Fix and log kinesis duplicates (#50)

* update kinesis stream. to log duplicates into sentry

* add docs to aws for kinesis stream

* update sentry to capture message

* add spec to check if it at least dedupes

app/models/kinesis_stream.rb

@@ -9,14 +9,84 @@ def initialize
 
   def create_events(payload)
     receive(payload)
-    # TO DO (possibly?): We may want to consider doing these upserts/inserts in batches to improve performance.
-    CommentEvent.upsert_all(@comment_events, unique_by: %i[comment_id event_time]) unless @comment_events.empty?
-    unless @classification_events.empty?
-      ClassificationEvent.upsert_all(@classification_events,
-                                     unique_by: %i[classification_id event_time])
+
+    # Because ERAS is one of the ONLY receiving apps that receives from kinesis and BULK UPSERTS (feature of Rails 6+), it has caught duplicates on payload from kinesis stream
+    # See: https://zooniverse-27.sentry.io/issues/4717869260/?project=4506117954011141&query=is%3Aunresolved&referrer=issue-stream&statsPeriod=14d&stream_index=3
+    # EVEN THOUGH de-duping the payload by id before upserting should resolve issues with ERAS, (since ERAS only cares about counting the classification/comment once),
+    # UNFORTUNATELY, there are other apps (eg. Caesar, Tove) that rely on the kinesis stream and where duplicates in payload may affect results.
+    # Since ERAS is one of the only places we can catch this error (because of how it can bulk upsert), the team has decided to log the error to Sentry when the duplicates in payload occurs
+    # and also log the payload to Sentry.
+    ## Should note that this duplicate error has been seen before:
+    ## SEE: https://github.com/zooniverse/zoo-stats-api-graphql/pull/128
+    ## ALSO NOTING: THIS CATCH, LOG, DEDUPE AND TRY UPSERTING AGAIN TO DB SITUATION IS TEMPORARY AND ONLY USED
+    ## TO SEE WHAT THE DUPES IN THE KINESIS PAYLOAD ARE.
+    ## ONE MORE NOTE: per Kinesis docs, it is VERY possible for Kinesis stream to send duplicates and
+    ## the recommendation of AWS is to appropriately handle process records.
+    ## SEE: https://docs.aws.amazon.com/streams/latest/dev/kinesis-record-processor-duplicates.html
+
+    upsert_comments unless @comment_events.empty?
+    upsert_classifications unless @classification_events.empty?
+    upsert_classification_user_groups unless @classification_user_groups.empty?
+  end
+
+  def upsert_comments
+    CommentEvent.upsert_all(@comment_events, unique_by: %i[comment_id event_time])
+  rescue StandardError => e
+    crumb = Sentry::Breadcrumb.new(
+      category: 'upsert_error_in_comments',
+      message: 'Comment Events Upsert Error',
+      data: {
+        payload: @comment_events,
+        error_message: e.message
+      },
+      level: 'warning'
+    )
+    Sentry.add_breadcrumb(crumb)
+    Sentry.capture_exception(e)
+    if e.message.include?('ON CONFLICT DO UPDATE command cannot affect row a second time')
+      @comment_events = @comment_events.uniq { |comment| comment[:comment_id] }
+      retry
     end
+  end
 
-    ClassificationUserGroup.upsert_all(@classification_user_groups.flatten, unique_by: %i[classification_id event_time user_group_id user_id]) unless @classification_user_groups.empty?
+  def upsert_classifications
+    ClassificationEvent.upsert_all(@classification_events, unique_by: %i[classification_id event_time])
+  rescue StandardError => e
+    crumb = Sentry::Breadcrumb.new(
+      category: 'upsert_error_in_classifications',
+      message: 'Classification Events Upsert Error',
+      data: {
+        payload: @classification_events,
+        error_message: e.message
+      },
+      level: 'warning'
+    )
+    Sentry.add_breadcrumb(crumb)
+    Sentry.capture_exception(e)
+    if e.message.include?('ON CONFLICT DO UPDATE command cannot affect row a second time')
+      @classification_events = @classification_events.uniq { |classification| classification[:classification_id] }
+      retry
+    end
+  end
+
+  def upsert_classification_user_groups
+    ClassificationUserGroup.upsert_all(@classification_user_groups.flatten, unique_by: %i[classification_id event_time user_group_id user_id])
+  rescue StandardError => e
+    crumb = Sentry::Breadcrumb.new(
+      category: 'upsert_error_in_classifications_user_groups',
+      message: 'Classification User Groups Upsert Error',
+      data: {
+        payload: @classification_user_groups,
+        error_message: e.message
+      },
+      level: 'warning'
+    )
+    Sentry.add_breadcrumb(crumb)
+    Sentry.capture_exception(e)
+    if e.message.include?('ON CONFLICT DO UPDATE command cannot affect row a second time')
+      @classification_user_groups = @classification_user_groups.uniq { |cug| [cug[:classification_id], cug[:user_group_id]] }
+      retry
+    end
   end
 
   def receive(payload)

spec/models/kinesis_stream_spec.rb

@@ -71,5 +71,17 @@
         end.to change(ClassificationUserGroup, :count).from(0).to(1)
       end
     end
+
+    context 'payload has duplicates' do
+      it 'dedupes classification_events' do
+        kinesis_stream.create_events([classification_payload, classification_payload])
+        expect(kinesis_stream.instance_variable_get(:@classification_events).length).to eq(1)
+      end
+
+      it 'dedupes comment_events' do
+        kinesis_stream.create_events([comment_payload, comment_payload, comment_payload])
+        expect(kinesis_stream.instance_variable_get(:@comment_events).length).to eq(1)
+      end
+    end
   end
 end