One Content API request that selects a very large result can exhaust the memory of the whole Engine process. The request memory budget estimates how much memory the result of a single request retains and stops the request before it grows past a limit.

The budget is disabled by default.

Configuration

Set both thresholds, in bytes:

CONTEMBER_HTTP_REQUEST_MEMORY_BUDGET_WARN_BYTES=67108864
CONTEMBER_HTTP_REQUEST_MEMORY_BUDGET_MAX_BYTES=268435456

or in the server config:

server:
  http:
    requestMemoryBudget:
      warnBytes: 67108864
      maxBytes: 268435456

Option	Meaning
warnBytes	A request whose estimate exceeds this value is logged as a warning (Content request memory usage) with the estimate breakdown. The request continues.
maxBytes	A request whose estimate exceeds this value is stopped.

Both options must be positive integers, they must be set together, and warnBytes must not exceed maxBytes. The Engine refuses to start otherwise.

What is counted

Only the data a request selects is counted: rows fetched for queries, for relations, and for the node of a mutation result, plus the objects built from them and a reserve for serializing the response. Internal queries of a mutation (unique lookups, RETURNING rows) are not counted, so a large import is not limited by the number of its operations.

The value is an estimate, not a measurement of the process heap. Start with warnBytes only set low and maxBytes high, watch the warnings, and then pick the limit.

What the client receives

A query that exceeds maxBytes fails as a whole with HTTP 422. The error names the root field that exceeded the budget:

{
	"data": null,
	"errors": [
		{
			"message": "Request memory budget exceeded",
			"path": ["listArticle"],
			"extensions": { "code": "RESOURCE_EXHAUSTED" }
		}
	]
}

A mutation that exceeds maxBytes is rolled back and reports the failure in its own result, like any other failed mutation. Mutations that were committed earlier in the same request keep their results:

{
	"data": {
		"first": { "ok": true, "errors": [] },
		"second": { "ok": false, "errors": [{ "type": "ResourceExhausted", "message": "Request memory budget exceeded" }] }
	}
}

Once a request exceeds its budget, its remaining mutations are not started and report the same error. Inside a transaction, the transaction and every mutation field in it report the failure. Retrying the same request does not help; select less data (add limit, narrow the selection) or wrap dependent mutations in a transaction so that they succeed or fail together.

Exceeding the budget is not counted as an SQL error in the Engine metrics.