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 no more than 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 { // Do not even have an outdated IP!
203 // This happens if have no cached IPs and resolve failed
204 GlobalLogger.info() << LogPrefix << "Not even outdated IP to ping "
205 << "-- treat like a failed ping.";
207 // skip the ping and directly call ping_done_handler
208 ping_done_handler(false);
212 uint32_t ttl = ip.get_ttl().get_updated_value();
215 expiry = "out of date!";
218 boost::posix_time::ptime now =
219 boost::posix_time::second_clock::local_time();
220 expiry = boost::posix_time::to_simple_string(now + seconds(ttl));
223 GlobalLogger.info() << LogPrefix << "pinging IP " << ip.get_ip()
224 << " with TTL " << ttl << "s (" << expiry << ")";
225 Ping->ping( ip.get_ip(),
227 boost::bind(&PingScheduler::ping_done_handler, this, _1) );
232 //------------------------------------------------------------------------------
233 // Post Processing of Ping result
234 //------------------------------------------------------------------------------
237 * @brief called when Ping::ping is done; calls functions to update
238 * statistics, ping interval and elapsed time;
239 * schedules a call to ping, thereby closing the loop
241 void PingScheduler::ping_done_handler( const bool ping_success )
243 GlobalLogger.info() << LogPrefix << "Ping done with success = "
247 // You must call these 3 methods exactly in this order
248 // TODO Fix this method, once it has a semantic dependency with the
249 // update_ping_statistics method, because it depends on the PingAnalyzer
250 // statistics to update the exceeded_ping_failed_limit
251 HostAnalyzer.update_ping_statistics( ping_success );
252 update_ping_interval();
253 update_ping_elapsed_time();
256 { // reset ContinueOnOutdatedIps
257 ContinueOnOutdatedIps = false;
261 // get next protocol, possibly start resolving IPs
262 update_ping_protocol();
264 // schedule next ping
265 (void) NextPingTimer.expires_from_now( seconds( PingIntervalInSec ) );
266 NextPingTimer.async_wait( bind( &PingScheduler::ping, this,
267 boost::asio::placeholders::error ) );
270 void PingScheduler::update_ping_interval()
272 // have to ping more often?
273 if ( HostAnalyzer.exceeded_ping_failed_limit() )
275 PingIntervalInSec.speed_up();
277 GlobalLogger.debug() << LogPrefix << "- Speeding up ping interval to: "
278 << PingIntervalInSec << "s";
282 PingIntervalInSec.back_to_original();
284 GlobalLogger.debug() << LogPrefix << "- Stick to the original ping "
285 << "interval: " << PingIntervalInSec << "s";
289 void PingScheduler::update_ping_elapsed_time()
291 ptime now = microsec_clock::universal_time();
292 time_resolution_traits_adapted64_impl::int_type elapsed_time_in_sec =
293 (now - TimeSentLastPing).total_seconds();
294 GlobalLogger.debug() << LogPrefix << "- Time elapsed since last ping: "
295 << elapsed_time_in_sec << "s";
297 TimeSentLastPing = microsec_clock::universal_time();
301 //------------------------------------------------------------------------------
302 // Ping Protocol Rotation
303 //------------------------------------------------------------------------------
305 void PingScheduler::init_ping_protocol()
307 ProtocolIter = Protocols.end();
308 get_next_ping_protocol();
311 void PingScheduler::update_ping_protocol()
313 if ( can_change_ping_protocol() )
315 get_next_ping_protocol();
319 void PingScheduler::get_next_ping_protocol()
323 Ping->stop_pinging();
327 GlobalLogger.debug() << LogPrefix
328 << "------------------------------------------------------------------";
330 if (ProtocolIter == Protocols.end())
331 ProtocolIter = Protocols.begin();
332 PingProtocol ping_protocol = *ProtocolIter;
333 // --> ProtocolIter still points to currently used protocol which is
334 // required in dns_resolve_callback
336 Ping = PingerFactory::createPinger(ping_protocol, IoService,
337 NetworkInterfaceName, PingReplyTimeout);
339 update_dns_resolver( ping_protocol );
343 bool PingScheduler::can_change_ping_protocol() const
345 // TODO can_change_ping_protocol() and get_next_ping_protocol() may be
346 // implemented in a Algorithm class that can be exchanged in this class to
347 // provide an algorithm neutral class
351 //------------------------------------------------------------------------------
352 // DNS host name resolution
353 //------------------------------------------------------------------------------
355 // show "!" after host name if running on outdated IPs
356 void PingScheduler::update_log_prefix()
358 std::stringstream temp;
359 temp << "Sched(" << DestinationAddress;
