/fa20-lab3-code

Primary LanguageHTML

Lab 3 and Lab 4: Raft

To get started go to https://classroom.github.com/a/t5hPLITs to create a repository and get the template code.

Introduction

In this lab you will implement a replicated state machine using Raft as the underlying consensus protocol. The RSM you are implementing models a queue supporting enqueue and dequeue operations. The RSM also provides a no-op operation that is used for testing and forcing commits.

NOTE: This project is significantly more complex than Lab 1 and 2. Despite being split across 2 labs, we expect that you will need to spend significant time on completing this. You should start early, and really spend some time thinking about how to best implement this code. The stencil code itself is very long, and reading through it will probably take a while.

NOTE 2 Figure 2 in the Raft paper is exactly what your implementation should do. If in doubt, you should figure out how and why you implementation differs from what is described in Figure 2.

You do not need to implement:

  • Snapshotting or anything else to reduce log sizes.
  • Reconfiguration or mechanisms to change membership.

For this project you can safely assume that the network does not lose packets and that messages between pairs of processes are delivered in order.

Lab 3: Get the RSM Working without Leader Election (AppendEntries)

For Lab 3 you must write all of the code necessary to launch a cluster with a pre-appointed leader and allow clients to commit changes. In particular this requires

  • Implementing AppendEntries so the leader can replicate log entries across the cluster.
  • Making sure followers respond to AppendEntries appropriately.
  • Making sure the leader can decide when an entry has been committed. The leader should only respond to the client once the appropriate operation has been applied to the queue, which of course requires that the corresponding log entry be committed. To help with this LogEntry's store information about the client that invoked a request.
  • Making sure that followers apply committed log entries eventually.

While you do not need to implement Heartbeats as a part of this portion, we strongly recommend doing so since the logic closely mirrors bits you will be implementing.

Also, you can start with the assumption that all processes are either leaders or followers, and hence not worry about candidates in this part of the lab.

Testing Lab 3

We have test cases for Lab 3 in test/lab3_test.exs which can be run using:

> mix test test/lab3_test.exs

You are definitely encouraged to add additional tests.

Handing in Lab 3

WARN WARN WARN PLEASE READ THESE INSTRUCTIONS CAREFULLY. YOU MAY RECEIVE A 0 (ZERO) IF YOU DO NOT, EVEN IF YOU COMPLETE EVERYTHING THUS FAR.

To handin this assignment:

  • First make sure mix test shows that you pass all tests. If not be aware that you will loose points.
  • Second, make sure you have updated this README.md file. This requires providing line numbers for the test you added in Part 2, potentially adding implementation notes to Part 3, and filling out the information below.
  • Commit and push all your changes.
  • Use git rev-parse --short HEAD to get a commit hash for your changes.
  • Fill out the submission form with all of the information requested.

We will be using information in the submission form to grade your lab, determine late days, etc. It is therefore crucial that you fill this out correctly.

Github username: (e.g., apanda)

NYU NetID: (e.g., ap191)

NYU N#:

Name:

Citations

Lab 3 Grading

We will be grading Lab 3 and Lab 4 at the same time. Half (50%) of your Lab 3 grade will be determined by how the final Lab 4 handing works with the lab 3 tests, while the rest will be determined the handin for Lab 3. Lab 3 largely exists to act as a forcing function fo your to being working on this lab.

Lab 4: Get Leader Election Working

In this second part you will complete your implementation so that leader election works correctly. This requires implementing everything required to handle RequestVote calls correctly, and for getting logs to match.

Testing Lab 4

We have test cases for Lab 4 in test/lab4test.exs which can be run using:

> mix test test/lab4_test.exs

You are definitely encouraged to add additional tests.

Handing in Lab 4

WARN WARN WARN PLEASE READ THESE INSTRUCTIONS CAREFULLY. YOU MAY RECEIVE A 0 (ZERO) IF YOU DO NOT, EVEN IF YOU COMPLETE EVERYTHING THUS FAR.

To handin this assignment:

  • First make sure mix test shows that you pass all tests. If not be aware that you will loose points.
  • Second, make sure you have updated this README.md file. This requires providing line numbers for the test you added in Part 2, potentially adding implementation notes to Part 3, and filling out the information below.
  • Commit and push all your changes.
  • Use git rev-parse --short HEAD to get a commit hash for your changes.
  • Fill out the submission form with all of the information requested.

We will be using information in the submission form to grade your lab, determine late days, etc. It is therefore crucial that you fill this out correctly.

Github username: (e.g., apanda)

NYU NetID: (e.g., ap191)

NYU N#:

Name:

Citations

Hints and Code Structure

The stencil code for this project is quite extensive. We recommend reading and understanding it, particularly the helper functions before beginning work. Below we provide a few notes that might help you as you look through the code:

  • We represent RPC calls and their returns using Elixir structs. These structs are defined in lib/messages.ex and you can see examples of their use in raft.ex.

    Utility functions in raft.ex including get_last_log_index, get_last_log_term, save_election_timer, etc. demonstrate how you can manipulate these structs.

  • Per-process state is defined in raft.ex which also provides a description of how the log is stored. Please make sure you read and understand that comment since it is important that you either (a) follow that log structure or (b) appropriately change the utility functions.

  • Log entries can be committed by calling the commit_log_entry or commit_log_index function.

  • You should use the save_heartbeat_timer and save_election_timer functions to store handles to the appropriate timers in process state. These handles are of course useful when cancelling timer.

  • You are required to implement the reset_election_timer and reset_heartbeat_time. Each should cancel any existing timer, and set a new one. You must use state.heartbeat_timeout as the heartbeat timeout, and call get_election_time to get the election timeout.

  • The code is structured so that each processes code appears as a state machine:

    • A process can be in one of three states: follower, leader and candidate. Each has a function that you should fill out with appropriate information.

      Of course some of the logic is shared across the three, and so you should make use of additional private functions to avoid code duplication.

      Each of these functions take both process state (state) and an additional extra_state parameter that you can use to track anything specific to a particular type of process. For example, a candidate can use the extra_state parameter to track the number of votes it has received.

    • We provide three functions become_leader, become_follower, and become_candidate that should be called when transitioning from one process state to another. You should used these functions to implement any one-time processing required, for instance you might want to add logic in become_leader to (a) start a heartbeat timer, and (b) assert leadership using an empty AppendEntry message.

      You should make sure your transitions call these become_* functions. For example when implementing the election timeout on a follower your code should roughly do the following:

      follower -(timer)-> become_candidate -----> candidate

      You might want to send out RequestVote requests in become_candidate.

  • Finally, start early, read through the code you are given, and write plan things out on paper before you start writing code.