I have been asked this question in an interview for a start up:

Design a counter for a website which tells how many visits happened to that website. A given website will have various webpages. How can we implement a counter for that.

I proposed using a counter variable which gets incremented every time there is a read request to that website. Then the interviewer gave me a signature of the following method and asked me to implement the following method which increments the count.
void incrementCount()

I suggested two approaches using:

1. use an atomic variable
2. use a lock

Then interviewer asked me following questions:

1. How do you take care of locking contention between the two methods:

   • getCount() which returns the count. This method is called from the thread which renders the page
   • incrementCount() which increments the count

2. What happens when the machine on which the above routines are running is crashed, then the count will be lost.

For (1), I suggested using a read-write lock
For (2), I suggested updating the values in a db(or filesystem) occasionally

He asked what if there is a crash just before the machine is trying to update the db. In that case the count will not be exact. I mentioned that these counts are mostly used for analytics and I think it is okay if these are not accurate.