ping_service/manager/dyfi.go

package main

import (
	"crypto/tls"
	"fmt"
	"io"
	"net"
	"net/http"
	"strings"
	"time"
)

// parseDyfiResponse interprets dy.fi update response codes
func parseDyfiResponse(response string) (string, string) {
	errorCodes := map[string]string{
		"abuse":   "The service feels YOU are ABUSING it!",
		"badauth": "Authentication failed",
		"nohost":  "No hostname given for update, or hostname not yours",
		"notfqdn": "The given hostname is not a valid FQDN",
		"badip":   "The client IP address is not valid or permitted",
		"dnserr":  "Update failed due to a problem at dy.fi",
		"good":    "The update was processed successfully",
		"nochg":   "The successful update did not cause a DNS data change",
	}

	// Response format: "code" or "code ipaddress"
	parts := strings.Fields(response)
	if len(parts) == 0 {
		return "", "Empty response from dy.fi"
	}

	code := parts[0]
	description, exists := errorCodes[code]
	if !exists {
		description = response
	}

	return code, description
}

// getCurrentDNSIP looks up the current IP address the hostname points to
func getCurrentDNSIP(hostname string) (string, error) {
	ips, err := net.LookupIP(hostname)
	if err != nil {
		return "", err
	}

	// Return first IPv4 address
	for _, ip := range ips {
		if ipv4 := ip.To4(); ipv4 != nil {
			return ipv4.String(), nil
		}
	}

	return "", fmt.Errorf("no IPv4 address found for %s", hostname)
}

// getOurPublicIP attempts to determine our own public IP address
func getOurPublicIP() (string, error) {
	// Try to get our public IP from a reliable source
	services := []string{
		"https://api.ipify.org",
		"https://checkip.amazonaws.com",
		"https://icanhazip.com",
	}

	client := &http.Client{Timeout: 5 * time.Second}

	for _, service := range services {
		resp, err := client.Get(service)
		if err != nil {
			continue
		}
		defer resp.Body.Close()

		body, err := io.ReadAll(resp.Body)
		if err != nil {
			continue
		}

		ip := strings.TrimSpace(string(body))
		// Validate it's an IP
		if net.ParseIP(ip) != nil {
			return ip, nil
		}
	}

	return "", fmt.Errorf("failed to determine public IP")
}

// checkManagerHealthAt checks if a manager instance is responding at the given IP
func checkManagerHealthAt(ip string, port string) bool {
	// Try HTTPS first, then HTTP
	schemes := []string{"https", "http"}

	for _, scheme := range schemes {
		url := fmt.Sprintf("%s://%s:%s/health", scheme, ip, port)

		// Create client with relaxed TLS verification (self-signed certs)
		transport := &http.Transport{
			TLSClientConfig: &tls.Config{InsecureSkipVerify: true},
		}
		client := &http.Client{
			Timeout:   5 * time.Second,
			Transport: transport,
		}

		resp, err := client.Get(url)
		if err != nil {
			continue
		}
		resp.Body.Close()

		// Consider 200 OK as healthy
		if resp.StatusCode == 200 {
			return true
		}
	}

	return false
}

func startDyfiUpdater(hostname, username, password, managerPort string) {
	if hostname == "" || username == "" || password == "" {
		return
	}

	logger.Info("Starting dy.fi updater for %s", hostname)
	logger.Info("Update interval: 20 hours (dy.fi requires update at least every 7 days)")
	logger.Info("Multi-instance mode: will only update if current pointer is down (failover)")

	// Default to 443 if not specified
	if managerPort == "" {
		managerPort = "443"
	}

	update := func() {
		// Step 1: Check where DNS currently points
		currentIP, err := getCurrentDNSIP(hostname)
		if err != nil {
			logger.Warn("dy.fi: failed to lookup current DNS for %s: %v", hostname, err)
			logger.Info("dy.fi: assuming initial state, proceeding with update")
			// Continue to update since we can't verify
		} else {
			logger.Info("dy.fi: %s currently points to %s", hostname, currentIP)

			// Step 2: Get our own public IP
			ourIP, err := getOurPublicIP()
			if err != nil {
				logger.Warn("dy.fi: failed to determine our public IP: %v", err)
				logger.Info("dy.fi: proceeding with cautious update")
			} else {
				logger.Info("dy.fi: our public IP is %s", ourIP)

