User:Akumarm

Fallback Algorithm

Introduction

The Fallback Algorithm in Expertiza is designed to ensure automatic topic assignment when the primary Bidding Algorithm fails due to Web Service unavailability.

Normally, Expertiza allows reviewers to bid on topics, and the ReviewBiddingAlgorithmService processes these bids to assign topics. However, when this web service fails (due to API downtime, server issues, or unexpected errors), the system must switch to fallback and automatically assign topics to reviewers.

Solution

It is triggered when the web service is unavailable. It ensures fair topic assignment by:

• Prioritizing topics with the largest teams.
• Using a round-robin approach to distribute topics among reviewers.
• Ensuring reviewers do not get their own team’s topic.

When is this Used?

• The ReviewBiddingAlgorithmService fails due to an API issue.
• The web service times out or returns an error.
• Unexpected exceptions occur in the bidding process.

Instead of stopping the review process, the system automatically falls-back to an internal algorithm to distribute topics.

How Does the Fallback Algorithm Work?

The algorithm follows these **four main steps:

• Fetch available topics – Retrieves all topics for the given assignment.
• Sort topics by team size – Topics with more members are given priority.
• Create a topic queue – Topics are arranged in descending order of team size.
• Assign topics in a round-robin manner – Reviewers are assigned topics while avoiding their own team's topic.

This ensures fair and balanced distribution of topics.

Implementation - Controller Code (Triggering the Fallback)

If the primary bidding service fails, the system catches the error and calls the Fallback Algorithm.

   def assign_bidding
     assignment_id = params[:assignment_id].to_i
     reviewer_ids = AssignmentParticipant.where(parent_id: assignment_id).ids
 
     begin
       # Attempt to use the web service
       bidding_data = ReviewBid.bidding_data(assignment_id, reviewer_ids)
       matched_topics = ReviewBiddingAlgorithmService.run_bidding_algorithm(bidding_data)
     rescue StandardError => e
       # If web service fails, trigger fallback
       Rails.logger.error "Web service unavailable: #{e.message}"
       matched_topics = ReviewBid.fallback_algorithm(assignment_id, reviewer_ids)
     end
 
     # Ensure valid topic assignments
     matched_topics ||= {}
     reviewer_ids.each { |reviewer_id| matched_topics[reviewer_id.to_s] ||= [] }
 
     Rails.logger.debug "Final matched topics after fallback: #{matched_topics.inspect}"
   
     # Assign topics to reviewers
     ReviewBid.assign_review_topics(assignment_id, reviewer_ids, matched_topics)
 
     # Disable topic selection after assignment
     Assignment.find(assignment_id).update(can_choose_topic_to_review: false)
 
     redirect_back fallback_location: root_path
 end

Key Takeaway: If ReviewBiddingAlgorithmService fails, the system automatically calls fallback_algorithm.

Implementation - Model Code (Fallback Algorithm)

This method ensures fair topic assignment when the bidding system fails.

 def fallback_algorithm(assignment_id, reviewer_ids)
   Rails.logger.debug "Fallback algorithm triggered for assignment_id: #{assignment_id}"
   matched_topics = {}
   # Step 1: Get available topics
   topics = SignUpTopic.where(assignment_id: assignment_id).pluck(:id)
   Rails.logger.debug "Available topics: #{topics}"
   # Step 2: Get team sizes and sort by largest teams first
   teams = SignedUpTeam.where(topic_id: topics)
                       .joins(:team)
                       .joins("LEFT JOIN teams_users ON teams.id = teams_users.team_id")
                       .group(:topic_id)
                       .count("teams_users.user_id")
   # Sort teams by size (Descending Order)
   sorted_teams = teams.sort_by { |_, count| -count }
   Rails.logger.debug "Teams sorted by size: #{sorted_teams}"
   # Step 3: Create topic queue (largest teams first)
   topic_queue = sorted_teams.map(&:first)
   # Step 4: Assign topics in a round-robin manner
   topic_index = 0
   reviewer_ids.each do |reviewer_id|
       assigned_topic = nil
       self_topic = fetch_self_topic(assignment_id, reviewer_id)
       attempts = 0
       while assigned_topic.nil? && attempts < topic_queue.size
           topic_id = topic_queue[topic_index % topic_queue.size]
           unless topic_id == self_topic
               assigned_topic = topic_id
               Rails.logger.debug "Assigned topic #{assigned_topic} to reviewer #{reviewer_id}"
               topic_index += 1
           end
           attempts += 1
       end
       matched_topics[reviewer_id.to_s] = assigned_topic ? [assigned_topic] : []
   end
   Rails.logger.debug "Final matched topics after fallback: #{matched_topics.inspect}"
   matched_topics
 end

Key Takeaway: This function ensures every reviewer receives a topic without requiring manual intervention.

RSpec Testing of Fallback Algorithm

To verify correctness, we use unit tests in review_bid_spec.rb:

describe '.fallback_algorithm' do
 let(:assignment_id) { 2085 }
 let(:reviewer_ids) { [1, 2, 3] }
 let(:topics) { [101, 102, 103] }
 let(:teams) { { 101 => 5, 102 => 3, 103 => 1 } }
 before do
   allow(SignUpTopic).to receive(:where).with(assignment_id: assignment_id).and_return(double(pluck: topics))
   allow(SignedUpTeam).to receive_message_chain(:where, :joins, :group, :count).and_return(teams)
   allow(SignedUpTeam).to receive(:topic_id).with(assignment_id, 1).and_return(101)
   allow(SignedUpTeam).to receive(:topic_id).with(assignment_id, 2).and_return(102)
   allow(SignedUpTeam).to receive(:topic_id).with(assignment_id, 3).and_return(nil)
 end
 it 'assigns topics in a round-robin manner while avoiding self-assignment' do
   result = ReviewBid.fallback_algorithm(assignment_id, reviewer_ids)
   expect(result['1']).not_to include(101)
   expect(result['2']).not_to include(102)
   expect([101, 102, 103]).to include(result['3'].first)
 end
end

The Fallback Algorithm ensures that topics are assigned fairly even when the primary bidding service fails. By prioritizing larger teams, avoiding self-review, and using a round-robin strategy, the fallback mechanism keeps the review process smooth and automatic.