qdrant-docker-tools

qdrant-tool.sh

#!/bin/bash

# ==============================================================================
# Unified Qdrant Docker Troubleshooter (v3)
#
# This script combines various diagnostic checks for a Qdrant instance
# running in Docker into a single, easy-to-use tool.
#
# It now includes a comprehensive batch test and report option.
#
# Usage:
#   ./qdrant_troubleshooter.sh [command]
#
# If no command is provided, it will display an interactive menu.
# ==============================================================================

# --- Configuration ---
# Please edit these variables to match your environment.

# The name or ID of your primary Qdrant container.
QDRANT_CONTAINER_NAME="qdrant"

# The host and port for the Qdrant API.
QDRANT_HOST="localhost:6333"

# --- End of Configuration ---


# --- Helper Functions ---

# Function to print a colored header
print_header() {
    echo
    echo "================================================="
    echo "  $1"
    echo "================================================="
    echo
}

# Function to dynamically get the list of collections and have the user select one
select_collection() {
    print_header "Fetching Collections"
    # Fetch collections and parse names. This avoids needing jq.
    collections=($(curl -s "http://$QDRANT_HOST/collections" | grep -o '"name":"[^"]*"' | sed 's/"name":"//;s/"//'))

    if [ ${#collections[@]} -eq 0 ]; then
        echo "No collections found or could not connect to Qdrant at http://$QDRANT_HOST."
        return 1
    fi

    echo "Please select a collection:"
    select col in "${collections[@]}"; do
        if [ -n "$col" ]; then
            SELECTED_COLLECTION=$col
            echo "You selected: $SELECTED_COLLECTION"
            return 0
        else
            echo "Invalid selection. Please try again."
        fi
    done
    return 1
}


# --- Diagnostic Functions ---

# 1. Check the status of all Docker containers.
check_container_status() {
    print_header "Docker Container Status"
    docker ps -a
}

# 2. View the logs of the Qdrant container.
view_container_logs() {
    print_header "Logs for $QDRANT_CONTAINER_NAME"
    docker logs "$QDRANT_CONTAINER_NAME"
}

# 3. Check for port conflicts on the host.
check_port_conflicts() {
    print_header "Checking for Port Conflicts (6333, 6334)"
    echo "--- Checking Port 6333 ---"
    lsof -i :6333 || echo "Port 6333 is not in use or lsof is not installed."
    echo
    echo "--- Checking Port 6334 ---"
    lsof -i :6334 || echo "Port 6334 is not in use or lsof is not installed."
}

# 4. Check the Qdrant API health endpoint.
check_health_endpoint() {
    print_header "Checking Qdrant API Health Endpoint"
    http_code=$(curl -s -o /dev/null -w "%{http_code}" "http://$QDRANT_HOST/healthz")
    if [ "$http_code" -eq 200 ]; then
        echo "Health check successful (HTTP 200 OK)"
    else
        echo "Health check failed. Received HTTP status $http_code."
        echo "Is Qdrant running and the port correctly mapped?"
    fi
    echo
}

# 5. Check permissions of the data volume (Dynamically discovered).
check_volume_permissions() {
    print_header "Checking Data Volume Permissions"
    # Dynamically find the host path mapped to /qdrant/storage in the container
    local volume_path
    volume_path=$(docker inspect -f '{{range .Mounts}}{{if eq .Destination "/qdrant/storage"}}{{.Source}}{{end}}{{end}}' "$QDRANT_CONTAINER_NAME")

    if [ -z "$volume_path" ]; then
        echo "Could not find a volume mounted to /qdrant/storage for container '$QDRANT_CONTAINER_NAME'."
        return 1
    fi

    echo "Discovered data volume path: $volume_path"
    if [ ! -d "$volume_path" ]; then
      echo "Error: Directory not found: $volume_path"
      return 1
    fi

    echo "--- Permissions for $volume_path ---"
    ls -ld "$volume_path"
}

# 6. Create a snapshot for a user-selected collection.
create_snapshot() {
    if ! select_collection; then return 1; fi
    print_header "Creating Snapshot for '$SELECTED_COLLECTION'"
    curl -X POST "http://$QDRANT_HOST/collections/$SELECTED_COLLECTION/snapshots"
    echo
}

# 7. List all snapshots for a user-selected collection.
list_snapshots() {
    if ! select_collection; then return 1; fi
    print_header "Listing Snapshots for '$SELECTED_COLLECTION'"
    curl "http://$QDRANT_HOST/collections/$SELECTED_COLLECTION/snapshots"
    echo
}

