PostHog · bill-ph · Sep 8, 2026 · Sep 8, 2026 · Sep 8, 2026
diff --git a/.github/workflows/scenario-dev.yml b/.github/workflows/scenario-dev.yml
@@ -1,4 +1,5 @@
 name: scenario-dev
+run-name: scenario-dev / ${{ inputs.scenario || 'posthog_frozen_perf' }} / Trino ${{ inputs.trino_perf_shape || 'baseline' }}
 
 on:
   workflow_dispatch:
@@ -13,6 +14,17 @@ on:
         required: false
         default: ""
         type: string
+      trino_perf_shape:
+        description: Trino execution shape (experiments require posthog_frozen_perf)
+        required: false
+        default: baseline
+        type: choice
+        options:
+          - baseline
+          - large
+          - scaleout
+          - large-scaleout
+          - all
   schedule:
     - cron: "17 8 * * *"
 
@@ -25,7 +37,24 @@ concurrency:
   cancel-in-progress: false
 
 jobs:
+  plan:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 5
+    outputs:
+      matrix: ${{ steps.shapes.outputs.matrix }}
+    env:
+      SCENARIO_NAME: ${{ github.event_name == 'schedule' && 'posthog_frozen_perf' || inputs.scenario }}
+      TRINO_PERF_SHAPE: ${{ inputs.trino_perf_shape || 'baseline' }}
+    steps:
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+      - name: Select Trino shapes
+        id: shapes
+        run: |
+          matrix="$(bash scripts/scenario_trino_shapes.sh)"
+          echo "matrix=$matrix" >> "$GITHUB_OUTPUT"
+
   scenario-runner-image:
+    needs: [plan]
     uses: ./.github/workflows/_image-build.yml
     with:
       dockerfile: tests/mw-dev/scenario/Dockerfile
@@ -37,6 +66,7 @@ jobs:
       ecr-role: ${{ vars.AWS_ECR_PRS_PUBLISH_IAM_ROLE }}
 
   duckgres-image:
+    needs: [plan]
     if: ${{ github.event_name != 'workflow_dispatch' || inputs.duckgres_image == '' }}
     uses: ./.github/workflows/_image-build.yml
     with:
@@ -57,12 +87,20 @@ jobs:
       ecr-role: ${{ vars.AWS_ECR_PRS_PUBLISH_IAM_ROLE }}
 
   scenario:
-    needs: [scenario-runner-image, duckgres-image]
-    if: ${{ always() && needs.scenario-runner-image.result == 'success' && (needs.duckgres-image.result == 'success' || (github.event_name == 'workflow_dispatch' && inputs.duckgres_image != '')) }}
+    name: scenario (${{ matrix.shape }})
+    needs: [plan, scenario-runner-image, duckgres-image]
+    if: ${{ always() && needs.plan.result == 'success' && needs.scenario-runner-image.result == 'success' && (needs.duckgres-image.result == 'success' || (github.event_name == 'workflow_dispatch' && inputs.duckgres_image != '')) }}
+    strategy:
+      # Each job includes teardown. A fresh runner also refreshes credentials
+      # and the timeout budget before admitting the next shape.
+      max-parallel: 1
+      fail-fast: false
+      matrix: ${{ fromJSON(needs.plan.outputs.matrix) }}
     runs-on: ubuntu-24.04
     timeout-minutes: 270
     env:
       SCENARIO_NAME: ${{ github.event_name == 'schedule' && 'posthog_frozen_perf' || inputs.scenario }}
+      TRINO_PERF_SHAPE: ${{ matrix.shape }}
       SCENARIO_RUNNER_IMAGE: ${{ needs.scenario-runner-image.outputs.image }}
       WORKER_IMAGE: ${{ (github.event_name == 'workflow_dispatch' && inputs.duckgres_image) || needs.duckgres-image.outputs.image }}
       CONTROLPLANE_IMAGE: ${{ (github.event_name == 'workflow_dispatch' && inputs.duckgres_image) || needs.duckgres-image.outputs.image }}
@@ -76,8 +114,8 @@ jobs:
       TRINO_POD_IDENTITY_ROLE: ${{ secrets.MW_DEV_TRINO_POD_IDENTITY_ROLE }}
       TRINO_IMAGE: ghcr.io/posthog/trino:4505364c570d6b51edecd299b603fca4b6693d86@sha256:ac80c275fd18a439d25da5652ab5cd3c80bcbdd2d88d64c9722dc3e8bb68ba07
       E2E_SUITE: ${{ (github.event_name == 'schedule' || inputs.scenario == 'posthog_frozen_perf') && 'trino' || 'neutral' }}
-      PR_NUMBER: ${{ github.run_id }}
-      NAMESPACE: duckgres-ci-pr-${{ github.run_id }}
+      PR_NUMBER: ${{ github.run_id }}${{ matrix.suffix }}
+      NAMESPACE: duckgres-ci-pr-${{ github.run_id }}${{ matrix.suffix }}
       DUCKGRES_SCENARIO_MAX_RUNTIME: 4h
       DUCKGRES_SCENARIO_GO_TEST_TIMEOUT: 4h15m
       # Add process headroom for repeated full-dataset pgwire aggregates.
@@ -95,7 +133,7 @@ jobs:
           go-version-file: go.mod
 
