Description
Hi there,
I have a graphql schema like this:
D has many S has many SW has one W has many C.
D is available via a root level query.
(The underlying SQL has the same set of entities and relationships between them).
Using apollo server (not sure if relevant), postgres and knex.
I started by adding a dataloader for C. I then made a root level query for D, asking for all the properties of each element in the above schema. Testing revealed that the caching worked - each distinct C was only queried for once. But batching didn't work - for each distinct C, there was a separate SELECT.
When debugging this, added batchScheduleFn: (callback) => setTimeout(callback, 100) to the dataloader for C. At that point, batching started to work.
This made me think that somehow the C resolution was taking place on different event loop tics. On a hunch, I added a dataloader for W, SW and S. (none for D, as there is just one root level query for D).
After doing this, batching worked for all of S, SW, W and C - even with the batchSchedule function removed.
So I wanted to ask - is this possibly a bug (either in dataloader or, for a separate matter, Apollo Server), or is it expected that batching at the leaf node C would not work unless all ancestor nodes are also resolved with dataloaders?
If this isn't a bug, I wonder if adding this information to the documentation would be possible, to help other people who find themselves in this situation - implementing dataloader only at the leaf of a deep schema, and having batching not work they way they expected it would?
Anyhow, thanks for the library and its ongoing maintenance :)