# 8. Restore a collection from a user-selected snapshot.
restore_snapshot() {
    if ! select_collection; then return 1; fi
    local collection_to_restore=$SELECTED_COLLECTION

    print_header "Fetching Snapshots for '$collection_to_restore'"
    # Fetch snapshots for the selected collection
    snapshots=($(curl -s "http://$QDRANT_HOST/collections/$collection_to_restore/snapshots" | grep -o '"name":"[^"]*"' | sed 's/"name":"//;s/"//'))

    if [ ${#snapshots[@]} -eq 0 ]; then
        echo "No snapshots found for collection '$collection_to_restore'."
        return 1
    fi

    echo "Please select a snapshot to restore:"
    select snap in "${snapshots[@]}"; do
        if [ -n "$snap" ]; then
            local snapshot_to_restore=$snap
            break
        else
            echo "Invalid selection. Please try again."
        fi
    done

    if [ -z "$snapshot_to_restore" ]; then
      echo "No snapshot selected. Aborting."
      return 1
    fi

    # Construct the snapshot location URI inside the container
    local snapshot_uri="file:///qdrant/storage/snapshots/$collection_to_restore/$snapshot_to_restore"

    print_header "Restoring '$collection_to_restore' from Snapshot"
    echo "Attempting to restore from: $snapshot_uri"
    curl -X PUT "http://$QDRANT_HOST/collections/$collection_to_restore/snapshots/recover" \
      -H "Content-Type: application/json" \
      -d "{\"location\": \"$snapshot_uri\"}"
    echo
}

# 9. Check the container's resource usage.
check_resource_usage() {
    print_header "Resource Usage for $QDRANT_CONTAINER_NAME"
    docker stats "$QDRANT_CONTAINER_NAME" --no-stream
}

# 10. Inspect the container's environment variables.
inspect_env_vars() {
    print_header "Environment Variables for $QDRANT_CONTAINER_NAME"
    docker inspect "$QDRANT_CONTAINER_NAME" --format '{{range .Config.Env}}{{println .}}{{end}}'
}

# 11. Check the status of the Qdrant cluster.
check_cluster_status() {
    print_header "Qdrant Cluster Status"
    curl "http://$QDRANT_HOST/cluster"
    echo
}

# 12. Run a comprehensive batch test and generate a report.
run_comprehensive_check() {
    print_header "Comprehensive System Report"
    echo "Running all key diagnostic checks..."

    check_container_status
    check_health_endpoint
    check_port_conflicts
    check_volume_permissions
    check_resource_usage
    check_cluster_status

    echo
    echo "--- Comprehensive Report End ---"
    echo
}


# --- Main Logic ---

# Function to display the main menu.
show_menu() {
    echo
    echo "=============================================="
    echo "      Qdrant Docker Troubleshooter Menu"
    echo "=============================================="
    echo "  0) Run Comprehensive Health Report"
    echo "----------------------------------------------"
    echo "  Container Health:"
    echo "    1) Check Container Status"
    echo "    2) View Container Logs"
    echo "  API & Connectivity:"
    echo "    3) Check for Port Conflicts"
    echo "    4) Check API Health Endpoint"
    echo "  Data & Persistence:"
    echo "    5) Check Data Volume Permissions (Dynamic)"
    echo "    6) Create Collection Snapshot (Interactive)"
    echo "    7) List Collection Snapshots (Interactive)"
    echo "    8) Restore Collection from Snapshot (Interactive)"
    echo "  Performance & Configuration:"
    echo "    9) Check Container Resource Usage"
    echo "    10) Inspect Environment Variables"
    echo "  Clustering:"
    echo "    11) Check Cluster Status"
    echo
    echo "  q) Quit"
    echo "=============================================="
    echo
}

# Function to read user choice and execute the corresponding command.
read_choice() {
    local choice
    read -p "Enter your choice [0-11 or q]: " choice
    case $choice in
        0) run_comprehensive_check ;;
        1) check_container_status ;;
        2) view_container_logs ;;
        3) check_port_conflicts ;;
        4) check_health_endpoint ;;
        5) check_volume_permissions ;;
        6) create_snapshot ;;
        7) list_snapshots ;;
        8) restore_snapshot ;;
        9) check_resource_usage ;;
        10) inspect_env_vars ;;
        11) check_cluster_status ;;
        q|Q) echo "Exiting."; exit 0 ;;
        *) echo "Invalid option. Please try again." ;;
    esac
}

# --- Script Execution ---

# If a command is passed as an argument, run it directly.
# Note: Snapshot commands will now be interactive.
if [ -n "$1" ]; then
    case $1 in
        report|all) run_comprehensive_check ;;
        status) check_container_status ;;
        logs) view_container_logs ;;
        ports) check_port_conflicts ;;
        health) check_health_endpoint ;;
        permissions) check_volume_permissions ;;
        create_snapshot) create_snapshot ;;
        list_snapshots) list_snapshots ;;
        restore_snapshot) restore_snapshot ;;
        resources) check_resource_usage ;;
        env) inspect_env_vars ;;
        cluster) check_cluster_status ;;
        *) echo "Invalid command: $1" >&2; exit 1 ;;
    esac
    exit 0
fi

# If no command is passed, show the interactive menu.
while true
do
    show_menu
    read_choice
done