There is nothing specific to replication in the Stripe issue. They noticed a well-known problem which is that, with large datasets and some categories of hardware, forking to snapshot to RDB takes a long time (can be a few seconds) and blocks the Redis process. This also happens on single-instance Redis.
EDIT: actually I am talking about the second problem mentioned in their post here: https://stripe.com/blog/game-day-exercises-at-stripe (latency spikes). The first problem (data loss) is related to replication.
EDIT: actually I am talking about the second problem mentioned in their post here: https://stripe.com/blog/game-day-exercises-at-stripe (latency spikes). The first problem (data loss) is related to replication.