Improve resilience against clock adjustments (#5018) (#5189)

* Improve resilience against clock adjustments (#5018)

* Use steady_clock instead of high_resolution_clock for status checks (high_resolution_clock might not be steady depending on STL impl)

Signed-off-by: Matthias Schneider <[email protected]>

* Use steady_clock instead for system_clock for calculating timeouts

Signed-off-by: Matthias Schneider <[email protected]>

* Use correct clock's duration for duration_cast

Signed-off-by: Matthias Schneider <[email protected]>

* Use Time_t::now()

Signed-off-by: Matthias Schneider <[email protected]>

* Fix build.

Signed-off-by: Miguel Company <[email protected]>

* Refs #21314. Refactor on DataWriterImpl.

Signed-off-by: Miguel Company <[email protected]>

* Refs #21314. Refactor on DataReaderImpl.

Signed-off-by: Miguel Company <[email protected]>

* Refs #21314. Refactor on StatefulWriter.

Signed-off-by: Miguel Company <[email protected]>

* Refs #21314. Protect current_time_since_unix_epoch against clock adjustments.

Signed-off-by: Miguel Company <[email protected]>

* Revert "Use steady_clock instead of high_resolution_clock for status checks (high_resolution_clock might not be steady depending on STL impl)"

This reverts commit d69eb91.

---------

Signed-off-by: Matthias Schneider <[email protected]>
Signed-off-by: Miguel Company <[email protected]>
Co-authored-by: Miguel Company <[email protected]>
(cherry picked from commit ccc690c)

* Refs #21314: Fix conflicts and build issues

Signed-off-by: JesusPoderoso <[email protected]>

* Refs #21314: Apply rev suggestions

Signed-off-by: JesusPoderoso <[email protected]>

* Refs #21314: Please uncrustify

Signed-off-by: JesusPoderoso <[email protected]>

* Keep comparison

Signed-off-by: Miguel Company <[email protected]>

* Use rtps data types

Signed-off-by: Miguel Company <[email protected]>

---------

Signed-off-by: Matthias Schneider <[email protected]>
Signed-off-by: Miguel Company <[email protected]>
Signed-off-by: JesusPoderoso <[email protected]>
Co-authored-by: Matthias Schneider <[email protected]>
Co-authored-by: Miguel Company <[email protected]>
Co-authored-by: JesusPoderoso <[email protected]>

Author: 4 people

include/fastdds/rtps/writer/StatefulWriter.h

@@ -515,8 +515,8 @@ class StatefulWriter : public RTPSWriter
     bool disable_heartbeat_piggyback_;
     //! True to disable positive ACKs
     bool disable_positive_acks_;
-    //! Keep duration for disable positive ACKs QoS, in microseconds
-    std::chrono::duration<double, std::ratio<1, 1000000>> keep_duration_us_;
+    //! Keep duration for disable positive ACKs QoS
+    Duration_t keep_duration_;
     //! Last acknowledged cache change (only used if using disable positive ACKs QoS)
     SequenceNumber_t last_sequence_number_;
     //! Biggest sequence number removed from history

src/cpp/fastdds/domain/DomainParticipantImpl.cpp

@@ -1221,14 +1221,7 @@ bool DomainParticipantImpl::contains_entity(
 ReturnCode_t DomainParticipantImpl::get_current_time(
         fastrtps::Time_t& current_time) const
 {
-    auto now = std::chrono::system_clock::now();
-    auto duration = now.time_since_epoch();
-    auto seconds = std::chrono::duration_cast<std::chrono::seconds>(duration);
-    duration -= seconds;
-    auto nanos = std::chrono::duration_cast<std::chrono::nanoseconds>(duration);
-
-    current_time.seconds = static_cast<int32_t>(seconds.count());
-    current_time.nanosec = static_cast<uint32_t>(nanos.count());
+    fastrtps::Time_t::now(current_time);
 
