Building a highly scalable, failure tolerant flash sales backend

Asked Mar 27 '18 at 09:53

Active Mar 27 '18 at 09:53

Viewed 134 times

I have concert tickets to sell, 1.000.000 of them. They are very popular, and as soon as the order button goes live, I will have a very high load of people trying to get them. Lets say 10.000 requests in parallel as worst case.

I want to sell at most 1.000.000 tickets, not 1.000.001. The actual order processing can be asynchronous using a queue, but the customer should have instant feedback if he got the ticket or not.

The system should survive the failure of a single server instance, so I want a fully distributed solution or at least master-slave.

My first thought: a CountDownLatch built using Zookeeper + Curator. Are there better approaches?

asked Mar 27 '18 at 09:53

Michael Böckling

7,341
6
55
76

There's also a distributed `CountDownLatch` in Apache Ignite. (No experience with it, just googled "distributed CountDownLatch".) – lexicore Mar 27 '18 at 10:48
Saw that, but found no Jepsen test results. Might also be overkill, as we require no persistence. – Michael Böckling Mar 27 '18 at 11:06
As there could be many solution, I would suggest you could try `redis`'s increment with **automic lua** script. – Pralgomathic Apr 06 '18 at 15:27

Building a highly scalable, failure tolerant flash sales backend

0 Answers0