As a newcomer to this code, I was reading callsites of RecoveryUnit::getPointInTimeReadTimestamp, like this one as if it were a normal getter function - one with no side effects that just returned the read timestamp. But it turns out that it actually opens a storage transaction in some cases!

That was unexpected. In addition there are other usages like here where I wonder if the author knew that this was an anticipated side effect. Personally, I would not have expected this line to sometimes open a snapshot.

Renaming this function something like establishPointInTimeReadTimestamp or something that sounded less like a passive getter might make this behavior less surprising to future users/readers.