chore: add non-cached benchmark to find the price of costs control

router-tests/protocol/integration_test.go

@@ -1580,6 +1580,7 @@ func BenchmarkSequentialBig(b *testing.B) {
 			res := xEnv.MakeGraphQLRequestOK(testenv.GraphQLRequest{
 				Query: bigEmployeesQuery,
 			})
+			// Not worth stopping the timer.
 			if len(res.Body) < 3000 {
 				b.Errorf("unexpected result %q, expecting \n\n%q", res.Body, bigEmployeesResponse)
 			}
@@ -1593,7 +1594,69 @@ func BenchmarkSequentialBigCostControl(b *testing.B) {
 			cfg.CostControl = &config.CostControl{
 				Enabled:           true,
 				Mode:              config.CostControlModeMeasure,
-				EstimatedListSize: 15,
+				EstimatedListSize: 5,
+				ExposeHeaders:     true,
+			}
+		},
+	}, func(b *testing.B, xEnv *testenv.Environment) {
+		b.SetBytes(int64(len(bigEmployeesResponse)))
+		b.ReportAllocs()
+		for b.Loop() {
+			res := xEnv.MakeGraphQLRequestOK(testenv.GraphQLRequest{
+				Query: bigEmployeesQuery,
+			})
+			b.StopTimer()
+			if len(res.Body) < 3000 {
+				b.Errorf("unexpected result %q, expecting \n\n%q", res.Body, bigEmployeesResponse)
+			}
+			estimated := res.Response.Header.Get(core.CostEstimatedHeader)
+			require.Equal(b, "4650", estimated)
+			actual := res.Response.Header.Get(core.CostActualHeader)
+			require.Equal(b, "189", actual)
+			b.StartTimer()
+		}
+	})
+}
+
+func BenchmarkSequentialBigNotCached(b *testing.B) {
+	testenv.Bench(b, &testenv.Config{
+		ModifyEngineExecutionConfiguration: func(cfg *config.EngineExecutionConfiguration) {
+			cfg.EnableNormalizationCache = false
+			cfg.EnableValidationCache = false
+			cfg.EnablePersistedOperationsCache = false
+			cfg.ExecutionPlanCacheSize = 0
+			cfg.OperationHashCacheSize = 0
+		},
+	}, func(b *testing.B, xEnv *testenv.Environment) {
+		b.SetBytes(int64(len(bigEmployeesResponse)))
+		b.ReportAllocs()
+		for b.Loop() {
+			res := xEnv.MakeGraphQLRequestOK(testenv.GraphQLRequest{
+				Query: bigEmployeesQuery,
+			})
+			// Not worth stopping the timer.
+			if len(res.Body) < 3000 {
+				b.Errorf("unexpected result %q, expecting \n\n%q", res.Body, bigEmployeesResponse)
+			}
+		}
+	})
+}
+
+func BenchmarkSequentialBigNotCachedCostControl(b *testing.B) {
+	testenv.Bench(b, &testenv.Config{
+		ModifyEngineExecutionConfiguration: func(cfg *config.EngineExecutionConfiguration) {
+			cfg.EnableNormalizationCache = false
+			cfg.EnableValidationCache = false
+			cfg.EnablePersistedOperationsCache = false
+			cfg.ExecutionPlanCacheSize = 0
+			cfg.OperationHashCacheSize = 0
+		},
+		ModifySecurityConfiguration: func(cfg *config.SecurityConfiguration) {
+			cfg.CostControl = &config.CostControl{
+				Enabled:           true,
+				Mode:              config.CostControlModeMeasure,
+				EstimatedListSize: 5,
+				ExposeHeaders:     true,
 			}
 		},
 	}, func(b *testing.B, xEnv *testenv.Environment) {
@@ -1603,9 +1666,15 @@ func BenchmarkSequentialBigCostControl(b *testing.B) {
 			res := xEnv.MakeGraphQLRequestOK(testenv.GraphQLRequest{
 				Query: bigEmployeesQuery,
 			})
+			b.StopTimer()
 			if len(res.Body) < 3000 {
 				b.Errorf("unexpected result %q, expecting \n\n%q", res.Body, bigEmployeesResponse)
 			}
+			estimated := res.Response.Header.Get(core.CostEstimatedHeader)
+			require.Equal(b, "4650", estimated)
+			actual := res.Response.Header.Get(core.CostActualHeader)
+			require.Equal(b, "189", actual)
+			b.StartTimer()
 		}
 	})
 }

