directly if ollama pull "$model_name"; then log_info "Successfully pulled ${model_type} model: ${model_name}" return 0 fi log_error "Failed to pull ${model_type} model: ${model_name}" return 1 } # Main execution main() { log_info "Ollama Entrypoint Script Starting..." log_info "Embedding Model: ${EMBEDDING_MODEL}" log_info "Inference Model: ${LLM_MODEL}" # Start Ollama server in background log_info "Starting Ollama server..." ollama serve & OLLAMA_PID=$! # Wait for Ollama to be ready if ! wait_for_ollama_ready; then log_error "Ollama failed to start. Exiting." exit 1 fi # Pull embedding model (required) if ! pull_model "$EMBEDDING_MODEL" "embedding"; then log_error "Failed to pull required embedding model. Exiting." kill $OLLAMA_PID 2>/dev/null || true exit 1 fi # Pull inference model (optional - warn but don't fail) if ! pull_model "$LLM_MODEL" "inference"; then log_warn "Inference model not available. AI-augmented reasoning may be limited." fi log_info "All models ready. Ollama initialization complete." log_info "Ollama server running (PID: ${OLLAMA_PID})" # Wait for Ollama process (keep container running) wait $OLLAMA_PID } # Run main function main "$@"