feat: ability to set dedupe globally for io (#2036)

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
Co-authored-by: meskill <8974488+meskill@users.noreply.github.com>
Co-authored-by: Tushar Mathur <tusharmath@gmail.com>

.github/workflows/nginx-benchmark.yml

@@ -31,7 +31,8 @@ jobs:
 
       - name: Run Tailcall
         run: |
-          TAILCALL_LOG_LEVEL=error ./target/release/tailcall start ci-benchmark/nginx-benchmark.graphql &
+          TAILCALL_LOG_LEVEL=error ./target/release/tailcall start ci-benchmark/nginx-benchmark.graphql > tailcall.log 2>&1 &
+          echo $! > tailcall_pid.txt
 
       - name: Install Nginx
         run: |
@@ -59,7 +60,19 @@ jobs:
         name: Run Wrk
         working-directory: ci-benchmark
         run: |
-          wrk -d 30 -t 4 -c 100 -s wrk.lua http://localhost:8000/graphql > wrk-output.txt
+          wrk -d 30 -t 4 -c 100 -s wrk.lua http://localhost:8000/graphql | tee wrk-output.txt
+
+      - id: check_tailcall
+        name: Check Tailcall Status
+        run: |
+          tailcall_pid=$(cat tailcall_pid.txt)
+          if ! kill -0 $tailcall_pid > /dev/null 2>&1; then
+            echo "Tailcall process has crashed. Log output:" >&2
+            cat tailcall.log >&2
+            exit 1
+          else
+            echo "Tailcall process is still running"
+          fi
 
       - id: convert_wrk_output_markdown
         name: Convert Output to Markdown

benches/impl_path_string_for_evaluation_context.rs

@@ -187,6 +187,10 @@ impl<'a> ResolverContextLike<'a> for MockGraphqlContext {
         None
     }
 
+    fn is_query(&'a self) -> bool {
+        false
+    }
+
     fn add_error(&'a self, _: async_graphql::ServerError) {}
 }
 

generated/.tailcallrc.graphql

@@ -360,6 +360,11 @@ directive @upstream(
   """
   dedupe: Boolean
   """
+  When set to `true`, it will ensure no HTTP, GRPC, or any other IO call is made more 
+  than once if similar request is inflight across the server's lifetime.
+  """
+  dedupeInFlight: Boolean
+  """
   The `http2Only` setting allows you to specify whether the client should always issue 
   HTTP2 requests, without checking if the server supports it or not. By default it 
   is set to `false` for all HTTP requests made by the server, but is automatically 

generated/.tailcallrc.schema.json

@@ -1435,6 +1435,13 @@
             "null"
           ]
         },
+        "dedupeInFlight": {
+          "description": "When set to `true`, it will ensure no HTTP, GRPC, or any other IO call is made more than once if similar request is inflight across the server's lifetime.",
+          "type": [
+            "boolean",
+            "null"
+          ]
+        },
         "http2Only": {
           "description": "The `http2Only` setting allows you to specify whether the client should always issue HTTP2 requests, without checking if the server supports it or not. By default it is set to `false` for all HTTP requests made by the server, but is automatically set to true for GRPC.",
           "type": [

src/core/app_context.rs

@@ -2,16 +2,17 @@ use std::sync::Arc;
 
 use async_graphql::dynamic::{self, DynamicRequest};
 use async_graphql::Response;
+use async_graphql_value::ConstValue;
 
 use crate::core::auth::context::GlobalAuthContext;
 use crate::core::blueprint::Type::ListType;
 use crate::core::blueprint::{Blueprint, Definition, SchemaModifiers};
-use crate::core::data_loader::DataLoader;
+use crate::core::data_loader::{DataLoader, DedupeResult};
 use crate::core::graphql::GraphqlDataLoader;
 use crate::core::grpc;
 use crate::core::grpc::data_loader::GrpcDataLoader;
 use crate::core::http::{DataLoaderRequest, HttpDataLoader};
-use crate::core::ir::{DataLoaderId, IO, IR};
+use crate::core::ir::{DataLoaderId, EvaluationError, IoId, IO, IR};
 use crate::core::rest::{Checked, EndpointSet};
 use crate::core::runtime::TargetRuntime;
 
@@ -24,6 +25,7 @@ pub struct AppContext {
     pub grpc_data_loaders: Arc<Vec<DataLoader<grpc::DataLoaderRequest, GrpcDataLoader>>>,
     pub endpoints: EndpointSet<Checked>,
     pub auth_ctx: Arc<GlobalAuthContext>,
+    pub dedupe_handler: Arc<DedupeResult<IoId, ConstValue, EvaluationError>>,
 }
 
 impl AppContext {
@@ -128,6 +130,7 @@ impl AppContext {
             grpc_data_loaders: Arc::new(grpc_data_loaders),
             endpoints,
             auth_ctx: Arc::new(auth_ctx),
+            dedupe_handler: Arc::new(DedupeResult::new(false)),
         }
     }
 

src/core/blueprint/upstream.rs

@@ -28,6 +28,7 @@ pub struct Upstream {
     pub batch: Option<Batch>,
     pub http2_only: bool,
     pub dedupe: bool,
+    pub dedupe_in_flight: bool,
     pub on_request: Option<String>,
 }
 
@@ -83,6 +84,7 @@ impl TryFrom<&ConfigModule> for Upstream {
                 batch,
                 http2_only: (config_upstream).get_http_2_only(),
                 dedupe: (config_upstream).get_dedupe(),
+                dedupe_in_flight: (config_upstream).get_dedupe_in_flight(),
                 on_request: (config_upstream).get_on_request(),
             })
             .to_result()

src/core/config/upstream.rs

@@ -145,6 +145,12 @@ pub struct Upstream {
     /// When set to `true`, it will ensure no HTTP, GRPC, or any other IO call
     /// is made more than once within the context of a single GraphQL request.
     pub dedupe: Option<bool>,
+
+    #[serde(default, skip_serializing_if = "is_default")]
+    /// When set to `true`, it will ensure no HTTP, GRPC, or any other IO call
+    /// is made more than once if similar request is inflight across the
+    /// server's lifetime.
+    pub dedupe_in_flight: Option<bool>,
 }
 
 impl Upstream {
@@ -201,6 +207,10 @@ impl Upstream {
         self.dedupe.unwrap_or(false)
     }
 
+    pub fn get_dedupe_in_flight(&self) -> bool {
+        self.dedupe_in_flight.unwrap_or(false)
+    }
+
     pub fn get_on_request(&self) -> Option<String> {
         self.on_request.clone()
     }