Status: 3+YOE , Mtech from Tier-1
Position: SSE at MNC
Location: Bengaluru
Date: Last week Sep to Nov

Virtual Onsite Round 1 (1 hour):

• Algorithm question
  https://leetcode.com/problems/course-schedule-ii/
  Interviewer was very friendly , Aced this round

Virtual Onsite Round 2 (1 hour):

• Algorithm question
  https://leetcode.com/problems/all-nodes-distance-k-in-binary-tree/
  Interviewer was very friendly , solved optimally with his little help

gap of a month. (No response to follow up emails.)

Virtual Onsite Round 3 (1 hour):
After past work experience discussion of 15 mins.

• Algorithm question
  Design LFU cache.

Interviewer was very friendly , could not finish coding it completely , but was able to give optimal data structure and write insert API with all edge cases.

Virtual Onsite Round 4 (1 hour):

Explain current role. and discussion around current job.

Algorithm question-

1. Delete nodes from a linkedlist which has non unique values both recusively and iteratively. (could not complete both with all edge cases , panicked).

2. Extension to above , how would you delete if deleteNode API called in multple threads (Brain responded OutOfBound !! ).

   • Had a discussion about what code should be in critical section.
     Which locks to use , why spin lock , why semaphore.

   • I asked for a hint : He gave to use fine grained locking on every node. (My brain hanged instantly !!!!!!)
     Couldnt understand how to use locks on every node.

Moved further.

1. How to count number of users connected to a service , service has n nodes in a distributed network.

   • We can add a server which authenticates and adds sessions of users in RDBMS . Every request to service is accepted by this server and it redirects to different nodes.
     we can query database to know number of unique users.

2. What data structure to use know unique users?

   • Create a set or Relational database at the server.

3. what if we cannot use a seperate machine or server and can use only nodes to which users are connected.

   • Make one node a master and all other nodes ping this on adding and deleting a session. It results in lot of RPC calls.

Got a rejection mail within 10 mins. :(

Can someone please help to understand what could be optimal solutions or discussions for fourth round. What could have been done better to solve multithreading and distributed systems question.
I know my answers were poor.

Thanks.