Right now the initial sync process will drop all databases and restart upon losing the connectivity to the sync source. It can be a very big issue for big databases syncing from another DC.

There are a few places where we can improve the behavior:

• if we are in the index build phase, we don't really care about the connection as data has already been fetched
• if a collection has been already successfully fetched, there is no point in dropping it and fetching it all over again if there was a temporary connectivity problem.

Possibly there are other places, too. But these should be fairly simple and a good start.