router-tests/testenv/testenv.go

@@ -173,13 +173,15 @@ func Bench(b *testing.B, cfg *Config, f func(b *testing.B, xEnv *Environment)) {
 	}
 	b.StartTimer()
 	f(b, env)
+	b.StopTimer()
 	if cfg.AssertCacheMetrics != nil {
 		assertCacheMetrics(b, env, cfg.AssertCacheMetrics.BaseGraphAssertions, "")
 
 		for ff, v := range cfg.AssertCacheMetrics.FeatureFlagAssertions {
 			assertCacheMetrics(b, env, v, ff)
 		}
 	}
+	b.StartTimer()
 }
 
 const letterBytes = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ"

router/__schemas/graph.yaml

@@ -18,3 +18,5 @@ subgraphs:
   - name: employeeUpdates
     schema:
       file: ../../demo/pkg/subgraphs/employeeupdated/subgraph/schema.graphqls
+  - name: test1
+    routing_url: http://localhost:4006/graphql

router/bench-big-response.js

@@ -0,0 +1,79 @@
+import http from 'k6/http';
+import { check } from 'k6';
+
+// Load test for LARGE responses through the router.
+//
+// PREREQUISITES:
+//  1. test1 subgraph running on 4006 port: `cd ../demo && go run cmd/all/main.go`
+//  2. test1 present in the router's execution config
+//
+// GOTCHA: the leaf object is a shared singleton of identical lorem-ipsum text,
+// so gzip compresses the payload to almost nothing. To genuinely push 20-60 MB
+// over the wire we request `identity` (uncompressed) by default.
+// Set ENCODING=gzip to measure the compressed path instead.
+
+const URL = __ENV.URL || 'http://localhost:3002/graphql';
+const BIG_OBJECTS = __ENV.BIG_OBJECTS || '200';
+const NESTED_OBJECTS = __ENV.NESTED_OBJECTS || '72';
+const DEEPLY_NESTED = __ENV.DEEPLY_NESTED || '10';
+const ENCODING = __ENV.ENCODING || 'identity'; // 'identity' = full bytes on the wire; 'gzip' = compressed
+const DISCARD = (__ENV.DISCARD || 'false') === 'true';
+
+export const options = {
+  discardResponseBodies: DISCARD,
+  scenarios: {
+    big: {
+      executor: 'constant-vus',
+      vus: parseInt(__ENV.VUS || '1', 10),
+      duration: __ENV.DURATION || '60s',
+    },
+  },
+  thresholds: {
+    http_req_failed: ['rate<0.01'],
+  },
+};
+
+const query = `query Big($big: Int!, $nested: Int!, $deep: Int!) {
+  bigResponse(bigObjects: $big, nestedObjects: $nested, deeplyNestedObjects: $deep) {
+    nestedObjects {
+      deeplyNestedObjects {
+        aFieldOnDeeplyNestedObject bFieldOnDeeplyNestedObject cFieldOnDeeplyNestedObject
+        dFieldOnDeeplyNestedObject eFieldOnDeeplyNestedObject fFieldOnDeeplyNestedObject
+        gFieldOnDeeplyNestedObject hFieldOnDeeplyNestedObject iFieldOnDeeplyNestedObject
+        jFieldOnDeeplyNestedObject kFieldOnDeeplyNestedObject lFieldOnDeeplyNestedObject
+        mFieldOnDeeplyNestedObject nFieldOnDeeplyNestedObject oFieldOnDeeplyNestedObject
+        pFieldOnDeeplyNestedObject qFieldOnDeeplyNestedObject rFieldOnDeeplyNestedObject
+        sFieldOnDeeplyNestedObject tFieldOnDeeplyNestedObject uFieldOnDeeplyNestedObject
+        vFieldOnDeeplyNestedObject wFieldOnDeeplyNestedObject xFieldOnDeeplyNestedObject
+        yFieldOnDeeplyNestedObject zFieldOnDeeplyNestedObject
+      }
+    }
+  }
+}`;
+
+const body = JSON.stringify({
+  query,
+  operationName: 'Big',
+  variables: {
+    big: parseInt(BIG_OBJECTS, 10),
+    nested: parseInt(NESTED_OBJECTS, 10),
+    deep: parseInt(DEEPLY_NESTED, 10),
+  },
+});
+
+const headers = {
+  'Content-Type': 'application/json',
+  'Accept-Encoding': ENCODING,
+  'GraphQL-Client-Name': 'k6-bigresponse',
+  'GraphQL-Client-Version': '0.0.1',
+};
+
+export default function () {
+  const res = http.post(URL, body, { headers });
+  // When DISCARD=true, r.body is null — only assert status. When DISCARD=false,
+  // also assert the payload is large and error-free.
+  check(res, {
+    'status is 200': (r) => r.status === 200,
+    'no graphql errors': (r) => DISCARD || (r.status === 200 && r.body.includes('errors') === false)
+  });
+}