360 if (ContinueOnOutdatedIps)
363 LogPrefix = temp.str();
366 void PingScheduler::update_dns_resolver( PingProtocol current_protocol )
368 if (Resolver && Resolver->is_resolving())
369 cancel_resolve(false);
371 if (ContinueOnOutdatedIps)
373 ContinueOnOutdatedIps = false;
377 // DNS master caches created resolvers and resolved IPs, so this will
378 // probably just return an existing resolver with already resolved IPs for
379 // requested protocol ( ICMP/TCP is ignored, only IPv4/v6 is important)
380 Resolver = DnsMaster::get_instance()->get_resolver_for(DestinationAddress,
383 // get number of up-to-date IPs
384 // TODO should check here, if they will be up to date in PingIntervalInSec
385 bool check_up_to_date = true;
386 int ip_count = Resolver->get_resolved_ip_count(check_up_to_date);
389 GlobalLogger.info() << LogPrefix << "Set resolved_ip_count to "
390 << ip_count << " (IPs may be outdated=" << !check_up_to_date << ")";
391 HostAnalyzer.set_resolved_ip_count( ip_count );
393 if (Resolver->is_resolving())
394 GlobalLogger.warning() << LogPrefix << "have up to date IPs but "
395 << "resolver seems to be resolving all the same... "
396 << "Start pinging anyway!";
401 GlobalLogger.info() << LogPrefix
402 << "No up-to-date IPs --> start resolve";
403 start_resolving_ping_address();
404 // set resolved_ip_count will be called in resolve callback
408 void PingScheduler::start_resolving_ping_address()
410 Resolver->async_resolve( boost::bind(&PingScheduler::dns_resolve_callback,
414 void PingScheduler::dns_resolve_callback(const bool was_success,
415 const int recursion_count)
417 GlobalLogger.info() << LogPrefix << "dns resolution finished "
418 << "with success = " << was_success << " "
419 << "after " << recursion_count << " recursions";
423 // trust that a successfull DNS resolve means we have an IP with TTL>0
424 int ip_count = Resolver->get_resolved_ip_count(!ContinueOnOutdatedIps);
426 { // this will create trouble in HostAnalyzer
427 GlobalLogger.warning() << LogPrefix
428 << "Should not have reached this case: resolve was "
429 << "successfull but still have no IPs (up-to-date="
430 << !ContinueOnOutdatedIps << ")!";
431 if (DnsMaster::get_instance()->get_resolved_ip_ttl_threshold() > 0)
432 GlobalLogger.warning() << LogPrefix << "This probably happened "
433 << "because you specified a TTL threshold > 0 but resolving"
434 << " had no effect on TTLs since external cache is only "
435 << "updated when TTL=0 is reached.";
439 GlobalLogger.info() << LogPrefix << "Set resolved_ip_count to "
440 << ip_count << " (IPs may be outdated="
441 << ContinueOnOutdatedIps << ") --> could ping now";
442 HostAnalyzer.set_resolved_ip_count( ip_count );
447 { // host name resolution failed; try again bypassing first outdated CNAME
448 // or using cached IP
449 ContinueOnOutdatedIps = true;
452 std::string skip_host = Resolver->get_skip_cname();
454 if (skip_host.empty())
455 { // continue with IP
456 GlobalLogger.notice() << LogPrefix << "DNS failed, "
457 << "try anyway with cached data";
459 int ip_count = Resolver->get_resolved_ip_count(false);
460 GlobalLogger.info() << LogPrefix << "Set resolved_ip_count to "
461 << ip_count << " (IPs may be outdated=" << true << ")";
462 HostAnalyzer.set_resolved_ip_count( ip_count );
467 { // have CNAME to continue
468 GlobalLogger.notice() << LogPrefix << "DNS failed, "
469 << "try again skipping a CNAME and resolving "
470 << skip_host << " directly";
472 cancel_resolve(false);
474 // now create new resolver
475 Resolver = DnsMaster::get_instance()
476 ->get_resolver_for(skip_host, *ProtocolIter);
477 start_resolving_ping_address();
483 * cancel resolver if force_cancel or if it is not resolving DestinationAddress
485 * Resolvers have a life on their own: they are cached by DnsMaster so never go
486 * out of scope and even after calling callbacks, there might still be a
487 * longterm timer active to re-try resolving.
488 * We want to cancel that long-term timer only if the Resolver is not for our
489 * real, original DestinationAddress but a CNAME, which can happen when trying
490 * to skip cnames and working on out-dated IPs
492 void PingScheduler::cancel_resolve(const bool force_cancel)
496 GlobalLogger.info() << "Cancelling resolver (forced)";
497 Resolver->cancel_resolve();
499 else if ( Resolver->get_hostname() == DestinationAddress )
500 GlobalLogger.info() << LogPrefix
501 << "Leave original resolver active in background";
504 GlobalLogger.info() << LogPrefix << "Cancel resolver for "
505 << Resolver->get_hostname() << " since is not the original "
506 << DestinationAddress;
507 Resolver->cancel_resolve();