merged PingRotate into PingScheduler; fixed save/load of cache to/from file; started...
[pingcheck] / src / host / pingscheduler.cpp
1 /*
2 The software in this package is distributed under the GNU General
3 Public License version 2 (with a special exception described below).
4
5 A copy of GNU General Public License (GPL) is included in this distribution,
6 in the file COPYING.GPL.
7
8 As a special exception, if other files instantiate templates or use macros
9 or inline functions from this file, or you compile this file and link it
10 with other works to produce a work based on this file, this file
11 does not by itself cause the resulting work to be covered
12 by the GNU General Public License.
13
14 However the source code for this file must still be made available
15 in accordance with section (3) of the GNU General Public License.
16
17 This exception does not invalidate any other reasons why a work based
18 on this file might be covered by the GNU General Public License.
19 */
20 #include "host/pingscheduler.h"
21
22 #include <iostream>
23 #include <limits>
24
25 #include <boost/bind.hpp>
26 #include <boost/foreach.hpp>
27
28 #include <logfunc.hpp>
29
30 #include "boost_assert_handler.h"
31 #include "host/pingerfactory.h"
32 #include "dns/dnsmaster.h"
33 #include "icmp/icmppinger.h"
34 #include "link/linkstatus.h"
35
36 using namespace std;
37 using boost::asio::io_service;
38 using boost::bind;
39 using boost::date_time::time_resolution_traits_adapted64_impl;
40 using boost::posix_time::microsec_clock;
41 using boost::posix_time::ptime;
42 using boost::posix_time::seconds;
43 using boost::shared_ptr;
44 using I2n::Logger::GlobalLogger;
45
46 //-----------------------------------------------------------------------------
47 // PingScheduler
48 //-----------------------------------------------------------------------------
49
50 /**
51  * @brief Parameterized constructor.
52  *
53  * @param io_serv The one @c io_serv object that controls async processing
54  * @param network_interface The name of the network interface originating the pings.
55  * @param destination_address The remote address to ping.
56  * @param destination_port The remote port to ping.
57  * @param ping_protocol_list A list of protocols to use.
58  * @param ping_interval_in_sec Amount of time between each ping.
59  * @param ping_fail_percentage_limit Maximum amount of pings that can fail.
60  * @param link_analyzer The object to monitor the link status.
61  * @param first_delay Delay in seconds from start_pinging to first ping attempt
62  */
63 PingScheduler::PingScheduler(
64         const IoServiceItem io_serv,
65         const string &network_interface,
66         const string &destination_address,
67         const uint16_t destination_port,
68         const PingProtocolList &ping_protocol_list,
69         const long ping_interval_in_sec,
70         const int ping_fail_percentage_limit,
71         const int ping_reply_timeout,
72         LinkStatusItem link_analyzer,
73         const int first_delay
74
75 ) :
76     IoService( io_serv ),
77     NetworkInterfaceName( network_interface ),
78     DestinationAddress( destination_address ),
79     DestinationPort( destination_port ),
80     Protocols( ping_protocol_list ),
81     ProtocolIter(),
82     PingIntervalInSec( ping_interval_in_sec ),
83     FirstDelay( first_delay ),
84     NextPingTimer( *io_serv ),
85     TimeSentLastPing( microsec_clock::universal_time() ),
86     PingReplyTimeout( ping_reply_timeout ),
87     HostAnalyzer( destination_address, ping_fail_percentage_limit, link_analyzer ),
88     Resolver(),
89     Ping(),
90     WantToPing( false ),
91     LogPrefix(),
92     ContinueOnOutdatedIps( false );
93 {
94     BOOST_ASSERT( !network_interface.empty() );
95     BOOST_ASSERT( !destination_address.empty() );
96     BOOST_ASSERT( ( 0 < destination_port ) &&
97                   ( destination_port < numeric_limits<uint16_t>::max() ) );
98     BOOST_ASSERT( 0 < ping_interval_in_sec );
99     BOOST_ASSERT( (0 <= ping_fail_percentage_limit) &&
100                   ( ping_fail_percentage_limit <= 100) );
101
102     update_log_prefix();
103
104     init_ping_protocol();
105 }
106
107 /**
108  * @brief Destructor.