     return ReturnCode_t::RETCODE_OK;
 }

src/cpp/fastdds/domain/DomainParticipantImpl.hpp

@@ -111,10 +111,10 @@ class DomainParticipantImpl
             DomainParticipantListener* listener,
             const std::chrono::seconds timeout = std::chrono::seconds::max())
     {
-        auto time_out = std::chrono::time_point<std::chrono::system_clock>::max();
+        auto time_out = std::chrono::time_point<std::chrono::steady_clock>::max();
         if (timeout < std::chrono::seconds::max())
         {
-            auto now = std::chrono::system_clock::now();
+            auto now = std::chrono::steady_clock::now();
             time_out = now + timeout;
         }
 

src/cpp/fastdds/publisher/DataWriterImpl.cpp

@@ -34,6 +34,7 @@
 #include <fastdds/publisher/filtering/DataWriterFilteredChangePool.hpp>
 #include <fastdds/publisher/PublisherImpl.hpp>
 #include <fastdds/rtps/builtin/liveliness/WLP.h>
+#include <fastdds/rtps/common/Time_t.h>
 #include <fastdds/rtps/participant/RTPSParticipant.h>
 #include <fastdds/rtps/resources/ResourceEvent.h>
 #include <fastdds/rtps/resources/TimedEvent.h>
@@ -1027,7 +1028,7 @@ ReturnCode_t DataWriterImpl::perform_create_new_change(
         {
             if (!history_.set_next_deadline(
                         handle,
-                        steady_clock::now() + duration_cast<system_clock::duration>(deadline_duration_us_)))
+                        steady_clock::now() + duration_cast<steady_clock::duration>(deadline_duration_us_)))
             {
                 EPROSIMA_LOG_ERROR(DATA_WRITER, "Could not set the next deadline in the history");
             }
@@ -1499,7 +1500,7 @@ bool DataWriterImpl::deadline_missed()
 
     if (!history_.set_next_deadline(
                 timer_owner_,
-                steady_clock::now() + duration_cast<system_clock::duration>(deadline_duration_us_)))
+                steady_clock::now() + duration_cast<steady_clock::duration>(deadline_duration_us_)))
     {
         EPROSIMA_LOG_ERROR(DATA_WRITER, "Could not set the next deadline in the history");
         return false;
@@ -1549,39 +1550,24 @@ bool DataWriterImpl::lifespan_expired()
 {
     std::unique_lock<RecursiveTimedMutex> lock(writer_->getMutex());
 
+    fastrtps::rtps::Time_t current_ts;
+    fastrtps::rtps::Time_t::now(current_ts);
+
     CacheChange_t* earliest_change;
     while (history_.get_earliest_change(&earliest_change))
     {
-        auto source_timestamp = system_clock::time_point() + nanoseconds(earliest_change->sourceTimestamp.to_ns());
-        auto now = system_clock::now();
+        fastrtps::rtps::Time_t expiration_ts = earliest_change->sourceTimestamp + qos_.lifespan().duration;
 
         // Check that the earliest change has expired (the change which started the timer could have been removed from the history)
-        if (now - source_timestamp < lifespan_duration_us_)
+        if (current_ts < expiration_ts)
         {
-            auto interval = source_timestamp - now + lifespan_duration_us_;
-            lifespan_timer_->update_interval_millisec(static_cast<double>(duration_cast<milliseconds>(interval).count()));
+            fastrtps::rtps::Time_t interval = expiration_ts - current_ts;
+            lifespan_timer_->update_interval_millisec(interval.to_ns() * 1e-6);
             return true;
         }
 
         // The earliest change has expired
         history_.remove_change_pub(earliest_change);
-
-        // Set the timer for the next change if there is one
-        if (!history_.get_earliest_change(&earliest_change))
-        {
-            return false;
-        }
-
-        // Calculate when the next change is due to expire and restart
-        source_timestamp = system_clock::time_point() + nanoseconds(earliest_change->sourceTimestamp.to_ns());
-        now = system_clock::now();
-        auto interval = source_timestamp - now + lifespan_duration_us_;
-
-        if (interval.count() > 0)
-        {
-            lifespan_timer_->update_interval_millisec(static_cast<double>(duration_cast<milliseconds>(interval).count()));
-            return true;
-        }
     }
 
     return false;

