Add extensions.service for all subgraph errors

.changesets/feat_error_extensions_service.md

@@ -0,0 +1,29 @@
+### Add `extensions.service` for all subgraph errors ([PR #6191](https://github.com/apollographql/router/pull/6191))
+
+If `include_subgraph_errors` is `true` for a particular subgraph, all errors originating in this subgraph will have the subgraph's name exposed as a `service` extension.
+
+For example, if subgraph errors are enabled, like so:
+```yaml title="router.yaml"
+include_subgraph_errors:
+  all: true # Propagate errors from all subgraphs
+```
+Note: This option is enabled by default in the [dev mode](./configuration/overview#dev-mode-defaults).
+
+And this `products` subgraph returns an error, it will have a `service` extension:
+
+```json
+{
+  "data": null,
+  "errors": [
+    {
+      "message": "Invalid product ID",
+      "path": [],
+      "extensions": {
+        "service": "products"
+      }
+    }
+  ]
+}
+```
+
+By [@IvanGoncharov](https://github.com/IvanGoncharov) in https://github.com/apollographql/router/pull/6191

apollo-router/src/axum_factory/snapshots/apollo_router__axum_factory__tests__defer_is_not_buffered.snap

@@ -26,7 +26,8 @@ expression: parts
           "@"
         ],
         "extensions": {
-          "code": "FETCH_ERROR"
+          "code": "FETCH_ERROR",
+          "service": "reviews"
         }
       }
     ],

apollo-router/src/plugins/cache/snapshots/apollo_router__plugins__cache__tests__missing_entities-2.snap

@@ -28,7 +28,10 @@ expression: response
         "currentUser",
         "allOrganizations",
         2
-      ]
+      ],
+      "extensions": {
+        "service": "orga"
+      }
     }
   ]
 }

apollo-router/src/plugins/include_subgraph_errors.rs

@@ -42,36 +42,47 @@ impl Plugin for IncludeSubgraphErrors {
     }
 
     fn subgraph_service(&self, name: &str, service: subgraph::BoxService) -> subgraph::BoxService {
-        // Search for subgraph in our configured subgraph map.
-        // If we can't find it, use the "all" value
-        if !*self.config.subgraphs.get(name).unwrap_or(&self.config.all) {
-            let sub_name_response = name.to_string();
-            let sub_name_error = name.to_string();
-            return service
-                .map_response(move |mut response: SubgraphResponse| {
-                    if !response.response.body().errors.is_empty() {
+        // Search for subgraph in our configured subgraph map. If we can't find it, use the "all" value
+        let include_subgraph_errors = *self.config.subgraphs.get(name).unwrap_or(&self.config.all);
+
+        let sub_name_response = name.to_string();
+        let sub_name_error = name.to_string();
+        return service
+            .map_response(move |mut response: SubgraphResponse| {
+                let errors = &mut response.response.body_mut().errors;
+                if !errors.is_empty() {
+                    if include_subgraph_errors {
+                        for error in errors.iter_mut() {
+                            error
+                                .extensions
+                                .entry("service")
+                                .or_insert(sub_name_response.clone().into());
+                        }
+                    } else {
                         tracing::info!("redacted subgraph({sub_name_response}) errors");
-                        for error in response.response.body_mut().errors.iter_mut() {
+                        for error in errors.iter_mut() {
                             error.message = REDACTED_ERROR_MESSAGE.to_string();
                             error.extensions = Object::default();
                         }
                     }
-                    response
-                })
-                // _error to stop clippy complaining about unused assignments...
-                .map_err(move |mut _error: BoxError| {
+                }
+
+                response
+            })
+            .map_err(move |error: BoxError| {
+                if include_subgraph_errors {
+                    error
+                } else {
                     // Create a redacted error to replace whatever error we have
                     tracing::info!("redacted subgraph({sub_name_error}) error");
-                    _error = Box::new(crate::error::FetchError::SubrequestHttpError {
+                    Box::new(crate::error::FetchError::SubrequestHttpError {
                         status_code: None,
                         service: "redacted".to_string(),
                         reason: "redacted".to_string(),
-                    });
-                    _error
-                })
-                .boxed();
-        }
-        service
+                    })
+                }
+            })
+            .boxed();
     }
 }
 
@@ -104,7 +115,7 @@ mod test {
     use crate::Configuration;
 
     static UNREDACTED_PRODUCT_RESPONSE: Lazy<Bytes> = Lazy::new(|| {
-        Bytes::from_static(r#"{"data":{"topProducts":null},"errors":[{"message":"couldn't find mock for query {\"query\":\"query($first: Int) { topProducts(first: $first) { __typename upc } }\",\"variables\":{\"first\":2}}","path":[],"extensions":{"test":"value","code":"FETCH_ERROR"}}]}"#.as_bytes())
+        Bytes::from_static(r#"{"data":{"topProducts":null},"errors":[{"message":"couldn't find mock for query {\"query\":\"query($first: Int) { topProducts(first: $first) { __typename upc } }\",\"variables\":{\"first\":2}}","path":[],"extensions":{"test":"value","code":"FETCH_ERROR","service":"products"}}]}"#.as_bytes())
     });
 
     static REDACTED_PRODUCT_RESPONSE: Lazy<Bytes> = Lazy::new(|| {

apollo-router/src/services/supergraph/snapshots/apollo_router__services__supergraph__tests__errors_from_primary_on_deferred_responses-2.snap