109  */
110 PingScheduler::~PingScheduler()
111 {
112 }
113
114 void PingScheduler::stop_pinging()
115 {
116     // stop pinger, which will probably call ping_done_handler --> re-new NextPingTimer
117     GlobalLogger.debug() << "scheduler: stop pinging" << endl;
118     Ping->stop_pinging();
119     Resolver->cancel_resolve();
120
121     // now cancel the own timer
122     GlobalLogger.debug() << "scheduler: cancel timer" << endl;
123     NextPingTimer.cancel();
124 }
125
126 /**
127  * @brief Start into infinite loop of calls to ping
128  *
129  * Does not start yet but set NextPingTimer (possibly to 0), so action starts
130  *   when io_service is started
131  */
132 void PingScheduler::start_pinging()
133 {
134     if ( FirstDelay > 0 )
135         GlobalLogger.info() << "Delaying first ping by " << FirstDelay << "s";
136     else
137         GlobalLogger.info() << "Schedule ping as soon as possible";
138
139     (void) NextPingTimer.expires_from_now( seconds( FirstDelay ) );
140     NextPingTimer.async_wait( bind( &PingScheduler::ping, this,
141                                           boost::asio::placeholders::error ) );
142 }
143
144
145 /**
146  * @brief call Ping::ping and schedule a call to ping_done_handler when finished
147  */
148 void PingScheduler::ping(const boost::system::error_code &error)
149 {
150     if ( error )
151     {   // get here, e.g. by NextPingTimer.cancel in stop_pinging
152         if ( error ==  boost::asio::error::operation_aborted )
153             GlobalLogger.error() << "Timer for ping was cancelled! "
154                                  << "Stopping" << endl;
155         else
156             GlobalLogger.error() << "Received error " << error
157                                  << " waiting for ping! Stopping"
158                                  << endl;
159         return;
160     }
161
162     // ping as soon as dns is ready
163     WantToPing = true;
164     try_to_ping();
165 }
166
167
168 void PingScheduler::try_to_ping()
169 {
170     if ( !WantToPing )
171     {
172         GlobalLogger.info() << "PingScheduler: not pinging (not requested to)";
173         return;
174     }
175     else if ( Resolver && Resolver->is_resolving() )
176     {
177         GlobalLogger.info() << "PingScheduler: not pinging (DNS not finished)";
178         return;
179     }
180     else if ( !Resolver )
181         // should not happen, but check anyway
182         GlobalLogger.warning() << LogPrefix << "Have no resolver!";
183
184     GlobalLogger.info() << "PingScheduler: start ping";
185     WantToPing = false;
186
187     HostAddress ip = Resolver->get_next_ip();
188     if ( !ip.is_valid() && ContinueOnOutdatedIps)
189     {
190         GlobalLogger.info() << LogPrefix << "Checking for outdated IPs";
191         bool check_up_to_date = false;
192         ip = Resolver->get_next_ip(check_up_to_date);
193     }
194     if ( ip.is_valid() )
195         Ping->ping(
196                 Resolver->get_next_ip().get_ip(),
197                 DestinationPort,
198                 boost::bind(&PingScheduler::ping_done_handler, this, _1)
199         );
200     else
201     {   // should not happen
202         GlobalLogger.error() << LogPrefix << "No IP to ping "
203                              << "-- this should not have happened!!";
204         WantToPing = true;
205         if ( !Resolver.is_resolving() )
206             start_resolving_ping_address();
207     }
208 }
209
210
211 //------------------------------------------------------------------------------
212 // Post Processing of Ping result
213 //------------------------------------------------------------------------------
214
215 /**
216  * @brief called when Ping::ping is done; calls functions to update
217  *   statistics, ping interval and elapsed time;
218  *   schedules a call to ping, thereby closing the loop
219  */
220 void PingScheduler::ping_done_handler( const bool ping_success )
221 {
222     // post-processing
223     // You must call these 3 methods exactly in this order
224     // TODO Fix this method, once it has a semantic dependency with the
225     // update_ping_statistics method, because it depends on the PingAnalyzer
226     // statistics to update the exceeded_ping_failed_limit
227     HostAnalyzer.update_ping_statistics( ping_success );
228     update_ping_interval();
229     update_ping_elapsed_time();
230
231     // get next protocol, possibly start resolving IPs
232     update_ping_protocol();
233
234     // schedule next ping
235     (void) NextPingTimer.expires_from_now( seconds( PingIntervalInSec ) );
236     NextPingTimer.async_wait( bind( &PingScheduler::ping, this,
237                                          boost::asio::placeholders::error ) );
238 }
239
240 void PingScheduler::update_ping_interval()
241 {
242     // have to ping more often?
