2 The software in this package is distributed under the GNU General
3 Public License version 2 (with a special exception described below).
5 A copy of GNU General Public License (GPL) is included in this distribution,
6 in the file COPYING.GPL.
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.
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.
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.
20 #include "host/pingscheduler.h"
25 #include <boost/bind.hpp>
26 #include <boost/foreach.hpp>
28 #include <logfunc.hpp>
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"
37 using boost::asio::io_service;
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;
46 //-----------------------------------------------------------------------------
48 //-----------------------------------------------------------------------------
51 * @brief Parameterized constructor.
53 * @param io_serv The one @c io_serv object that controls async processing
54 * @param network_interface The name of the network interface sending 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
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,
77 NetworkInterfaceName( network_interface ),
78 DestinationAddress( destination_address ),
79 DestinationPort( destination_port ),
80 Protocols( ping_protocol_list ),
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,
93 ContinueOnOutdatedIps( false )
95 BOOST_ASSERT( !network_interface.empty() );
96 BOOST_ASSERT( !destination_address.empty() );
97 BOOST_ASSERT( ( 0 < destination_port ) &&
98 ( destination_port < numeric_limits<uint16_t>::max() ) );
99 BOOST_ASSERT( 0 < ping_interval_in_sec );
100 BOOST_ASSERT( (0 <= ping_fail_percentage_limit) &&
101 ( ping_fail_percentage_limit <= 100) );
105 init_ping_protocol();
111 PingScheduler::~PingScheduler()
115 void PingScheduler::stop_pinging()
117 // stop pinger and resolver
118 GlobalLogger.debug() << LogPrefix << "scheduler: stop pinging";
119 Ping->stop_pinging();
120 cancel_resolve(true);
122 // now cancel the own timer in case that pinger cancelation called callback
123 GlobalLogger.debug() << LogPrefix << "scheduler: cancel timer";
124 NextPingTimer.cancel();
128 * @brief Start into infinite loop of calls to ping
130 * Does not start yet but set NextPingTimer (possibly to 0), so action starts
131 * when io_service is started
133 void PingScheduler::start_pinging()
135 if ( FirstDelay > 0 )
136 GlobalLogger.info() << LogPrefix << "Delaying first ping by "
137 << FirstDelay << "s";
139 GlobalLogger.info() << LogPrefix << "Schedule ping as soon as possible";
141 (void) NextPingTimer.expires_from_now( seconds( FirstDelay ) );
142 NextPingTimer.async_wait( bind( &PingScheduler::ping, this,
143 boost::asio::placeholders::error ) );
148 * @brief call Ping::ping and schedule a call to ping_done_handler when finished
150 void PingScheduler::ping(const boost::system::error_code &error)
153 { // get here, e.g. by NextPingTimer.cancel in stop_pinging
154 if ( error == boost::asio::error::operation_aborted )
155 GlobalLogger.error() << LogPrefix << "Timer for ping was cancelled!"
158 GlobalLogger.error() << LogPrefix << "Received error " << error
159 << " waiting for ping! Stopping";
163 // ping as soon as dns is ready
169 void PingScheduler::ping_when_ready()
173 GlobalLogger.info() << LogPrefix << "waiting for ping request "
174 << "(should take less than " << PingIntervalInSec << "s)";
177 else if ( Resolver && Resolver->is_resolving() )
179 GlobalLogger.info() << LogPrefix << "waiting for DNS to finish";
182 else if ( !Resolver )
183 // should not happen, but check anyway
184 GlobalLogger.warning() << LogPrefix << "Have no resolver!";
186 GlobalLogger.info() << LogPrefix << "start ping";
189 // try to get an up-to-date IP (ContinueOnOutdatedIps may only be set
190 // because a CNAME was out of date -- IPs may still be current)
191 HostAddress ip = Resolver->get_next_ip();
193 if ( !ip.is_valid() )
194 { // this can happen in 2 cases: if ContinueOnOutdatedIps==true
195 // or when ip went out of date between resolve and now
196 // --> try to use outdated IP
197 GlobalLogger.info() << LogPrefix << "Checking for outdated IPs";
198 bool check_up_to_date = false;
199 ip = Resolver->get_next_ip(check_up_to_date);
201 if ( !ip.is_valid() )
202 { // still no valid IP --> should not happen
203 GlobalLogger.error() << LogPrefix << "Not even outdated IP to ping "
204 << "-- this should not have happened!! Treat like a failed ping.";
205 if (DnsMaster::get_instance()->get_resolved_ip_ttl_threshold() > 0)
