Let's say I have a replicated storage system and clients make requests to multiple active servers.
If a client wants to change some data then this request is received by only one server and the change is replicated to all other servers. All servers acknowledge the change to the server the request originated from, and this one commits the change and confirms to the client.
But when exactly is the change activated/committed at all servers? If I had a client asking for the data being changed just two times in a row very quickly at different servers it might first get an answer with the new data and then get the old data from another server that just was not finished committing it. Is that not a problem? I assume after I get back the new data from one server I should never again get the old data from any server.
Would you lock the data under change at one server and all requests from clients go to this one server until every server has committed the change? So that answers to the outside are always consistent for that data?