Incorrect arguments ExecutionParams in graphql should throw proper GraphQL errors

[schrockn]

In create_execution_params:

def create_execution_params(graphene_info, graphql_execution_params):

    preset_name = graphql_execution_params.get('preset')
    selector = PipelineSelector.from_graphql_input(
        graphene_info.context, graphql_execution_params['selector']
    )
    if preset_name:
        check.invariant(
            not graphql_execution_params.get('environmentConfigData'),
            'Invalid ExecutionParams. Cannot define environment_dict when using preset',
        )
        check.invariant(
            not graphql_execution_params.get('mode'),
            'Invalid ExecutionParams. Cannot define mode when using preset',
        )

        check.invariant(
            not selector.solid_subset,
            'Invalid ExecutionParams. Cannot define selector.solid_subset when using preset',
        )

We do invariant checks that fail when we pass in properly formatted graphql. We should be throwing proper GraphQL errors. "check" errors are the moral equivalent of something akin to a typecheck failure, where as the errors above are more "business domain" errors. The computation has completed normally and has failed in an expected manner.

The fact that this is so odd is exposed here:

    def test_basic_start_pipeline_execution_with_preset_failure(self, graphql_context):
        with pytest.raises(check.CheckError):
            execute_dagster_graphql(
                graphql_context,
                START_PIPELINE_EXECUTION_QUERY,
                variables={
                    'executionParams': {
                        'selector': {'name': 'csv_hello_world', 'solidSubset': 'sum_sq_solid'},
                        'preset': 'test_inline',
                    }
                },
            )

[helloworld]

For a bit more additional context, we want to introduce a new error type, similar to the PresetNotFoundError that's right after the code snippet defined above.

  if not external_pipeline.has_preset(preset_name):
            raise UserFacingGraphQLError(
                graphene_info.schema.type_named('PresetNotFoundError')(
                    preset=preset_name, selector=selector
                )
            )

Instead of doing check.invariant, we'll want to raise a new UserFacingGraphQLError with a type describing this class of error where both a preset and one of environmentConfigData, mode or solidSubset is defined.