Try running a basic comparison benchmark between base and pr commit

.github/workflows/benchmarks/pr_benchmarks.yml

@@ -0,0 +1,35 @@
+name: Run and Cache Benchmarks
+
+on:
+  pull_request:
+    types: [labeled, opened, reopened, synchronize]
+
+jobs:
+  benchmark:
+    name: Run Benchmarks
+    runs-on: ubuntu-latest
+    steps:
+      - name: Dump GitHub context
+        env:
+          GITHUB_CONTEXT: ${{ toJSON(github) }}
+        run: echo "$GITHUB_CONTEXT"
+
+      - name: Setup data and benchmark PR changes
+        uses: actions/checkout@v4
+        run: |
+          cd benchmarks
+          ./bench.sh data tpch10
+          ./bench.sh data clickbench_1
+          ./bench.sh run tpch10
+          ./bench.sh run clickbench_1
+
+      - name: Benchmark baseline and compare
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.base.sha }}
+        run: |
+          cd benchmarks
+          ./bench.sh run tpch10
+          ./bench.sh run clickbench_1
+          pip3 install rich
+          ./bench.sh compare ${{ github.event.pull_request.base.ref }}" ${{ github.ref }}"

benchmarks/bench.sh

@@ -166,7 +166,7 @@ main() {
             esac
             ;;
         run)
-            # Parse positional paraleters
+            # Parse positional parameters
             BENCHMARK=${ARG2:-"${BENCHMARK}"}
             BRANCH_NAME=$(cd ${DATAFUSION_DIR} && git rev-parse --abbrev-ref HEAD)
             BRANCH_NAME=${BRANCH_NAME//\//_} # mind blowing syntax to replace / with _
@@ -177,7 +177,7 @@ main() {
             echo "COMMAND: ${COMMAND}"
             echo "BENCHMARK: ${BENCHMARK}"
             echo "DATAFUSION_DIR: ${DATAFUSION_DIR}"
-            echo "BRACH_NAME: ${BRANCH_NAME}"
+            echo "BRANCH_NAME: ${BRANCH_NAME}"
             echo "DATA_DIR: ${DATA_DIR}"
             echo "RESULTS_DIR: ${RESULTS_DIR}"
             echo "CARGO_COMMAND: ${CARGO_COMMAND}"

datafusion/physical-plan/src/sorts/sort.rs

@@ -956,6 +956,9 @@ impl ExecutionPlan for SortExec {
                         let batch = batch?;
                         topk.insert_batch(batch)?;
                     }
+                    // Test whether benchmarks catch this
+                    // TODO: remove before merge!
+                    tokio::time::sleep(std::time::Duration::from_secs(3)).await;
                     topk.emit()
                 })
                 .try_flatten(),