Optimization: only issue new dependency tasks for "parent" documents which contain a pinned relation(list) to "child" document

[sandreae]

In our dependency task we need to account for cases when a newly reduced document may have skipped over a view which is pinned from another document (this happens when operations are batched). In this case we want to issue new dependency tasks for any of these "parent" documents (the document which contained the pinned relation). We currently do this by issuing dependency tasks for all documents where their schema contains a pinned relation to the schema of the newly reduced document. We could improve this a bit by identifying documents which actually contain the pinned relation we are looking for. This can be done by querying operation values which are of type pinned_relation or pinned_relation_list where the value matches any child_document_view_id for the current document.

It's really similar to what we do when identifying if a blob should be garbage collected (basically, it has no "parent" documents) here:

aquadoggo/aquadoggo/src/db/stores/blob.rs

Lines 324 to 370 in a24b2d8

	/// Helper for getting the document ids of any document which relates to the specified document.
	///
	/// Optionally pass in a `SchemaId` to restrict the results to documents of a certain schema.
	async fn reverse_relations(
	pool: &AnyPool,
	document_id: &DocumentId,
	schema_id: Option<SchemaId>,
	) -> Result<Vec<String>, SqlStoreError> {
	let schema_id_condition = match schema_id {
	Some(schema_id) => format!("AND document_views.schema_id = '{}'", schema_id),
	None => String::new(),
	};
	query_scalar(&format!(
	"
	SELECT
	document_view_fields.document_view_id
	FROM
	document_view_fields
	LEFT JOIN
	operation_fields_v1
	ON
	document_view_fields.operation_id = operation_fields_v1.operation_id
	AND
	document_view_fields.name = operation_fields_v1.name
	LEFT JOIN
	document_views
	ON
	document_view_fields.document_view_id = document_views.document_view_id
	WHERE
	operation_fields_v1.field_type
	IN
	('pinned_relation', 'pinned_relation_list', 'relation', 'relation_list')
	{schema_id_condition}
	AND
	operation_fields_v1.value IN (
	SELECT document_views.document_view_id
	FROM document_views
	WHERE document_views.document_id = $1
	) OR operation_fields_v1.value = $1
	",
	))
	.bind(document_id.to_string())
	.fetch_all(pool)
	.await
	.map_err(|e| SqlStoreError::Transaction(e.to_string()))
	}

[sandreae]

It would be worth discussing whether this is actually an optimization, we would be swapping fewer tasks issued for a more complex (but maybe not that complex?) query.