@@ -24,7 +24,10 @@ expression: stream.next_response().await.unwrap()
           "path": [
             "computer",
             "errorField"
-          ]
+          ],
+          "extensions": {
+            "service": "computers"
+          }
         }
       ]
     }

apollo-router/src/services/supergraph/snapshots/apollo_router__services__supergraph__tests__errors_on_deferred_responses-2.snap

@@ -17,7 +17,10 @@ expression: stream.next_response().await.unwrap()
           "message": "error user 0",
           "path": [
             "currentUser"
-          ]
+          ],
+          "extensions": {
+            "service": "user"
+          }
         }
       ]
     }

apollo-router/src/services/supergraph/snapshots/apollo_router__services__supergraph__tests__errors_on_incremental_responses-2.snap

@@ -23,7 +23,10 @@ expression: stream.next_response().await.unwrap()
             "activeOrganization",
             "suborga",
             0
-          ]
+          ],
+          "extensions": {
+            "service": "orga"
+          }
         }
       ]
     },
@@ -56,7 +59,10 @@ expression: stream.next_response().await.unwrap()
             "activeOrganization",
             "suborga",
             2
-          ]
+          ],
+          "extensions": {
+            "service": "orga"
+          }
         }
       ]
     }

apollo-router/src/services/supergraph/snapshots/apollo_router__services__supergraph__tests__errors_on_nullified_paths.snap

@@ -20,7 +20,8 @@ expression: stream.next_response().await.unwrap()
         "bar"
       ],
       "extensions": {
-        "code": "NOT_FOUND"
+        "code": "NOT_FOUND",
+        "service": "S2"
       }
     }
   ]

apollo-router/src/services/supergraph/snapshots/apollo_router__services__supergraph__tests__missing_entities.snap

@@ -18,7 +18,10 @@ expression: stream.next_response().await.unwrap()
       "path": [
         "currentUser",
         "activeOrganization"
-      ]
+      ],
+      "extensions": {
+        "service": "orga"
+      }
     }
   ]
 }

apollo-router/tests/integration/batching.rs

@@ -147,6 +147,8 @@ async fn it_batches_with_errors_in_single_graph() -> Result<(), BoxError> {
         - errors:
             - message: expected error in A
               path: []
+              extensions:
+                service: a
         - data:
             entryA:
               index: 2
@@ -200,9 +202,13 @@ async fn it_batches_with_errors_in_multi_graph() -> Result<(), BoxError> {
         - errors:
             - message: expected error in A
               path: []
+              extensions:
+                service: a
         - errors:
             - message: expected error in B
               path: []
+              extensions:
+                service: b
         - data:
             entryA:
               index: 2
@@ -256,6 +262,7 @@ async fn it_handles_short_timeouts() -> Result<(), BoxError> {
               path: []
               extensions:
                 code: REQUEST_TIMEOUT
+                service: b
         - data:
             entryA:
               index: 1
@@ -264,6 +271,7 @@ async fn it_handles_short_timeouts() -> Result<(), BoxError> {
               path: []
               extensions:
                 code: REQUEST_TIMEOUT
+                service: b
         "###);
     }
 
@@ -331,16 +339,19 @@ async fn it_handles_indefinite_timeouts() -> Result<(), BoxError> {
               path: []
               extensions:
                 code: REQUEST_TIMEOUT
+                service: b
         - errors:
             - message: Request timed out
               path: []
               extensions:
                 code: REQUEST_TIMEOUT
+                service: b
         - errors:
             - message: Request timed out
               path: []
               extensions:
                 code: REQUEST_TIMEOUT
+                service: b
         "###);
     }
 
@@ -568,6 +579,7 @@ async fn it_handles_cancelled_by_coprocessor() -> Result<(), BoxError> {
               path: []
               extensions:
                 code: ERR_NOT_ALLOWED
+                service: a
         - data:
             entryB:
               index: 0
@@ -576,6 +588,7 @@ async fn it_handles_cancelled_by_coprocessor() -> Result<(), BoxError> {
               path: []
               extensions:
                 code: ERR_NOT_ALLOWED
+                service: a
         - data:
             entryB:
               index: 1
@@ -725,6 +738,7 @@ async fn it_handles_single_request_cancelled_by_coprocessor() -> Result<(), BoxE
               path: []
               extensions:
                 code: ERR_NOT_ALLOWED
+                service: a
         - data:
             entryB:
               index: 2

apollo-router/tests/integration/snapshots/integration_tests__integration__traffic_shaping__subgraph_rate_limit-2.snap

@@ -2,4 +2,4 @@
 source: apollo-router/tests/integration/traffic_shaping.rs
 expression: response
 ---
-"{\"data\":null,\"errors\":[{\"message\":\"Your request has been rate limited\",\"path\":[],\"extensions\":{\"code\":\"REQUEST_RATE_LIMITED\"}}]}"
+"{\"data\":null,\"errors\":[{\"message\":\"Your request has been rate limited\",\"path\":[],\"extensions\":{\"code\":\"REQUEST_RATE_LIMITED\",\"service\":\"products\"}}]}"

apollo-router/tests/integration/snapshots/integration_tests__integration__traffic_shaping__subgraph_timeout.snap

@@ -2,4 +2,4 @@
 source: apollo-router/tests/integration/traffic_shaping.rs
 expression: response
 ---
-"{\"data\":null,\"errors\":[{\"message\":\"Request timed out\",\"path\":[],\"extensions\":{\"code\":\"REQUEST_TIMEOUT\"}}]}"
+"{\"data\":null,\"errors\":[{\"message\":\"Request timed out\",\"path\":[],\"extensions\":{\"code\":\"REQUEST_TIMEOUT\",\"service\":\"products\"}}]}"