Skip to content

challenge-07-DistributedSystem.md

Latest commit

 

History

History
34 lines (23 loc) · 2.2 KB

challenge-07-DistributedSystem.md

File metadata and controls

34 lines (23 loc) · 2.2 KB

Challenge 07 -- Manual Distributed System Simulation

The Challenge

  • Design and implement a simple distributed booking system using Python’s socket library. Simulate multiple nodes (instances of your HTTP server) and manually handle the distribution of bookings across these nodes with basic load balancing (round-robin or random).
  • This task requires understanding of distributed system concepts without relying on any external libraries or frameworks

The UML Diagram

The Solution

Updated Files

  • created new classes DistributedBookingSystems, computational Node and NodeConnection in distibutedSystem
  • updated HttpClient-testing slightly
  • update httpServer
  • create fancy names for nodes in via CreativeNamer class
    • leading to welcome messages like Hello. This is serendipitous fantastical Galaxy. Let's go!
  • create manyClients to run multiple httpClients

Basic Idea

  • hardware-abstraction = Node (runs instance of HTTP-server)
  • LoadBalancer -> RoundRobin, Random, WeightedRoundRobin -> decides where to send request
  • DistributedBookingSystem -> distributes requests among nodes, thus is a relay that combines client socket with node (server) socket
  • central DB, connected to by all nodes individually -> concurrency!

Problems

  • it took me quite a while to make asyncio (server-side) and sockets (client side) work properly together as closing the connection on client side is/was a problem for me
  • finally introduced a hard waiting time and re-init of a connection on server side --> but: slows down the system response speed -> including the query respone is $<=150ms$ right now, but if a big query or inserting operation is used line await asyncio.wait_for(asyncio.gather(client_task, node_task), timeout=0.2) in DistributedBookingSystem.handle_client(...) will become a big problem
  • when running 10 clients the probability is high that the same node has the same DB-entry/query in its LRUCache, but when checking the cli all reads are from DB -> why???