Assignment 2 continues the work from the initial assignment — building a Map/Reduce library as a way to learn the Go programming language and as a way to learn about fault tolerance in distributed systems. In this assignment, you will tackle a distributed version of the Map/Reduce library, writing code for a master that hands out tasks to multiple workers and handles failures in workers. The interface to the library and the approach to fault tolerance is similar to the one described in the original MapReduce paper. As with the previous assignment, you will also complete a sample Map/Reduce application.
You will use the same mapreduce package as in Assignment 1 (assignment1.tar.gz), focusing this time on the distributed mode.
Over the course of this assignment, you will have to modify schedule from schedule.go, as well as mapF and reduceF in main/ii.go.
As with the previous assignment, you should not need to modify any other files, but reading them might be useful in order to understand how the other methods fit into the overall architecture of the system.
One of Map/Reduce's biggest selling points is that the developer should not need to be aware that their code is running in parallel on many machines. In theory, we should be able to take the word count code you wrote in Part II of Assignment 1, and automatically parallelize it!
Our current implementation runs all the map and reduce tasks one after another on the master. While this is conceptually simple, it is not great for performance. In this part of the assignment, you will complete a version of MapReduce that splits the work up over a set of worker threads, in order to exploit multiple cores. Computing the map tasks in parallel and then the reduce tasks can result in much faster completion, but is also harder to implement and debug. Note that for this assignment, the work is not distributed across multiple machines as in “real” Map/Reduce deployments, your implementation will be using RPC and channels to simulate a truly distributed computation.
To coordinate the parallel execution of tasks, we will use a special master thread, which hands out work to the workers and waits for them to finish. To make the assignment more realistic, the master should only communicate with the workers via RPC. We give you the worker code (mapreduce/worker.go), the code that starts the workers, and code to deal with RPC messages (mapreduce/common_rpc.go).
Your job is to complete schedule.go in the mapreduce package. In particular, you should modify schedule() in schedule.go to hand out the map and reduce tasks to workers, and return only when all the tasks have finished.
Look at run() in master.go. It calls your schedule() to run the map and reduce tasks, then calls merge() to assemble the per-reduce-task outputs into a single output file. schedule only needs to tell the workers the name of the original input file (mr.files[task]) and the task task; each worker knows from which files to read its input and to which files to write its output. The master tells the worker about a new task by sending it the RPC call Worker.DoTask, giving a DoTaskArgs object as the RPC argument.
When a worker starts, it sends a Register RPC to the master. master.go already implements the master's Master.Register RPC handler for you, and passes the new worker's information to mr.registerChannel. Your schedule should process new worker registrations by reading from this channel.
Information about the currently running job is in the Master struct, defined in master.go. Note that the master does not need to know which Map or Reduce functions are being used for the job; the workers will take care of executing the right code for Map or Reduce (the correct functions are given to them when they are started by main/wc.go).
To test your solution, you should use the same Go test suite as you did in Part I of Assignment 1, except swapping out -run Sequential with -run TestBasic. This will execute the distributed test case without worker failures instead of the sequential ones we were running before:
$ go test -run TestBasic mapreduce/... -vet=offAs before, you can get more verbose output for debugging if you set debugEnabled = true in mapreduce/common.go, and add -v to the test command above. You will get much more output along the lines of:
$ go test -v -run TestBasic mapreduce/... -vet=off === RUN TestBasic /var/tmp/824-32311/mr8665-master: Starting Map/Reduce task test Schedule: 100 Map tasks (50 I/Os) /var/tmp/824-32311/mr8665-worker0: given Map task #0 on file 824-mrinput-0.txt (nios: 50) /var/tmp/824-32311/mr8665-worker1: given Map task #11 on file 824-mrinput-11.txt (nios: 50) /var/tmp/824-32311/mr8665-worker0: Map task #0 done /var/tmp/824-32311/mr8665-worker0: given Map task #1 on file 824-mrinput-1.txt (nios: 50) /var/tmp/824-32311/mr8665-worker1: Map task #11 done /var/tmp/824-32311/mr8665-worker1: given Map task #2 on file 824-mrinput-2.txt (nios: 50) /var/tmp/824-32311/mr8665-worker0: Map task #1 done /var/tmp/824-32311/mr8665-worker0: given Map task #3 on file 824-mrinput-3.txt (nios: 50) /var/tmp/824-32311/mr8665-worker1: Map task #2 done ... Schedule: Map phase done Schedule: 50 Reduce tasks (100 I/Os) /var/tmp/824-32311/mr8665-worker1: given Reduce task #49 on file 824-mrinput-49.txt (nios: 100) /var/tmp/824-32311/mr8665-worker0: given Reduce task #4 on file 824-mrinput-4.txt (nios: 100) /var/tmp/824-32311/mr8665-worker1: Reduce task #49 done /var/tmp/824-32311/mr8665-worker1: given Reduce task #1 on file 824-mrinput-1.txt (nios: 100) /var/tmp/824-32311/mr8665-worker0: Reduce task #4 done /var/tmp/824-32311/mr8665-worker0: given Reduce task #0 on file 824-mrinput-0.txt (nios: 100) /var/tmp/824-32311/mr8665-worker1: Reduce task #1 done /var/tmp/824-32311/mr8665-worker1: given Reduce task #26 on file 824-mrinput-26.txt (nios: 100) /var/tmp/824-32311/mr8665-worker0: Reduce task #0 done ... Schedule: Reduce phase done Merge: read mrtmp.test-res-0 Merge: read mrtmp.test-res-1 ... Merge: read mrtmp.test-res-49 /var/tmp/824-32311/mr8665-master: Map/Reduce task completed --- PASS: TestBasic (25.60s) PASS ok mapreduce25.613s
In this part you will make the master handle failed workers. MapReduce makes this relatively easy because workers don't have persistent state. If a worker fails, any RPCs that the master issued to that worker will fail (e.g., due to a timeout). Thus, if the master's RPC to the worker fails, the master should re-assign the task given to the failed worker to another worker.
An RPC failure doesn't necessarily mean that the worker failed; the worker may just be unreachable but still computing. Thus, it may happen that two workers receive the same task and compute it. However, because tasks are idempotent, it doesn't matter if the same task is computed twice — both times it will generate the same output. So, you don't have to do anything special for this case. (Our tests never fail workers in the middle of task, so you don't even have to worry about several workers writing to the same output file.)
You don't have to handle failures of the master; we will assume it won't fail. Making the master fault-tolerant is more difficult because it keeps persistent state that would have to be recovered in order to resume operations after a master failure. Much of the rest of this course is devoted to this challenge.
Your implementation must pass the two remaining test cases in test_test.go. The first case tests the failure of one worker, while the second test case tests handling of many failures of workers. Periodically, the test cases start new workers that the master can use to make forward progress, but these workers fail after handling a few tasks. To run these tests:
$ go test -run Failure mapreduce/... -vet=off
Word count is a classical example of a Map/Reduce application, but it is not an application that many large consumers of Map/Reduce use. It is simply not very often you need to count the words in a really large dataset. For this application exercise, we will instead have you build Map and Reduce functions for generating an inverted index.
Inverted indices are widely used in computer science, and are particularly useful in document searching. Broadly speaking, an inverted index is a map from interesting facts about the underlying data, to the original location of that data. For example, in the context of search, it might be a map from keywords to documents that contain those words.
