hackercup online competition - Virtual offsite

"""

    Hacker cup competition (Online coding judge system)
    * registered users
    * contest has a start time and end time. (2-3 hrs)
    * upon start, users get access to a certain # of problems (10)
    * users submit a source code for a problem, in any order
    * system validates the code and either accepts it or rejects it.
    * ranking or a leaderboard.'

"""

- Requirments:
Registrations (V2):
  register users (across globe) for a competitions (some previous... few current competitions)
  online coding competition
Competition:
  problems + test cases + critieria (memory/time etc) --> score
   
  
  non-functional requirements:
    - isolation
    - scalability
    - highly available
    - durability
   
   
estimation:
  - 1M users
  
  storage:
    10MB * 1M --> 10GB
   
  Bandwidth:
    1GB/sec
   
     
APIs
  GET /problems
  
  GET /problems/{problemId}
      {  
        description:
        case1:
        case2:
      }
  
  POST /submit
   
  /GET /leaderboard
  
  -- users
  -- registration


Datamodel:

problems
  - problemid
  - competitionId
  - descritpion

users
  - userid
  - name
  - contact
  
testcases
  - problemId
  - testcaseid
  - input
  - output

usersubmissions
  - userid
  - problemid
  - code (BLOB) --> NoSQL
  - status
  - result
  - memory
  - time
  -> submitted
  
testcaseResults
  - problemId
  - testcaseid
  - output
  - memory
  - time
  
SQL
  --> ACID
  --> meets our scaling requirements
So choosing SQL instead of NoSQL
  
  --> master-master for high availablility


backend components
- listservice --> storage
   
- submitservice  --> storage
    stores the code
    calls evaluateService
   
- EvaluateSystemManager --> run the code against test cases
      manages submissions
    --> executors
          --> heartbeat
          --> asks for a job
          --> executes a job

How does browser get notifications:
browser --> calling service (AJAX)
  
  option1:
    polling:
  
    serverpush:
      
    websocket:
      
Assuming that coders are submitting code in java
  lang: java
  
  
Submit:
  code
    --> store code with status: NOT_STARTED

   
   EvaluateSystemManager
      --> 1 master (passive)
      --> continously usersubmissions (look for NOT_STARTED) --> RUNNING --> COMPLETED
      --> in-memory QUEUE
      --> sends a msg to executor --> user-id, code, problem-id
      
      --> list of nodes available to run --> SQL
      --> maintain status of nodes  --> SQL
            
  Executor
    connection to SystemManager
    --> process--> heartbeat & status code --> EvaluateSystemManager
    --> spins up a process
        generate code
          submitted-code
        TEMPLATE-code
          generate class
          run-it
           
        --> output --> communicates to SQL --> saves the data
      
        --> event to compute
        
        
     
Leaderboard
  --> users --> submitted
  
  sort 'usersubmissions' by submited_time, cputime, memory asc
    --> top 100 users in board
   
    rank
      --> gets stats of a particular user
      --> get how many users ahead curr-user stats
      

  leaderboard
    --> uses websocket to update leaderboard
Feedback: need not be a web socket and results can be shown in eventual consistent manner. On very submission of code, leaderboard can change. it is compute intensitive to compute leaderboard for every submissions. Instead, this can be updated periodically.