206 GlobalLogger.warning() << LogPrefix << "This probably happened "
207 << "because you specified a TTL threshold > 0 but resolving"
208 << " had no effect on TTLs since external cache is only "
209 << "updated when TTL=0 is reached.";
211 // skip the ping and directly call ping_done_handler
212 ping_done_handler(false);
216 uint32_t ttl = ip.get_ttl().get_updated_value();
219 expiry = "out of date!";
222 boost::posix_time::ptime now =
223 boost::posix_time::second_clock::local_time();
224 expiry = boost::posix_time::to_simple_string(now + seconds(ttl));
227 GlobalLogger.info() << LogPrefix << "pinging IP " << ip.get_ip()
228 << " with TTL " << ttl << "s (" << expiry << ")";
229 Ping->ping( ip.get_ip(),
231 boost::bind(&PingScheduler::ping_done_handler, this, _1) );
236 //------------------------------------------------------------------------------
237 // Post Processing of Ping result
238 //------------------------------------------------------------------------------
241 * @brief called when Ping::ping is done; calls functions to update
242 * statistics, ping interval and elapsed time;
243 * schedules a call to ping, thereby closing the loop
245 void PingScheduler::ping_done_handler( const bool ping_success )
248 // You must call these 3 methods exactly in this order
249 // TODO Fix this method, once it has a semantic dependency with the
250 // update_ping_statistics method, because it depends on the PingAnalyzer
251 // statistics to update the exceeded_ping_failed_limit
252 HostAnalyzer.update_ping_statistics( ping_success );
253 update_ping_interval();
254 update_ping_elapsed_time();
257 { // reset ContinueOnOutdatedIps
258 ContinueOnOutdatedIps = false;
262 // get next protocol, possibly start resolving IPs
263 update_ping_protocol();
265 // schedule next ping
266 (void) NextPingTimer.expires_from_now( seconds( PingIntervalInSec ) );
267 NextPingTimer.async_wait( bind( &PingScheduler::ping, this,
268 boost::asio::placeholders::error ) );
271 void PingScheduler::update_ping_interval()
273 // have to ping more often?
274 if ( HostAnalyzer.exceeded_ping_failed_limit() )
276 PingIntervalInSec.speed_up();
278 GlobalLogger.debug() << LogPrefix << "- Speeding up ping interval to: "
279 << PingIntervalInSec << "s";
283 PingIntervalInSec.back_to_original();
285 GlobalLogger.debug() << LogPrefix << "- Stick to the original ping "
286 << "interval: " << PingIntervalInSec << "s";
290 void PingScheduler::update_ping_elapsed_time()
292 ptime now = microsec_clock::universal_time();
293 time_resolution_traits_adapted64_impl::int_type elapsed_time_in_sec =
294 (now - TimeSentLastPing).total_seconds();
295 GlobalLogger.debug() << LogPrefix << "- Time elapsed since last ping: "
296 << elapsed_time_in_sec << "s";
298 TimeSentLastPing = microsec_clock::universal_time();
302 //------------------------------------------------------------------------------
303 // Ping Protocol Rotation
304 //------------------------------------------------------------------------------
306 void PingScheduler::init_ping_protocol()
308 ProtocolIter = Protocols.end();
309 get_next_ping_protocol();
312 void PingScheduler::update_ping_protocol()
314 if ( can_change_ping_protocol() )
316 get_next_ping_protocol();
320 void PingScheduler::get_next_ping_protocol()
322 GlobalLogger.debug() << LogPrefix
323 << "------------------------------------------------------------------";
325 if (ProtocolIter == Protocols.end())
326 ProtocolIter = Protocols.begin();
327 PingProtocol ping_protocol = *ProtocolIter;
328 // --> ProtocolIter still points to currently used protocol which is
329 // required in dns_resolve_callback
332 Ping->stop_pinging();
334 Ping = PingerFactory::createPinger(ping_protocol, IoService,
335 NetworkInterfaceName, PingReplyTimeout);
337 update_dns_resolver( ping_protocol );
341 bool PingScheduler::can_change_ping_protocol() const
343 // TODO can_change_ping_protocol() and get_next_ping_protocol() may be
344 // implemented in a Algorithm class that can be exchanged in this class to
345 // provide an algorithm neutral class
349 //------------------------------------------------------------------------------
350 // DNS host name resolution
351 //------------------------------------------------------------------------------
353 // show "!" after host name if running on outdated IPs
354 void PingScheduler::update_log_prefix()
356 std::stringstream temp;
357 temp << "Sched(" << DestinationAddress;
358 if (ContinueOnOutdatedIps)
361 LogPrefix = temp.str();