src/cpp/fastdds/subscriber/DataReaderImpl.cpp

@@ -35,6 +35,7 @@
 #include <fastdds/dds/topic/Topic.hpp>
 #include <fastdds/dds/topic/TypeSupport.hpp>
 #include <fastdds/domain/DomainParticipantImpl.hpp>
+#include <fastdds/rtps/common/Time_t.h>
 #include <fastdds/rtps/participant/RTPSParticipant.h>
 #include <fastdds/rtps/reader/RTPSReader.h>
 #include <fastdds/rtps/resources/ResourceEvent.h>
@@ -1103,7 +1104,7 @@ bool DataReaderImpl::on_new_cache_change_added(
     {
         if (!history_.set_next_deadline(
                     change->instanceHandle,
-                    steady_clock::now() + duration_cast<system_clock::duration>(deadline_duration_us_)))
+                    steady_clock::now() + duration_cast<steady_clock::duration>(deadline_duration_us_)))
         {
             EPROSIMA_LOG_ERROR(SUBSCRIBER, "Could not set next deadline in the history");
         }
@@ -1122,12 +1123,13 @@ bool DataReaderImpl::on_new_cache_change_added(
         return true;
     }
 
-    auto source_timestamp = system_clock::time_point() + nanoseconds(change->sourceTimestamp.to_ns());
-    auto now = system_clock::now();
+    fastrtps::rtps::Time_t expiration_ts = change->sourceTimestamp + qos_.lifespan().duration;
+    fastrtps::rtps::Time_t current_ts;
+    fastrtps::rtps::Time_t::now(current_ts);
 
     // The new change could have expired if it arrived too late
     // If so, remove it from the history and return false to avoid notifying the listener
-    if (now - source_timestamp >= lifespan_duration_us_)
+    if (expiration_ts <= current_ts)
     {
         history_.remove_change_sub(new_change);
         return false;
@@ -1149,11 +1151,10 @@ bool DataReaderImpl::on_new_cache_change_added(
         EPROSIMA_LOG_ERROR(SUBSCRIBER, "A change was added to history that could not be retrieved");
     }
 
-    auto interval = source_timestamp - now + duration_cast<nanoseconds>(lifespan_duration_us_);
-
     // Update and restart the timer
     // If the timer is already running this will not have any effect
-    lifespan_timer_->update_interval_millisec(interval.count() * 1e-6);
+    fastrtps::rtps::Time_t interval = expiration_ts - current_ts;
+    lifespan_timer_->update_interval_millisec(interval.to_ns() * 1e-6);
     lifespan_timer_->restart_timer();
     return true;
 }
@@ -1244,7 +1245,7 @@ bool DataReaderImpl::deadline_missed()
 
     if (!history_.set_next_deadline(
                 timer_owner_,
-                steady_clock::now() + duration_cast<system_clock::duration>(deadline_duration_us_), true))
+                steady_clock::now() + duration_cast<steady_clock::duration>(deadline_duration_us_), true))
     {
         EPROSIMA_LOG_ERROR(SUBSCRIBER, "Could not set next deadline in the history");
         return false;
@@ -1275,41 +1276,26 @@ bool DataReaderImpl::lifespan_expired()
 {
     std::unique_lock<RecursiveTimedMutex> lock(reader_->getMutex());
 
+    fastrtps::rtps::Time_t current_ts;
+    fastrtps::rtps::Time_t::now(current_ts);
+
     CacheChange_t* earliest_change;
     while (history_.get_earliest_change(&earliest_change))
     {
-        auto source_timestamp = system_clock::time_point() + nanoseconds(earliest_change->sourceTimestamp.to_ns());
-        auto now = system_clock::now();
+        fastrtps::rtps::Time_t expiration_ts = earliest_change->sourceTimestamp + qos_.lifespan().duration;
 
         // Check that the earliest change has expired (the change which started the timer could have been removed from the history)
-        if (now - source_timestamp < lifespan_duration_us_)
+        if (current_ts < expiration_ts)
         {
-            auto interval = source_timestamp - now + lifespan_duration_us_;
-            lifespan_timer_->update_interval_millisec(static_cast<double>(duration_cast<milliseconds>(interval).count()));
+            fastrtps::rtps::Time_t interval = expiration_ts - current_ts;
+            lifespan_timer_->update_interval_millisec(interval.to_ns() * 1e-6);
             return true;
         }
 