router/docs/Profiling.md

@@ -1,68 +1,78 @@
 # Profiling
 
-The router is set up with pprof, so users can run the router with `pprof` running, and get a thorough understanding of the active performance. 
+The router is set up with pprof, so users can run the router with `pprof` running, and get a
+thorough understanding of the active performance.
 
-> **We recommend that before adding a new feature, users should profile it to make sure that there aren't any surprise resource drains from the feature.**  
+> **We recommend that before adding a new feature, users should profile it to make sure that there
+aren't any surprise resource drains from the feature.**
 
 ## Running the router with pprof
-To run the router with pprof, follow the steps in [Contributing.md](../../CONTRIBUTING.md) (Local Development) to set up a local development environment, aside from running `make start-router`.
+
+To run the router with pprof, follow the steps in [Contributing.md](../../CONTRIBUTING.md) (Local
+Development) to set up a local development environment, aside from running `make start-router`.
 
 In the `router` directory, run the following command:
+
 ```shell
 go run cmd/router/main.go -pprof-addr=":6060"
 ```
 
 This will start the router with pprof running on port 6060.
 
-To run a solid workflow and get a sense of the routers performance, you can use [k6](https://grafana.com/docs/k6/latest/).
+To run a solid workflow and get a sense of the routers performance, you can
+use [k6](https://grafana.com/docs/k6/latest/).
 To do so, run:
+
 ```shell
 brew install k6
 k6 run bench.js
 ```
 
 This will run a benchmark on the router, and you can see the results in the terminal.
+
 ```
-     ✓ is status 200
-
-     checks.........................: 100.00% 40978 out of 40978
-     data_received..................: 294 MB  5.8 MB/s
-     data_sent......................: 114 MB  2.3 MB/s
-     http_req_blocked...............: avg=7.01µs  min=2µs    med=5µs     max=2.19ms   p(90)=6µs     p(95)=7µs
-     http_req_connecting............: avg=1.57µs  min=0s     med=0s      max=1.06ms   p(90)=0s      p(95)=0s
-     http_req_duration..............: avg=52.54ms min=8.68ms med=48.33ms max=235.82ms p(90)=93.36ms p(95)=106.39ms
-       { expected_response:true }...: avg=52.54ms min=8.68ms med=48.33ms max=235.82ms p(90)=93.36ms p(95)=106.39ms
-     http_req_failed................: 0.00%   0 out of 40978
-     http_req_receiving.............: avg=1.88ms  min=32µs   med=318µs   max=153.15ms p(90)=4.24ms  p(95)=6.37ms
-     http_req_sending...............: avg=24.66µs min=9µs    med=19µs    max=7.72ms   p(90)=29µs    p(95)=34µs
-     http_req_tls_handshaking.......: avg=0s      min=0s     med=0s      max=0s       p(90)=0s      p(95)=0s
-     http_req_waiting...............: avg=50.62ms min=8.42ms med=46.39ms max=230.97ms p(90)=90.72ms p(95)=103.53ms
-     http_reqs......................: 40978   816.55677/s
-     iteration_duration.............: avg=52.83ms min=8.95ms med=48.62ms max=236.13ms p(90)=93.65ms p(95)=106.71ms
-     iterations.....................: 40978   816.55677/s
-     vus............................: 99      min=2              max=99
-     vus_max........................: 100     min=100            max=100
-
-
-running (0m50.2s), 000/100 VUs, 40978 complete and 0 interrupted iterations
-default ✓ [======================================] 000/100 VUs  50s
+    ✓ is status 200
+
+    HTTP
+    http_req_duration..............: avg=2.04ms min=192µs    med=1.82ms max=27.02ms p(90)=3.5ms  p(95)=4.05ms
+      { expected_response:true }...: avg=2.04ms min=192µs    med=1.82ms max=27.02ms p(90)=3.5ms  p(95)=4.05ms
+    http_req_failed................: 0.00%   0 out of 1014298
+    http_reqs......................: 1014298 20284.858532/s
+
+    EXECUTION
+    iteration_duration.............: avg=2.12ms min=240.62µs med=1.89ms max=27.13ms p(90)=3.58ms p(95)=4.13ms
+    iterations.....................: 1014298 20284.858532/s
+    vus............................: 99      min=2            max=99
+    vus_max........................: 100     min=100          max=100
+
+    NETWORK
+    data_received..................: 7.3 GB  145 MB/s
+    data_sent......................: 2.8 GB  56 MB/s
 ```
 