364 void PingScheduler::update_dns_resolver( PingProtocol current_protocol )
366 if (Resolver && Resolver->is_resolving())
367 cancel_resolve(false);
369 if (ContinueOnOutdatedIps)
371 ContinueOnOutdatedIps = false;
375 // DNS master caches created resolvers and resolved IPs, so this will
376 // probably just return an existing resolver with already resolved IPs for
377 // requested protocol ( ICMP/TCP is ignored, only IPv4/v6 is important)
378 Resolver = DnsMaster::get_instance()->get_resolver_for(DestinationAddress,
381 // get number of up-to-date IPs
382 // TODO should check here, if they will be up to date in PingIntervalInSec
383 bool check_up_to_date = true;
384 int ip_count = Resolver->get_resolved_ip_count(check_up_to_date);
387 GlobalLogger.info() << LogPrefix << "Set resolved_ip_count to "
388 << ip_count << " where IPs may be outdated: " << !check_up_to_date;
389 HostAnalyzer.set_resolved_ip_count( ip_count );
391 if (Resolver->is_resolving())
392 GlobalLogger.warning() << LogPrefix << "have up to date IPs but "
393 << "resolver seems to be resolving all the same... "
394 << "Start pinging anyway!";
399 GlobalLogger.info() << LogPrefix
400 << "No up-to-date IPs --> start resolve";
401 start_resolving_ping_address();
402 // set resolved_ip_count will be called in resolve callback
406 void PingScheduler::start_resolving_ping_address()
408 Resolver->async_resolve( boost::bind(&PingScheduler::dns_resolve_callback,
412 void PingScheduler::dns_resolve_callback(const bool was_success,
413 const int recursion_count)
415 GlobalLogger.info() << LogPrefix << "dns resolution finished "
416 << "with success = " << was_success << " "
417 << "after " << recursion_count << " recursions";
421 // trust that a successfull DNS resolve means we have an IP with TTL>0
422 int ip_count = Resolver->get_resolved_ip_count(!ContinueOnOutdatedIps);
424 { // this will create trouble in HostAnalyzer
425 GlobalLogger.warning() << LogPrefix
426 << "Should not have reached this case: resolve was "
427 << "successfull but still have no IPs (up-to-date="
428 << !ContinueOnOutdatedIps << ")!";
429 if (DnsMaster::get_instance()->get_resolved_ip_ttl_threshold() > 0)
430 GlobalLogger.warning() << LogPrefix << "This probably happened "
431 << "because you specified a TTL threshold > 0 but resolving"
432 << " had no effect on TTLs since external cache is only "
433 << "updated when TTL=0 is reached.";
437 GlobalLogger.info() << LogPrefix << "Set resolved_ip_count to "
438 << ip_count << " where IPs may be outdated: "
439 << ContinueOnOutdatedIps << " --> could ping now";
440 HostAnalyzer.set_resolved_ip_count( ip_count );
445 { // host name resolution failed; try again bypassing first outdated CNAME
446 // or using cached IP
447 ContinueOnOutdatedIps = true;
450 std::string skip_host = Resolver->get_skip_cname();
452 if (skip_host.empty())
453 { // continue with IP
454 GlobalLogger.notice() << LogPrefix << "DNS failed, "
455 << "try anyway with cached data";
457 int ip_count = Resolver->get_resolved_ip_count(false);
458 GlobalLogger.info() << LogPrefix << "Set resolved_ip_count to "
459 << ip_count << " where IPs may be outdated: " << true;
460 HostAnalyzer.set_resolved_ip_count( ip_count );
465 { // have CNAME to continue
466 GlobalLogger.notice() << LogPrefix << "DNS failed, "
467 << "try again skipping a CNAME and resolving "
468 << skip_host << " directly";
470 cancel_resolve(false);
472 // now create new resolver
473 Resolver = DnsMaster::get_instance()
474 ->get_resolver_for(skip_host, *ProtocolIter);
475 start_resolving_ping_address();
481 * cancel resolver if force_cancel or if it is not resolving DestinationAddress
483 * Resolvers have a life on their own: they are cached by DnsMaster so never go
484 * out of scope and even after calling callbacks, there might still be a
485 * longterm timer active to re-try resolving.
486 * We want to cancel that long-term timer only if the Resolver is not for our
487 * real, original DestinationAddress but a CNAME, which can happen when trying
488 * to skip cnames and working on out-dated IPs
490 void PingScheduler::cancel_resolve(const bool force_cancel)
494 GlobalLogger.info() << "Cancelling resolver (forced)";
495 Resolver->cancel_resolve();
497 else if ( Resolver->get_hostname() == DestinationAddress )
498 GlobalLogger.info() << LogPrefix
499 << "Leave original resolver active in background";
502 GlobalLogger.info() << LogPrefix << "Cancel resolver for "
503 << Resolver->get_hostname() << " since is not the original "
504 << DestinationAddress;
505 Resolver->cancel_resolve();