In 3.6 all writes to WT tables are journaled. Thus if a node crashes coming out of initial sync, WT recovery will put the data into its correct post-initial sync state.

However, the recoverable rollback project removes journaling from some tables. This introduces a window where a node that crashes and restarts may not recognize it had completed initial sync. Specifically, this crash window is after a node comes out of initial sync, but before its first checkpoint is taken.

By default, WT checkpoints every 60 seconds. This ticket is for notifying the checkpoint thread right when replication communicates a stable timestamp is available that is >= the initial sync timestamp. This change will not guarantee that seeing a node report as a secondary implies it is resilient to a crash requiring a resync.