         // The earliest change has expired
         history_.remove_change_sub(earliest_change);
 
         try_notify_read_conditions();
-
-        // Set the timer for the next change if there is one
-        if (!history_.get_earliest_change(&earliest_change))
-        {
-            return false;
-        }
-
-        // Calculate when the next change is due to expire and restart
-        source_timestamp = system_clock::time_point() + nanoseconds(earliest_change->sourceTimestamp.to_ns());
-        now = system_clock::now();
-        auto interval = source_timestamp - now + lifespan_duration_us_;
-
-        if (interval.count() > 0)
-        {
-            lifespan_timer_->update_interval_millisec(static_cast<double>(duration_cast<milliseconds>(interval).count()));
-            return true;
-        }
     }
 
     return false;

src/cpp/rtps/writer/StatefulWriter.cpp

@@ -183,7 +183,7 @@ StatefulWriter::StatefulWriter(
     , may_remove_change_(0)
     , disable_heartbeat_piggyback_(att.disable_heartbeat_piggyback)
     , disable_positive_acks_(att.disable_positive_acks)
-    , keep_duration_us_(att.keep_duration.to_ns() * 1e-3)
+    , keep_duration_(att.keep_duration)
     , last_sequence_number_()
     , biggest_removed_sequence_number_()
     , sendBufferSize_(pimpl->get_min_network_send_buffer_size())
@@ -218,7 +218,7 @@ StatefulWriter::StatefulWriter(
     , may_remove_change_(0)
     , disable_heartbeat_piggyback_(att.disable_heartbeat_piggyback)
     , disable_positive_acks_(att.disable_positive_acks)
-    , keep_duration_us_(att.keep_duration.to_ns() * 1e-3)
+    , keep_duration_(att.keep_duration)
     , last_sequence_number_()
     , biggest_removed_sequence_number_()
     , sendBufferSize_(pimpl->get_min_network_send_buffer_size())
@@ -254,7 +254,7 @@ StatefulWriter::StatefulWriter(
     , may_remove_change_(0)
     , disable_heartbeat_piggyback_(att.disable_heartbeat_piggyback)
     , disable_positive_acks_(att.disable_positive_acks)
-    , keep_duration_us_(att.keep_duration.to_ns() * 1e-3)
+    , keep_duration_(att.keep_duration)
     , last_sequence_number_()
     , biggest_removed_sequence_number_()
     , sendBufferSize_(pimpl->get_min_network_send_buffer_size())
@@ -432,12 +432,13 @@ void StatefulWriter::unsent_change_added_to_history(
 
         if (disable_positive_acks_)
         {
-            auto source_timestamp = system_clock::time_point() + nanoseconds(change->sourceTimestamp.to_ns());
-            auto now = system_clock::now();
-            auto interval = source_timestamp - now + keep_duration_us_;
-            assert(interval.count() >= 0);
+            Time_t expiration_ts = change->sourceTimestamp + keep_duration_;
+            Time_t current_ts;
+            Time_t::now(current_ts);
+            assert(expiration_ts >= current_ts);
+            auto interval = (expiration_ts - current_ts).to_duration_t();
 
-            ack_event_->update_interval_millisec((double)duration_cast<milliseconds>(interval).count());
+            ack_event_->update_interval(interval);
             ack_event_->restart_timer(max_blocking_time);
         }
 
@@ -953,12 +954,13 @@ DeliveryRetCode StatefulWriter::deliver_sample_to_network(
             if ( !(ack_event_->getRemainingTimeMilliSec() > 0))
             {
                 // Restart ack_timer
-                auto source_timestamp = system_clock::time_point() + nanoseconds(change->sourceTimestamp.to_ns());
-                auto now = system_clock::now();
-                auto interval = source_timestamp - now + keep_duration_us_;
-                assert(interval.count() >= 0);
+                Time_t expiration_ts = change->sourceTimestamp + keep_duration_;
+                Time_t current_ts;
+                Time_t::now(current_ts);
+                assert(expiration_ts >= current_ts);
+                auto interval = (expiration_ts - current_ts).to_duration_t();
 
