#!/bin/bash

# Health check monitoring script for DocFast
# Runs every 5 minutes via cron

HEALTH_URL="https://docfast.dev/health"
LOG_FILE="/var/log/docfast-healthcheck.log"
DOWN_MARKER="/tmp/docfast-down"
STATE_FILE="/tmp/docfast-healthcheck-state"
MAX_LOG_LINES=1000

# Initialize state file if it doesn't exist
if [ ! -f "$STATE_FILE" ]; then
    echo "0" > "$STATE_FILE"
fi

# Function to rotate log if it exceeds max lines
rotate_log() {
    if [ -f "$LOG_FILE" ]; then
        local line_count=$(wc -l < "$LOG_FILE")
        if [ "$line_count" -gt "$MAX_LOG_LINES" ]; then
            # Keep only the last 1000 lines
            tail -n "$MAX_LOG_LINES" "$LOG_FILE" > "${LOG_FILE}.tmp"
            mv "${LOG_FILE}.tmp" "$LOG_FILE"
        fi
    fi
}

# Function to log messages
log_message() {
    local msg="$1"
    echo "[$(date '+%Y-%m-%d %H:%M:%S')] $msg" >> "$LOG_FILE"
    rotate_log
}

# Perform health check
check_health() {
    response=$(curl -s -w "\n%{http_code}" "$HEALTH_URL" 2>/dev/null)
    http_code=$(echo "$response" | tail -n1)
    body=$(echo "$response" | head -n-1)
    
    # Check if HTTP 200 and contains "status":"ok"
    if [ "$http_code" = "200" ] && echo "$body" | grep -q '"status":"ok"'; then
        return 0  # Success
    else
        return 1  # Failure
    fi
}

# Get current failure count
get_failure_count() {
    cat "$STATE_FILE"
}

# Increment failure count
increment_failure() {
    local count=$(get_failure_count)
    echo $((count + 1)) > "$STATE_FILE"
}

# Reset failure count
reset_failure() {
    echo "0" > "$STATE_FILE"
}

# Main logic
if check_health; then
    # Health check passed
    failure_count=$(get_failure_count)
    if [ "$failure_count" -gt 0 ]; then
        log_message "✓ Service recovered after $failure_count failure(s)"
        if [ -f "$DOWN_MARKER" ]; then
            rm -f "$DOWN_MARKER"
            log_message "Removed downtime marker"
        fi
    fi
    reset_failure
else
    # Health check failed
    increment_failure
    failure_count=$(get_failure_count)
    log_message "✗ Health check failed (attempt $failure_count)"
    
    # After 2 consecutive failures, mark as down
    if [ "$failure_count" -ge 2 ]; then
        if [ ! -f "$DOWN_MARKER" ]; then
            touch "$DOWN_MARKER"
            log_message "⚠ DOWNTIME DETECTED - Marker file created"
        fi
    fi
fi

exit 0