Cost and throttling - Darpan

Darpan works out a query’s cost before reading anything. A query over a limit is rejected without touching the database, with an error code that says which limit.

Limits

Limit	Default	Error code
Cost of one query	1000	`COST_EXCEEDED`
Depth (nesting of objects)	6	`DEPTH_EXCEEDED`
`first` or `last` on a connection	100	`FIRST_TOO_LARGE`
A connection without `first` or `last`	not allowed	`FIRST_REQUIRED`
Time for one request	30 seconds	`DEADLINE_EXCEEDED`

How a query is costed

Selection	Cost
A scalar field	1
An object field	1 + the cost of its selection
A connection	`first` × (1 + the cost of one node)
A service-backed query (`auditReport`, `auditFinding`)	25 + the cost of its selection
A nested list inside an audit document (`categories`, `statements`, `findings`, `records`)	1 + the cost of one element: the document is already built, so its lists are not multiplied
A `query:` search key without a database index	+50

A full auditReport query, every field, costs well under 1000.

Throttle

Each API key has a bucket of 1000 points that refills at 50 points a second. A query’s cost is taken from the bucket when it runs. A query that costs more than the bucket holds is rejected with THROTTLED; wait for the bucket to refill. Every response reports the cost and the bucket:

"extensions": {
  "cost": {
    "requestedQueryCost": 34,
    "actualQueryCost": 34,
    "throttleStatus": { "maximumAvailable": 1000, "currentlyAvailable": 966, "restoreRate": 50 }
  }
}

Your administrator may set different limits for a key.