-This can show you, for example, how many requests were sent, how long it took, and it can help us diagnose the proper configuration of the router for users. 
+This can show you, for example, how many requests were sent, how long it took, and it can help us
+diagnose the proper configuration of the router for users.
 
 ## Profiling the router
-There are many different things you can use pprof for, and we recommend reading the [pprof documentation](https://pkg.go.dev/net/http/pprof) to get a better understanding of what you can do with it.
+
+There are many different things you can use pprof for, and we recommend reading
+the [pprof documentation](https://pkg.go.dev/net/http/pprof) to get a better understanding of what
+you can do with it.
 
 As an example, to look at heap and memory usage, you can run the following commands:
 In a terminal, as the router is running, run:
+
 ```shell
 go tool pprof http://localhost:6060/debug/pprof/heap
 # or
 go tool pprof -http 127.0.0.1:6060 heap.out
 ```
 
-That will open a `pprof` shell, and in it, you can explore commands. Some useful ones are: `web` (which depends on graphviz, `brew install graphviz`), `top`, and `pdf`, which will give you different ways to look at the heap allocations.
+That will open a `pprof` shell, and in it, you can explore commands. Some useful ones are: `web` (
+which depends on graphviz, `brew install graphviz`), `top`, and `pdf`, which will give you different
+ways to look at the heap allocations.
 For example, `top20` will return:
+
 ```
 (pprof) top20
 Showing nodes accounting for 13458.23kB, 100% of 13458.23kB total
@@ -91,11 +101,11 @@ Showing top 20 nodes out of 69
 (pprof) 
 ```
 
-Then, you can also do `list <function>` to get a better understanding of where the memory is being allocated, line by line, in a function.
-
-In addition, we can also see how allocs work by running 
+Then, you can also do `list <function>` to get a better understanding of where the memory is being
+allocated, line by line, in a function.
 
 You can also run a profile for an amount of time (for example, 5 seconds), by running:
+
 ```
 go tool pprof ‘http://localhost:6060/debug/pprof/profile?seconds=5’
 ```
@@ -105,6 +115,8 @@ go tool pprof ‘http://localhost:6060/debug/pprof/profile?seconds=5’
 To use Pyroscope for continuous profiling of the router:
 
 1. Raise the infra-debug stack with `make infra-debug-up`
-2. Run the router with either `PYROSCOPE_ADDR=http://localhost:4040` or `-pyroscope-addr http://localhost:4040` to enable sending continuous profiling data to Pyroscope. You can view this data in Grafana.
+2. Run the router with either `PYROSCOPE_ADDR=http://localhost:4040` or
+   `-pyroscope-addr http://localhost:4040` to enable sending continuous profiling data to Pyroscope.
+   You can view this data in Grafana.
 3. Visit the drilldown profiles section in Grafana at `http://localhost:9300`
 4. Select the router from the service dropdown