Send systemd notifications properly

There's a bug in `systemd-notify` such that the command sends the message and then exits. Since its lifetime is so short, systemd doesn't have time to do the housekeeping to figure out which service unit to associate it with. The mitigation is to just open the socket and send the message directly ourselves, since our process is long-lived.

See systemd/systemd#2737

Fixes #16

lib/abalone/watchdog.rb

@@ -3,6 +3,11 @@ def initialize(period, options)
     raise 'Watchdog needs a port to be set' unless options.include? :port
     raise 'Watchdog needs a logger to be set' unless options.include? :logger
     logger = options[:logger]
+    pid    = Process.pid
+    socket = Socket.new(Socket::AF_UNIX, Socket::SOCK_DGRAM, 0).tap do |socket|
+      socket.connect(Socket.pack_sockaddr_un(ENV['NOTIFY_SOCKET']))
+      socket.close_on_exec = true
+    end
 
     Thread.new do
       require 'net/http'
@@ -17,7 +22,8 @@ def initialize(period, options)
           http.start
           http.request_get('/heartbeat/ping') do |res|
             logger.debug "Heartbeat response: #{res.read_body}"
-            system('systemd-notify WATCHDOG=1')
+            #system('systemd-notify WATCHDOG=1')
+            socket.write("WATCHDOG=1\nMAINPID=#{pid}")
           end
         rescue => e
           logger.warn 'Abalone service failed heartbeat check!'