We often don't know that server startup is slow with a large number of collections until a customer observes a problem. It is a pretty common workflow for backup use cases and upgrade/downgrade.

We should consider adding performance workloads for the following common operations with a large number of collections (at least 1000):

• Starting a replica set node after a clean shutdown
• Starting a replica set node after an unclean shutdown
• Starting a replica set node in standalone mode, and starting a standalone node in replica set mode
  • Both of these steps are required for backup procedures.
• Upgrading a replica set node from the last stable version to the current version
• Consider further separating tests into some that write to a subset of collections vs. all of them