Let's say a server in STARTUP2 is initial syncing from some other node. If something invokes restartCatalog on the sync source, every cursor on every collection is destroyed, as all of the CursorManagers are destructed when the Collection and Database objects go away. This will cause the initial syncing node to encounter CursorNotFound errors, and ultimately will end up with not all of the required data when exiting initial sync. This can manifest as a dbhash mismatch at the end of a test.

We should blacklist restart_catalog.js from any passthrough suite that involves initial sync. However, we should definitely keep it in at least the replica_sets_jscore_passthrough, sharding_jscore_passthrough and sharded_collections_jscore_passthrough, as it tests that various types of replicated/sharded collections survive a catalog restart.