Opentelemetry instrumentation for graphql-batch

[toneymathews]

TL;DR GraphQL Batch is a query batching executor for the GraphQL Ruby gem. It’s used to solve n+1 query problems. As maintainers of GraphQL Batch, we would like to implement tracing so that a shared implementation may be available within open-telemetry/opentelemtry-ruby-contrib.

Defining the problem space and how the gem solves this problem

By default, every field in a GraphQL query is resolved individually on each object, in total isolation. Consider the following GraphQL query, assuming a SQL based data source with pagination omitted for simplicity:

{
  orders(first: 50) {
    customer {
	name
    }
  }
}

On this query, first the orders field is resolved, which generates a SQL query for 50 orders from the database. Then for each individual order, the customer field is resolved, which generates an additional SQL query to load the customer for that order. In total, 51 SQL queries are executed (one to load the orders, and one per order to load the customer). This is an extremely costly way to resolve this data.

GraphQL Batch exposes a GraphQL::Batch::Loader interface. It exposes a #perform method that is used by a custom batch loader such as the CustomerLoader that aggregates the orders and performs a single SQL query.

class CustomerLoader < GraphQL::Batch::Loader
  def perform(orders)
    customer_ids = orders.map(&:customer_id)
    # Query to fetch all customers.
    customers_by_id = Customers.where(id: customer_ids).index_by(&:id)

    orders.each do |order| 
      customer = customers_by_id[order.customer_id]
      fulfill(order, customer)  # For every order, fan out the corresponding customer.
    end
  end
end

Loaders based on GraphQL::Batch::Loader may be generic such as RecordLoader and AssociationLoader or a custom one such as the CustomerLoader above, for a specific purpose.

What we propose

Add a batch loader within open-telemetry/opentelemtry-ruby-contrib that is traced so that a shared implementation is available. The perform method holds code that makes calls to remote data sources. We want to add tracing for the perform method to visualize and measure performance of a batch loader.

If it makes sense to do so, we can add a pull request to add instrumentation for graphql-batch as the next step for further feedback.

[arielvalentin]

Speaking for myself, adding instrumentation for graphql-batch would be a welcome addition to this repo.

I am looking forward to your contribution!

[github-actions]

👋 This issue has been marked as stale because it has been open with no activity. You can: comment on the issue or remove the stale label to hold stale off for a while, add the keep label to hold stale off permanently, or do nothing. If you do nothing this issue will be closed eventually by the stale bot.

[arielvalentin]

@toneymathews Any chance you may have time to work on this?

[toneymathews]

Apologies, I have to prioritize work in other areas and hence will not be able to focus on this piece of work soon.

[github-actions]

👋 This issue has been marked as stale because it has been open with no activity. You can: comment on the issue or remove the stale label to hold stale off for a while, add the keep label to hold stale off permanently, or do nothing. If you do nothing this issue will be closed eventually by the stale bot.