       - name: Test scenario workflow scripts
-        run: go test -count=1 ./tests/mw-dev/scenario ./tests/mw-dev ./tests/perf/publishercli
+        run: go test -count=1 ./tests/mw-dev/scenario ./tests/mw-dev ./tests/perf/publishercli ./tests/perf/shapecompare ./cmd/duckgres-perf-shape-summary
 
       - name: Configure AWS credentials (OIDC)
         uses: aws-actions/configure-aws-credentials@517a711dbcd0e402f90c77e7e2f81e849156e31d # v6.2.2
@@ -120,19 +158,32 @@ jobs:
         run: aws eks update-kubeconfig --name "$CLUSTER_NAME" --region "$AWS_REGION" --alias "$KUBE_CONTEXT"
 
       - name: Load Athena perf configuration
-        if: env.SCENARIO_NAME == 'posthog_frozen_perf'
+        if: env.SCENARIO_NAME == 'posthog_frozen_perf' && matrix.shape == 'baseline'
         run: bash scripts/scenario_athena_config.sh >> "$GITHUB_ENV"
 
       - name: Deploy isolated Duckgres stack
+        id: deploy
         run: tests/mw-dev/run.sh deploy
 
       - name: Run selected scenario
+        id: benchmark
         run: tests/mw-dev/run.sh test-scenario
 
       - name: Publish scenario summary
         if: always()
         run: |
           set -euo pipefail
+          shape_file=artifacts/scenario-dev/trino-perf-shape.json
+          if [ -f "$shape_file" ]; then
+            {
+              echo '## Trino execution shape'
+              echo
+              echo '```json'
+              cat "$shape_file"
+              echo '```'
+              echo
+            } >> "$GITHUB_STEP_SUMMARY"
+          fi
           summary_file="$(find artifacts/scenario-dev -type f -name scenario_summary.md -print -quit 2>/dev/null || true)"
           if [ -n "$summary_file" ]; then
             cat "$summary_file" >> "$GITHUB_STEP_SUMMARY"
@@ -149,20 +200,36 @@ jobs:
         run: tests/mw-dev/run.sh diagnostics
 
       - name: Teardown
+        id: teardown
         if: always()
         run: tests/mw-dev/run.sh teardown
 