				// Step 3: Decide what to do based on current state
				if currentIP == ourIP {
					// We are the active instance - normal refresh
					logger.Info("dy.fi: we are the ACTIVE instance, performing normal refresh")
				} else {
					// DNS points to a different IP - check if that instance is healthy
					logger.Info("dy.fi: DNS points to different IP, checking health of instance at %s", currentIP)

					if checkManagerHealthAt(currentIP, managerPort) {
						// Another instance is healthy and serving - we are standby
						logger.Info("dy.fi: manager instance at %s is HEALTHY - we are STANDBY", currentIP)
						logger.Info("dy.fi: skipping update to avoid DNS pointer conflict")
						return // Don't update, stay in standby mode
					} else {
						// The instance at current IP is not responding - failover!
						logger.Warn("dy.fi: manager instance at %s is NOT responding", currentIP)
						logger.Info("dy.fi: initiating FAILOVER - taking over DNS pointer")
					}
				}
			}
		}

		// If we reach here, we should perform the update
		url := fmt.Sprintf("https://www.dy.fi/nic/update?hostname=%s", hostname)
		req, err := http.NewRequest("GET", url, nil)
		if err != nil {
			logger.Error("dy.fi: failed to create request: %v", err)
			return
		}

		req.SetBasicAuth(username, password)
		req.Header.Set("User-Agent", "PingServiceManager/1.0")

		client := &http.Client{Timeout: 30 * time.Second}
		resp, err := client.Do(req)
		if err != nil {
			logger.Error("dy.fi: update request failed: %v", err)
			return
		}
		defer resp.Body.Close()

		// Read response body
		body, err := io.ReadAll(resp.Body)
		if err != nil {
			logger.Error("dy.fi: failed to read response: %v", err)
			return
		}

		responseText := strings.TrimSpace(string(body))

		// Check HTTP status
		if resp.StatusCode != 200 {
			logger.Error("dy.fi: HTTP error %d: %s", resp.StatusCode, responseText)
			return
		}

		// Check Content-Type
		contentType := resp.Header.Get("Content-Type")
		if !strings.HasPrefix(strings.ToLower(contentType), "text/plain") {
			logger.Warn("dy.fi: unexpected content-type: %s", contentType)
		}

		// Parse dy.fi response
		code, description := parseDyfiResponse(responseText)

		switch code {
		case "good":
			// Extract IP if present
			parts := strings.Fields(responseText)
			if len(parts) > 1 {
				logger.Info("dy.fi: ✅ SUCCESSFUL UPDATE for %s - DNS now points to %s", hostname, parts[1])
				logger.Info("dy.fi: we are now the ACTIVE instance")
			} else {
				logger.Info("dy.fi: ✅ SUCCESSFUL UPDATE for %s", hostname)
				logger.Info("dy.fi: we are now the ACTIVE instance")
			}
		case "nochg":
			logger.Info("dy.fi: ✅ SUCCESSFUL REFRESH for %s (no DNS change, we remain ACTIVE)", hostname)
		case "abuse":
			logger.Error("dy.fi: ABUSE DETECTED! The service is denying our requests for %s", hostname)
			logger.Error("dy.fi: This usually means the update script is running too frequently")
			logger.Error("dy.fi: Stopping dy.fi updater to prevent further abuse flags")
			return // Stop updating if abuse is detected
		case "badauth":
			logger.Error("dy.fi: authentication failed for %s - check username/password", hostname)
		case "nohost":
			logger.Error("dy.fi: hostname %s not found or not owned by this account", hostname)
		case "notfqdn":
			logger.Error("dy.fi: %s is not a valid FQDN", hostname)
		case "badip":
			logger.Error("dy.fi: client IP address is not valid or permitted", hostname)
		case "dnserr":
			logger.Error("dy.fi: DNS update failed due to a problem at dy.fi for %s", hostname)
		default:
			logger.Warn("dy.fi: unknown response for %s: %s (%s)", hostname, responseText, description)
		}
	}

	// Update immediately on start
	update()

	// Update every 20 hours (dy.fi deletes inactive domains after 7 days)
	go func() {
		ticker := time.NewTicker(20 * time.Hour)
		defer ticker.Stop()
		for range ticker.C {
			update()
		}
	}()
}