Fix Leak of Instance Handles in DataReader

Also document the `bit_autopurge_nowriter_samples_delay` and
`bit_autopurge_disposed_samples_delay` config props.

dds/DCPS/DataReaderImpl.cpp

@@ -585,19 +585,17 @@ DataReaderImpl::remove_associations_i(const WriterIdSeq& writers,
   }
 
   for (CORBA::ULong i = 0; i < updated_writers.length(); ++i) {
-    {
-      this->disassociate(updated_writers[i]);
-    }
+    disassociate(updated_writers[i]);
   }
 
   // Mirror the add_associations SUBSCRIPTION_MATCHED_STATUS processing.
   if (!this->is_bit_) {
     ACE_Guard<ACE_Recursive_Thread_Mutex> justMe(publication_handle_lock_);
 
     // Derive the change in the number of publications writing to this reader.
-    int matchedPublications = static_cast<int>(this->publication_id_to_handle_map_.size());
-    this->subscription_match_status_.current_count_change
-    = matchedPublications - this->subscription_match_status_.current_count;
+    const int matchedPublications = static_cast<int>(publication_id_to_handle_map_.size());
+    subscription_match_status_.current_count_change =
+      matchedPublications - subscription_match_status_.current_count;
 
     // Only process status if the number of publications has changed.
     if (this->subscription_match_status_.current_count_change != 0) {
@@ -606,15 +604,12 @@ DataReaderImpl::remove_associations_i(const WriterIdSeq& writers,
       /// Section 7.1.4.1: total_count will not decrement.
 
       /// @TODO: Reconcile this with the verbiage in section 7.1.4.1
-      this->subscription_match_status_.last_publication_handle
-      = handles[ wr_len - 1];
+      subscription_match_status_.last_publication_handle = handles[wr_len - 1];
 
       set_status_changed_flag(DDS::SUBSCRIPTION_MATCHED_STATUS, true);
 
-      DDS::DataReaderListener_var listener
-      = listener_for(DDS::SUBSCRIPTION_MATCHED_STATUS);
-
-      if (!CORBA::is_nil(listener.in())) {
+      DDS::DataReaderListener_var listener = listener_for(DDS::SUBSCRIPTION_MATCHED_STATUS);
+      if (listener) {
         listener->on_subscription_matched(this, this->subscription_match_status_);
 
         // Client will look at it so next time it looks the change should be 0
@@ -635,6 +630,13 @@ DataReaderImpl::remove_associations_i(const WriterIdSeq& writers,
   if (this->monitor_) {
     this->monitor_->report();
   }
+
+  {
+    const RcHandle<DomainParticipantImpl> participant = participant_servant_.lock();
+    for (unsigned int i = 0; i < handles.length(); ++i) {
+      participant->return_handle(handles[i]);
+    }
+  }
 }
 
 void

docs/devguide/quality_of_service.rst

@@ -544,6 +544,10 @@ The ``max_samples`` member specifies the maximum number of samples a single data
 The ``max_instances`` member specifies the maximum number of instances that a data writer or data reader can manage.
 The ``max_samples_per_instance`` member specifies the maximum number of samples that can be managed for an individual instance in a single data writer or data reader.
 
+.. warning::
+
+  Setting these to very high values may exhaust available memory.
+
 Resources are used by the data writer to queue samples written to the data writer but not yet sent to all data readers because of backpressure from the transport.
 Resources are used by the data reader to queue samples that have been received, but not yet read/taken from the data reader.
 
@@ -1226,6 +1230,10 @@ Reader Data Lifecycle QoS
 
 The reader data lifecycle QoS policy controls the lifecycle of :term:`instance`\s managed by a :term:`DataReader`.
 
+.. warning::
+
+  The default values (never purging) will result in increasing memory usage over time for each new instance received.
+
 .. important::
 
   This policy is :ref:`mutable <qos-changing>` and does not affect association.
@@ -1275,9 +1283,11 @@ This policy could, for example, permit a late-joining data writer to prolong the
 
 The ``autopurge_nowriter_samples_delay`` controls how long the data reader waits before reclaiming resources once an instance transitions to the ``NOT_ALIVE_NO_WRITERS`` state.
 By default, ``autopurge_nowriter_samples_delay`` is infinite.
+Use :cfg:prop:`bit_autopurge_nowriter_samples_delay` to set this for :ref:`built-in data readers <bit>`.
 
 The ``autopurge_disposed_samples_delay`` controls how long the data reader waits before reclaiming resources once an instance transitions to the ``NOT_ALIVE_DISPOSED`` state.
 By default, ``autopurge_disposed_samples_delay`` is infinite.
+Use :cfg:prop:`bit_autopurge_disposed_samples_delay` to set this for :ref:`built-in data readers <bit>`.
 
 .. _qos-time-based-filter:
 .. _quality_of_service--time-based-filter:

docs/devguide/run_time_configuration.rst

@@ -369,6 +369,16 @@ For example:
 
 .. sec:: common
 
+  .. prop:: bit_autopurge_nowriter_samples_delay
+    :default: ``DURATION_INFINITE_SEC`` (disabled)
+
+    Sets the ``autopurge_nowriter_samples_delay`` value of the :ref:`quality_of_service--reader-data-lifecycle` for :ref:`built-in data readers <bit>`.
+
+  .. prop:: bit_autopurge_disposed_samples_delay
+    :default: ``DURATION_INFINITE_SEC`` (disabled)
+
+    Sets the ``autopurge_disposed_samples_delay`` value of the :ref:`quality_of_service--reader-data-lifecycle` for :ref:`built-in data readers <bit>`.
+
   .. prop:: DCPSBidirGIOP=<boolean>
     :default: ``1`` (enabled)
 
@@ -3331,7 +3341,7 @@ See :ref:`config-common` for options controlling the destination and formatting
 The highest level logging is controlled by the general log levels listed in the following table.
 
 .. list-table::
-   :header-rows: 1
+  :header-rows: 1
 
   * - Level
 

docs/news.d/dr-instance-handles.rst

@@ -0,0 +1,9 @@
+.. news-prs: 4784
+
+.. news-start-section: Fixes
+- Fixed a leak of instance handles in data readers that prevented :ref:`quality_of_service--reader-data-lifecycle` from properly purging data it was supposed to.
+.. news-end-section
+
+.. news-start-section: Documentation
+- Documented the :cfg:prop:`bit_autopurge_nowriter_samples_delay` and :cfg:prop:`bit_autopurge_disposed_samples_delay` configuration properties.
+.. news-end-section

docs/news.d/relay-max-participants.rst

@@ -1,7 +1,7 @@
 .. news-prs: 4774
 
 .. news-start-section: Additions
-- Added -AdmissionMaxParticipantsRange to RtpsRelay options
+- Added :option:`RtpsRelay -AdmissionMaxParticipantsRange`
 
   - This option provides another mechanism for detecting load on each RtpsRelay instance