When running the YCSB 100 update workload, we can observe that checkpoint creation is reading in pages from the history store in order to transfer content to them. Those reads cause the checkpoint to wait for a round-trip to SLS, which slows checkpoint creation meaningfully.

We could identify which history store pages are going to be useful in advance of the checkpoint and pre-fetch them outside of the critical path, which would allow checkpoint to complete quickly and not be disruptive.

Note that this should be done after WT-16443, which aims to avoid evicting the history store pages in the first place. If that is successful, the work described here might not be necessary.