-                ack_event_->update_interval_millisec((double)duration_cast<milliseconds>(interval).count());
+                ack_event_->update_interval(interval);
                 ack_event_->restart_timer(max_blocking_time);
             }
         }
@@ -1675,13 +1677,9 @@ void StatefulWriter::updatePositiveAcks(
         const WriterAttributes& att)
 {
     std::lock_guard<RecursiveTimedMutex> guard(mp_mutex);
-    if (keep_duration_us_.count() != (att.keep_duration.to_ns() * 1e-3))
-    {
-        // Implicit conversion to microseconds
-        keep_duration_us_ = std::chrono::nanoseconds {att.keep_duration.to_ns()};
-    }
+    keep_duration_ = att.keep_duration;
     // Restart ack timer with new duration
-    ack_event_->update_interval_millisec(keep_duration_us_.count() * 1e-3);
+    ack_event_->update_interval(keep_duration_);
     ack_event_->restart_timer();
 }
 
@@ -2094,14 +2092,16 @@ bool StatefulWriter::ack_timer_expired()
     // The timer has expired so the earliest non-acked change must be marked as acknowledged
     // This will be done in the first while iteration, as we start with a negative interval
 
-    auto interval = -keep_duration_us_;
+    Time_t expiration_ts;
+    Time_t current_ts;
+    Time_t::now(current_ts);
 
     // On the other hand, we've seen in the tests that if samples are sent very quickly with little
     // time between consecutive samples, the timer interval could end up being negative
     // In this case, we keep marking changes as acknowledged until the timer is able to keep up, hence the while
     // loop
 
-    while (interval.count() < 0)
+    do
     {
         bool acks_flag = false;
         for_matched_readers(matched_local_readers_, matched_datasharing_readers_, matched_remote_readers_,
@@ -2141,13 +2141,13 @@ bool StatefulWriter::ack_timer_expired()
             return false;
         }
 
-        auto source_timestamp = system_clock::time_point() + nanoseconds(change->sourceTimestamp.to_ns());
-        auto now = system_clock::now();
-        interval = source_timestamp - now + keep_duration_us_;
+        Time_t::now(current_ts);
+        expiration_ts = change->sourceTimestamp + keep_duration_;
     }
-    assert(interval.count() >= 0);
+    while (expiration_ts < current_ts);
 
-    ack_event_->update_interval_millisec((double)duration_cast<milliseconds>(interval).count());
+    auto interval = (expiration_ts - current_ts).to_duration_t();
+    ack_event_->update_interval(interval);
     return true;
 }
 

src/cpp/utils/SystemInfo.cpp

@@ -174,7 +174,7 @@ bool SystemInfo::wait_for_file_closure(
         const std::string& filename,
         const std::chrono::seconds timeout)
 {
-    auto start = std::chrono::system_clock::now();
+    auto start = std::chrono::steady_clock::now();
 
 #ifdef _MSC_VER
     std::ofstream os;
@@ -184,7 +184,7 @@ bool SystemInfo::wait_for_file_closure(
         os.open(filename, std::ios::out | std::ios::app, _SH_DENYWR);
         if (!os.is_open()
                 // If the file is lock-opened in an external editor do not hang
-                && (std::chrono::system_clock::now() - start) < timeout )
+                && (std::chrono::steady_clock::now() - start) < timeout )
         {
             std::this_thread::yield();
         }
@@ -199,7 +199,7 @@ bool SystemInfo::wait_for_file_closure(
 
     while (flock(fd, LOCK_EX | LOCK_NB)
             // If the file is lock-opened in an external editor do not hang
-            && (std::chrono::system_clock::now() - start) < timeout )
+            && (std::chrono::steady_clock::now() - start) < timeout )
     {
         std::this_thread::yield();
     }
@@ -214,7 +214,7 @@ bool SystemInfo::wait_for_file_closure(
     (void)filename;
 #endif // ifdef _MSC_VER
 
-    return std::chrono::system_clock::now() - start < timeout;
+    return std::chrono::steady_clock::now() - start < timeout;
 }
 
 ReturnCode_t SystemInfo::set_environment_file()