Experiment requests should be ordered

[guseggert]

IIUC, each experiment gets its own SQS queue. Standard queues are sometimes ordered, but not always. To preserve order when replaying traffic, we should use FIFO queues (or Kinesis streams?). This would also prevent duplicate messages.

[iand]

All targets in an experiment get the requests in the same order since dealgood reads from the queue sequentially and sends that request to each target (or drops it if the target is overloaded). So in a single experiment all nodes see the same sequence of requests. However another independent experiment may see a different ordering.

[guseggert]

Yes but SQS queues themselves do not guarantee order unless they are FIFO queues.