Class: Messhy::HealthChecker

Inherits:
Object
  • Object
show all
Includes:
WireguardStatusParser
Defined in:
lib/messhy/health_checker.rb

Constant Summary collapse

HANDSHAKE_STALENESS_LIMIT =

seconds

180

Constants included from WireguardStatusParser

WireguardStatusParser::TIME_UNITS_IN_SECONDS

Instance Attribute Summary collapse

Instance Method Summary collapse

Methods included from WireguardStatusParser

extract_allowed_ips, extract_endpoint, extract_handshake_time, extract_peer_block, extract_transfer_stats, integer_token?, line_value, parse_handshake_seconds

Constructor Details

#initialize(config) ⇒ HealthChecker

Returns a new instance of HealthChecker.



14
15
16
17
 
# File 'lib/messhy/health_checker.rb', line 14

def initialize(config)
  @config = config
  @ssh_executor = SSHExecutor.new(config)
end

Instance Attribute Details

#config ⇒ Object (readonly)

Returns the value of attribute config.



12
13
14
 
# File 'lib/messhy/health_checker.rb', line 12

def config
  @config
end

Instance Method Details

#ping_node(node_or_ip) ⇒ Object 



64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
 
# File 'lib/messhy/health_checker.rb', line 64

def ping_node(node_or_ip)
  # Determine if input is node name or IP
  target_node = nil
  target_ip = nil

  if node_or_ip =~ /^\d+\.\d+\.\d+\.\d+$/
    # It's an IP
    target_ip = node_or_ip
    target_node = config.nodes.find { |_, cfg| cfg['private_ip'] == target_ip }&.first
  else
    # It's a node name
    target_node = node_or_ip
    node_config = config.node_config(target_node)
    target_ip = node_config['private_ip'] if node_config
  end

  unless target_ip
    puts "Node or IP not found: #{node_or_ip}"
    return
  end

  puts "Pinging #{target_node || target_ip} (#{target_ip})..."

  # Try pinging from each other node
  config.each_node do |source_node, _|
    next if source_node == target_node # Skip pinging self

    success = @ssh_executor.ping_node_from(source_node, target_ip)
    status = success ? '✓' : '✗'
    puts "  #{status} from #{source_node}"
  end
end

#show_dns_status ⇒ Object 



162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
 
# File 'lib/messhy/health_checker.rb', line 162

def show_dns_status
  puts '==> Mesh DNS Status'
  puts "Domain: #{config.dns_domain}"
  puts "Servers: #{config.dns_server_nodes.join(', ')}"
  puts

  config.dns_server_nodes.each do |node_name|
    node_config = config.node_config(node_name)
    next unless node_config

    label = node_config['label']
    label_display = label.to_s.strip.empty? ? '' : " (#{label})"

    begin
      @ssh_executor.execute_on_node(node_name) do
        service = capture(:systemctl, 'is-active', 'dnsmasq', raise_on_non_zero_exit: false).strip
        messhy_records = capture(:bash, '-c',
                                 "sudo awk 'BEGIN{c=0} /^address=\\//{c++} END{print c}' " \
                                 '/etc/dnsmasq.d/messhy.conf 2>/dev/null || true').strip
        ap_records = capture(:bash, '-c',
                             "sudo awk 'BEGIN{c=0} /^address=\\//{c++} END{print c}' " \
                             '/etc/dnsmasq.d/active_postgres.conf 2>/dev/null || true').strip

        status_icon = service == 'active' ? '✓' : '✗'
        puts "#{status_icon} #{node_name} (#{node_config['private_ip']})#{label_display} - dnsmasq #{service}"
        puts "  └─ records: messhy=#{messhy_records.to_i} active_postgres=#{ap_records.to_i}"
      end
    rescue StandardError => e
      puts "✗ #{node_name} (#{node_config['private_ip']})#{label_display} - DNS check failed: #{e.message}"
    end
  end

  puts
end

#show_node_status(node_name) ⇒ Object 



34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
 
# File 'lib/messhy/health_checker.rb', line 34

def show_node_status(node_name)
  node_config = config.node_config(node_name)
  label = node_config['label']
  label_display = label.to_s.strip.empty? ? '' : " (#{label})"

  begin
    status = @ssh_executor.get_wireguard_status(node_name)

    # Parse status output
    peers = status.scan(/peer: (.+?)$/).flatten

    if peers.any?
      puts "✓ #{node_name} (#{node_config['private_ip']})#{label_display} - connected to #{peers.size} peers"

      # Show basic peer info
      status.split('peer:').drop(1).each do |peer_block|
        endpoint = extract_endpoint(peer_block)
        next unless endpoint

        stats = extract_transfer_stats(peer_block)
        puts "  └─ Peer: #{endpoint} - #{stats[:received]} rx, #{stats[:sent]} tx"
      end
    else
      puts "✗ #{node_name} (#{node_config['private_ip']})#{label_display} - 0 peers (DOWN)"
    end
  rescue StandardError => e
    puts "✗ #{node_name} (#{node_config['private_ip']})#{label_display} - Error: #{e.message}"
  end
end

#show_stats(node: nil) ⇒ Object 



151
152
153
154
155
156
157
158
159
160
 
# File 'lib/messhy/health_checker.rb', line 151

def show_stats(node: nil)
  if node
    show_node_stats(node)
  else
    config.each_node do |node_name, _|
      show_node_stats(node_name)
      puts
    end
  end
end

#show_status ⇒ Object 



19
20
21
22
23
24
25
26
27
28
29
30
31
32
 
# File 'lib/messhy/health_checker.rb', line 19

def show_status
  puts '==> WireGuard Mesh Status'
  puts "Environment: #{config.environment}"
  puts

  config.each_node do |node_name, _node_config|
    show_node_status(node_name)
    puts
  end

  show_dns_status if config.dns_enabled?

  show_latency_matrix
end

#test_all ⇒ Object 



97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
 
# File 'lib/messhy/health_checker.rb', line 97

def test_all
  puts '==> Testing mesh connectivity...'
  puts
  puts 'Note: This test may take a while. WireGuard status shows all peers connected.'
  puts

  all_ok = true
  tested_pairs = Set.new
  test_count = 0
  total_tests = config.node_names.size * (config.node_names.size - 1) / 2

  status_cache = {}
  config.each_node do |source_name, _source_config|
    config.each_node do |target_name, target_config|
      next if source_name == target_name

      pair_key = [source_name, target_name].sort.join('-')
      next if tested_pairs.include?(pair_key)

      tested_pairs.add(pair_key)
      test_count += 1
      target_ip = target_config['private_ip']

      print "[#{test_count}/#{total_tests}] Testing #{source_name} → #{target_name} (#{target_ip})... "
      $stdout.flush

      success = false
      begin
        Timeout.timeout(3) do
          success = @ssh_executor.ping_node_from(source_name, target_ip) ||
                    @ssh_executor.test_tcp_connectivity(source_name, target_ip, 22)
        end
      rescue StandardError
        success = false
      end

      if success
        puts '✓'
      elsif handshake_recent?(source_name, target_config['private_ip'], status_cache)
        puts '✓ (handshake)'
      else
        puts '✗ (ICMP/TCP may be blocked, and no recent WireGuard handshake)'
        all_ok = false
      end
      $stdout.flush
    end
  end

  puts
  puts 'Note: When ICMP/TCP probes fail, we fall back to recent WireGuard handshakes.'
  puts 'If a pair still reports a failure, there has been no recent handshake—check UDP 51820 and ' \
       'keepalive/route settings.'
end