SERVER-124309 adds more startup tests for us . However, as we increase the number of collections, we face more standby lag that causes the test to occasionally fail due to timeouts. We should tweak the test structure so that we can try avoid this. Since the standby lag exists in the creation of the collections, one solution could be to create the collections with just the primary, then startup the standby to read from the latest checkpoint instead of having to apply the oplog one by one. We should also investigate if we can just shut down the standby altogether, and only restart our singular node, so we can avoid the issue of the standby lag.

Here are some runs with the creation startup, with varying success.

• all tasks complete
• fail on 25k standby
• fail on 5k, 25k, 50k standby
• fail on 25k, 50k standby