thoughts on implementing Query Complexity Analysis?

[roippi]

This is a feature that's popping up in a few other framework implementations, and I think it's pretty killer - it lets you do a reasonable job at statically detecting pathologic queries.

There are some more complex ideas floating out there in the community, namely graphql/graphql-js#304. This could be used both for complexity/cost analysis in addition to query resolution optimization, though it seems more useful for optimization. Probably overkill here.

Has anyone already thought of implementing this here, or should I take a crack at it?

[viniychuk]

Hi @roippi , I started to work on it but the first priority is to get the docs updated. We can definitely discuss your ideas (let's say in gitter) so I can have as much input as possible.

[roippi]

Thanks. I have a working POC here: master...roippi:complexity_analysis_wip

You can implement a 'cost' function on fields, e.g.:

cost: function($args, $config, $childScore) {
  return 10 + $args['count'] * $childScore;
}

For a query:

{
  things (count:5) {
    someField
    someExpensiveSubQuery {
      someOtherField
    }
  }
}

And things would have access to the combined calculated score of all its subfields/subqueries.

Usage is pretty easy:

$processor = new Processor(new $mySchema());
$processor->setMaxComplexity(42);
// process normally, if complexity is exceeded it will return an error instead of trying to resolve the query

I'll hop into gitter to discuss. This is still a WIP so I'm open to feedback on the API, implementation, or test coverage 😄 .