Investigate how Gateway performs GraphQL query normalization

[tkornuta-semiotic]

No description provided.

[tkornuta-semiotic]

@aasseman can you please point @tumaysem to the AutoAgora processor code that does normalization "our way"?

[tumayt]

After some gateway code investigation. queries are matched against agora models with CostModel.cost_with_context method which maps to agora library. I found a good unit test to investigate further.
https://github.com/graphprotocol/agora/blob/bbe0c3c5882eb8685697f4615e285572706bde88/lang/src/tests.rs#L69

[tumayt]

If we assume query { a(skip: 10), b(bob: 5) } has two separate top fields as a and b. For each top field and predicate in the model, match_field is executed.

1- Matches top field names with predicate name
2- Check top field is not excluded due to query directives skip:if true or include:false
3- Matches predicate argument vs query variable names and values
4- Matches selection fields. This is checked recursively and some graphQL structures such as fragments.

For each top field first matched cost is aggregated to calculate the total cost.

[tumayt]

We should be using directly agora library because the code complexity and heavy graphQL language dependent match is prone to frequent changes. Thus required a patch to model providing a more detailed cost result with the matched predicates.

[tumayt]

After wrapping the agora and extend it to provide cost details, I have come to a conclusion that current processing of queries inside agora cost models is fully heuristic and without any normalization similar to semiotic-ai gql utils.
🔖 Cost details leads to a new idea. Gateway can provide list of tuples containing hash of selected cost model branch (top field in agora lang.) and cost to indexers with query instead of aggregated cost as price. Thus enables indexers to validate both provided price for a query (model validation) and payment verification (@aasseman) .

[tkornuta-semiotic]

should we try to upstream AutoAgora processing/normalization to the Gateway?

[aasseman]

should we try to upstream AutoAgora processing/normalization to the Gateway?

We can't upstream the Python code as Agora is in Rust. However @tumaysem is working on documenting (upstream issue) the disparities between how Agora compares queries, and how graph-node actually manipulates the GQL.