We are considering approximating ACID behavior in a MongoDB replication set for one of our use cases. We want a write to hit all of our non-hidden replication set members so that a subsequent read from any member will definitely reflect the update. We found that the write concern setting does not consider which nodes are hidden, so we are going to tag our non-hidden instances and use a getLastError mode which specifies a write concern on the number of tagged members to hit (set to the number of non-hidden members).

We are concerned about potential downtime after running rs.reconfig(conf) to add the tags and getLastError mode. Is that just a metadata update that will happen in less than a few minutes? Or does it depend on our topology and specs? We have three non-hidden nodes, one hidden node, and 130GB of data.

Chris