We have created a second binary in main/ii.go that is very similar to the wc.go you built earlier. You should modify mapF and reduceF in main/ii.go so that they together produce an inverted index. Running ii.go should output a list of tuples, one per line, in the following format:
$ go run ii.go master sequential pg-*.txt $ head -n5 mrtmp.iiseq A: 16 pg-being_ernest.txt,pg-dorian_gray.txt,pg-dracula.txt,pg-emma.txt,pg-frankenstein.txt,pg-great_expectations.txt,pg-grimm.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-metamorphosis.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt ABC: 2 pg-les_miserables.txt,pg-war_and_peace.txt ABOUT: 2 pg-moby_dick.txt,pg-tom_sawyer.txt ABRAHAM: 1 pg-dracula.txt ABSOLUTE: 1 pg-les_miserables.txtIf it is not clear from the listing above, the format is:
word: #documents documents,sorted,and,separated,by,commas
We will test your implementation's correctness with the following command, which should produce these resulting last 10 items in the index:
$ sort -k1,1 mrtmp.iiseq | sort -snk2,2 mrtmp.iiseq | grep -v '16' | tail -10 women: 15 pg-being_ernest.txt,pg-dorian_gray.txt,pg-dracula.txt,pg-emma.txt,pg-frankenstein.txt,pg-great_expectations.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-metamorphosis.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt won: 15 pg-being_ernest.txt,pg-dorian_gray.txt,pg-dracula.txt,pg-frankenstein.txt,pg-great_expectations.txt,pg-grimm.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-metamorphosis.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt wonderful: 15 pg-being_ernest.txt,pg-dorian_gray.txt,pg-dracula.txt,pg-emma.txt,pg-frankenstein.txt,pg-great_expectations.txt,pg-grimm.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt words: 15 pg-dorian_gray.txt,pg-dracula.txt,pg-emma.txt,pg-frankenstein.txt,pg-great_expectations.txt,pg-grimm.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-metamorphosis.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt worked: 15 pg-dorian_gray.txt,pg-dracula.txt,pg-emma.txt,pg-frankenstein.txt,pg-great_expectations.txt,pg-grimm.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-metamorphosis.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt worse: 15 pg-being_ernest.txt,pg-dorian_gray.txt,pg-dracula.txt,pg-emma.txt,pg-frankenstein.txt,pg-great_expectations.txt,pg-grimm.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt wounded: 15 pg-being_ernest.txt,pg-dorian_gray.txt,pg-dracula.txt,pg-emma.txt,pg-frankenstein.txt,pg-great_expectations.txt,pg-grimm.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt yes: 15 pg-being_ernest.txt,pg-dorian_gray.txt,pg-dracula.txt,pg-emma.txt,pg-great_expectations.txt,pg-grimm.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-metamorphosis.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt younger: 15 pg-being_ernest.txt,pg-dorian_gray.txt,pg-dracula.txt,pg-emma.txt,pg-frankenstein.txt,pg-great_expectations.txt,pg-grimm.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt yours: 15 pg-being_ernest.txt,pg-dorian_gray.txt,pg-dracula.txt,pg-emma.txt,pg-frankenstein.txt,pg-great_expectations.txt,pg-grimm.txt,pg-huckleberry_finn.txt,pg-les_miserables.txt,pg-moby_dick.txt,pg-sherlock_holmes.txt,pg-tale_of_two_cities.txt,pg-tom_sawyer.txt,pg-ulysses.txt,pg-war_and_peace.txt(this sample result is also found in main/mr-challenge.txt)
All submissions will be through the course's Inginious site, which will run and return your score in real time. You will need to sign in with you KAUST username and password; the same used for Portal and Blackboard. Once logged in, click assignment 1, which will ask you to upload a file. This file should be a zip that contains exclusively these files: schedule.go, and ii.go. You may submit multiple times. You will receive the highest score recorded prior to the assignment deadline. If you face any issues, especially students using Windows, please post a public question on Campuswire and include in the title "Assignment submission issue".
Before submitting, please run the full tests given above for both parts one final time. You are responsible for making sure your code works.
You will receive full credit for Part I if your software passes TestBasic from test_test.go (the test given in Part I) on our servers. You will receive full credit for Part II if your software passes the tests with worker failures (the Failure pattern to go test given in Part II) on our servers. You will receive full credit for Part II if your index output matches the correct output when run on our servers.
The final portion of your credit is determined by code quality tests, using the standard tools gofmt and go vet. You will receive full credit for this portion if all files submitted conform to the style standards set by gofmt and the report from go vet is clean for your mapreduce package (that is, produces no errors). If your code does not pass the gofmt test, you should reformat your code using the tool. You can also use the Go Checkstyle tool for advice to improve your code's style, if applicable. Additionally, though not part of the graded cheks, it would also be advisable to produce code that complies with Golint where possible.
This assignment is adapted from Princeton's COS-418, and previously from MIT's 6.824 course. Thanks to Mike Freedman, Frans Kaashoek, Robert Morris, and Nickolai Zeldovich for their support.
Last updated: 2023-08-31 22:00