Distributed system design

https://github.com/donnemartin/system-design-primer

You trade-off between high availability and consistency. Strong consistency will require dual writes at data layer or reads going to leader/master partition to guarantee read-after-write symantics. This means that some clients requests may take longer to serve or becomes unavailable as system can handle fewer qps of traffic and have more failures points in the system. For example, you could use MySQL or Oracle db to store records. You can still partition into multiple dbs, but you lose availability since all writes and reads wait until they are propagated to slave db instances. MySQL and Oracle both have ACID properties by default. If you're using a non-acid db such as Cassandra, you'll have to ensure that each write is written to at least two places for durability, and reads are always sent to leader/master. Again you'll lose on availability in this case. You could also use a consensus algorithm such as paxos to drive reads and writes, again trading off availability for consistency. The primary trait an interviewer is looking for with such interviews is whether you're aware and considering the different trade-offs. Most interviewer know that you may not have "working" knowledge of everything, but having the necessary theoretical knowledge is important. Grokking the system design course (designgurus.org) talks about this in brief and also has an example solution for a ticket booking system.

Are you familiar with the CAP theorem?

The book designing data intensive applications.

Read grokking the system design in educative io