Why

While monitoring operations is a great start at monitoring our GQL Server for inefficiencies, the main culprits for any performance issues lie with individual fields that can degrade over time and can be individually responsible for breaching SLOs of particular operations.

The page for individual fields only mentions Total calls and RPM which can become time consuming to find inefficiencies in the operations and in our schema.

What

GraphQL Hive should report SLO metrics for fields that are in the main or different sub-schema, so we can monitor if a particular field in a stitched-in schema (called Service in Hive’s UI) is the culprit of any performance issues.

Currently, Hive has great support for metrics for unique operations that includes the following; however, these metrics are NOT available for individual fields.

We want to see the following for fields in our graph schema:

• p99
• p95
• p90
• Success Rate
• Failure Rate

Additionally, if possible, we would also want to know how long each field took, when looking at an individual operation.

Seems like with Apollo Studio / Graph OS, we can sort fields by latency or error rates.

Is this something possible with supported GraphQL servers and Hive together?

Example

Given a page for a particular operation with fields that can come from a different sub-schema.

• cart comes from cart-service (stitched)
• user comes from user-service (stitched)

query UserCart {
  cart {
    productCount
  }
  user {
    name
    email
  }
}

It would be helpful to see the following information, which helps us see that cart in particular is taking the longest for this operation.

• cart; p50: 50ms, p90: 100ms, p99: 200ms
• user: p50: 10ms, p90: 20ms, p99: 40ms
• <any_nested_field?>: p50: 5ms, p90: 8ms, p99: 12ms