243     if ( HostAnalyzer.exceeded_ping_failed_limit() )
244     {
245         PingIntervalInSec.speed_up();
246
247         GlobalLogger.debug() << "- Speeding up ping interval to: " << PingIntervalInSec << "s"
248                 << endl;
249     }
250     else
251     {
252         PingIntervalInSec.back_to_original();
253
254         GlobalLogger.debug() << "- Stick to the original ping interval: " << PingIntervalInSec << "s"
255                 << endl;
256     }
257 }
258
259 void PingScheduler::update_ping_elapsed_time()
260 {
261     ptime now = microsec_clock::universal_time();
262     time_resolution_traits_adapted64_impl::int_type elapsed_time_in_sec =
263             (now - TimeSentLastPing).total_seconds();
264     GlobalLogger.debug() << "- Time elapsed since last ping: " << elapsed_time_in_sec << "s" << endl;
265
266     TimeSentLastPing = microsec_clock::universal_time();
267 }
268
269
270 //------------------------------------------------------------------------------
271 // Ping Protocol Rotation
272 //------------------------------------------------------------------------------
273
274 void PingScheduler::init_ping_protocol()
275 {
276     ProtocolIter = Protocols.end();
277     get_next_ping_protocol();
278 }
279
280 void PingScheduler::update_ping_protocol()
281 {
282     if ( can_change_ping_protocol() )
283     {
284         get_next_ping_protocol();
285     }
286 }
287
288 void PingScheduler::get_next_ping_protocol()
289 {
290     ++ProtocolIter;
291     if (ProtocolIter == Protocols.end())
292         ProtocolIter = Protocols.begin();
293     PingProtocol ping_protocol = *ProtocolIter;
294     // --> ProtocolIter still points to currently used protocol which is 
295     //     required in dns_resolve_callback
296
297     if (Ping)
298         Ping->stop_pinging();
299
300     Ping = PingerFactory::createPinger(ping_protocol, IoService,
301                                        NetworkInterfaceName, PingReplyTimeout);
302
303     update_dns_resolver( ping_protocol );
304 }
305
306 bool PingScheduler::can_change_ping_protocol() const
307 {
308     // TODO can_change_ping_protocol() and get_next_ping_protocol() may be implemented in a Algorithm
309     // class that can be exchanged in this class to provide an algorithm neutral class
310     return true;
311 }
312
313 //------------------------------------------------------------------------------
314 // DNS host name resolution
315 //------------------------------------------------------------------------------
316
317 // show "!" after host name if running on outdated IPs
318 void update_log_prefix()
319 {
320     std::stringstream temp;
321     temp << "PS(" << DestinationAddress;
322     if (ContinueOnOutdatedIps)
323         temp << "!";
324     temp << "): ";
325     LogPrefix = temp.str();
326 }
327
328 void PingScheduler::update_dns_resolver( PingProtocol current_protocol )
329 {
330     if (Resolver && Resolver->is_resolving())
331     {
332         GlobalLogger.warning() << "Resolver still seems to be resolving "
333                                << "--> cancel!";
334         Resolver->cancel_resolve();
335     }
336
337     // DNS master caches created resolvers and resolved IPs, so this will
338     // probably just return an existing resolver with already resolved IPs for
339     // requested protocol ( ICMP/TCP is ignored, only IPv4/v6 is important)
340     Resolver = DnsMaster::get_instance()->get_resolver_for(DestinationAddress,
341                                                            current_protocol);
342     // start resolving if no ips available
343     if ( Resolver->have_up_to_date_ip() )
344     {
345         if (!Resolver->is_resolving())
346             GlobalLogger.warning() << "PingScheduler: have up to date IPs but "
347                 << "resolver seems to be resolving all the same... "
348                 << "Start pinging anyway!";
349         try_to_ping();
350     }
351     else
352         start_resolving_ping_address();
353 }
354
355 void PingScheduler::start_resolving_ping_address()
356 {
357     Resolver->async_resolve( boost::bind(&PingScheduler::dns_resolve_callback,
358                                           this, _1, _2) );
359 }
360
361 void PingScheduler::dns_resolve_callback(const bool was_success,
362                                          const int cname_count)
363 {
364     GlobalLogger.info() << "PingScheduler: dns resolution finished "
365                         << "with success = " << was_success << " "
366                         << "and cname_count = " << cname_count;
367
368     // TODO this is too simple, but need to think more about how to update here!
369     // (may have to switch back some time to resolver for original host or so
370     ContinueOnOutdatedIps = true;
371     update_log_prefix();
372
373     if ( was_success )
374     {
375         HostAnalyzer.set_resolved_ip_count( Resolver->get_resolved_ip_count());
376         try_to_ping();
377     }
378     else
379     {   // host name resolution failed; try again bypassing first outdated CNAME
380
381         std::string skip_host = Resolver->get_skip_cname();
382
383         if (skip_host.empty())
384         {
385             GlobalLogger.notice() << LogPrefix << "DNS failed, "
386                 << "try anyway with cached data";
387             HostAnalyzer.set_resolved_ip_count(0);
388             try_to_ping();
389         }
390         else
391         {
392             GlobalLogger.notice() << LogPrefix << "DNS failed, "
393                 << "try again skipping a CNAME and resolving "
394                 << skip_host << " directly";
395             Resolver = DnsMaster::get_instance()
396                                    ->get_resolver_for(skip_host, *ProtocolIter);
397             // the original resolver is still alive and cached by DnsMaster
398             //   and counting down the time to re-try on its own
399             start_resolving_ping_address();
400         }
401     }
402 }