Shards are not splitted into smaller ones

[tuk]

@ajbeamon

The remaining two steps are what I referred to when I mentioned deferred cleanup. The first of these is something we call lazy deletion, where we traverse the sub-trees removed in the first step and properly free each page. Among other things, this puts each page onto the free list and makes it available for reuse when new pages need to be allocated. See https://github.com/apple/foundationdb/blob/f418a85390734373b74f022ae6a61bed23eb92ee/fdbserver/sqlite/btree.c#L7073 .

The last step is what we call vacuuming, and it’s doing what you describe; namely it fills in the holes in the file with pages from the end and then truncates off the empty pages. See https://github.com/apple/foundationdb/blob/f418a85390734373b74f022ae6a61bed23eb92ee/fdbserver/sqlite/btree.c#L3042

@ajbeamon If new writes are continuously coming (along with deletes) will the vaccum back off? If yes how does setting SPRING_CLEANING_VACUUMS_PER_LAZY_DELETE_PAGE changes the behavior as discussed here?

FoundationDB Version - 6.2.30.