+      - name: Record shape outcome
+        if: always()
+        env:
+          DEPLOY_OUTCOME: ${{ steps.deploy.outcome }}
+          SCENARIO_OUTCOME: ${{ steps.benchmark.outcome }}
+          TEARDOWN_OUTCOME: ${{ steps.teardown.outcome }}
+        run: |
+          mkdir -p artifacts/scenario-dev
+          jq -n --arg shape "$TRINO_PERF_SHAPE" \
+            --arg deploy "$DEPLOY_OUTCOME" --arg scenario "$SCENARIO_OUTCOME" \
+            --arg teardown "$TEARDOWN_OUTCOME" \
+            '$ARGS.named' > artifacts/scenario-dev/shape-result.json
+
       - name: Upload scenario artifacts
         if: always()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
-          name: scenario-dev-${{ github.run_id }}-${{ github.run_attempt }}
+          name: scenario-dev-${{ github.run_id }}-${{ github.run_attempt }}-${{ matrix.shape }}
           path: artifacts/scenario-dev/
           if-no-files-found: warn
           retention-days: 14
 
       - name: Publish scenario perf results
-        if: ${{ always() && github.ref == 'refs/heads/main' }}
+        # Shape experiments remain downloadable artifacts; publishing them
+        # would mix different resource budgets into the daily baseline.
+        if: ${{ always() && github.ref == 'refs/heads/main' && env.TRINO_PERF_SHAPE == 'baseline' && inputs.trino_perf_shape != 'all' }}
         timeout-minutes: 10
         env:
           MW_DEV_SCENARIO_PERF_SECRET_ID: ${{ vars.MW_DEV_SCENARIO_PERF_SECRET_ID }}
@@ -195,3 +262,40 @@ jobs:
             fi
           done
           exit "$publish_failed"
+
+  compare-shapes:
+    needs: [plan, scenario]
+    if: ${{ always() && needs.plan.result == 'success' && inputs.trino_perf_shape == 'all' }}
+    runs-on: ubuntu-24.04
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+      - uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16 # v6.5.0
+        with:
+          go-version-file: go.mod
+      - name: Download shape artifacts
+        # Missing uploads are reported as missing shapes by the comparison.
+        continue-on-error: true
+        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
+        with:
+          pattern: scenario-dev-${{ github.run_id }}-${{ github.run_attempt }}-*
+          path: artifacts/trino-shapes
+          merge-multiple: false
+      - name: Compare Trino shapes
+        run: go run ./cmd/duckgres-perf-shape-summary --artifacts-dir artifacts/trino-shapes > trino-shape-comparison.md
+      - name: Publish Trino shape comparison
+        if: always()
+        run: |
+          if [ -s trino-shape-comparison.md ]; then
+            cat trino-shape-comparison.md >> "$GITHUB_STEP_SUMMARY"
+          else
+            echo 'Trino shape comparison unavailable. Inspect the shape jobs and their artifacts.' >> "$GITHUB_STEP_SUMMARY"
+          fi
+      - name: Upload Trino shape comparison
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: trino-shape-comparison-${{ github.run_id }}-${{ github.run_attempt }}
+          path: trino-shape-comparison.md
+          if-no-files-found: warn
+          retention-days: 14
diff --git a/cmd/duckgres-perf-shape-summary/main.go b/cmd/duckgres-perf-shape-summary/main.go
@@ -0,0 +1,33 @@
+package main
+
+import (
+	"flag"
+	"fmt"
+	"io"
+	"os"
+
+	"github.com/posthog/duckgres/tests/perf/shapecompare"
+)
+
+func main() {
+	if err := run(os.Args[1:], os.Stdout); err != nil {
+		_, _ = fmt.Fprintln(os.Stderr, err)
+		os.Exit(1)
+	}
+}
+
+func run(args []string, stdout io.Writer) error {
+	flags := flag.NewFlagSet("duckgres-perf-shape-summary", flag.ContinueOnError)
+	dir := flags.String("artifacts-dir", "", "required directory containing one artifact directory per Trino shape")
+	if err := flags.Parse(args); err != nil {
+		return err
+	}
+	if *dir == "" || flags.NArg() != 0 {
+		return fmt.Errorf("usage: duckgres-perf-shape-summary --artifacts-dir <directory>")
+	}
+	report, err := shapecompare.Generate(*dir)
+	if _, writeErr := io.WriteString(stdout, report); writeErr != nil {
+		return writeErr
+	}
+	return err
+}
diff --git a/cmd/duckgres-perf-shape-summary/main_test.go b/cmd/duckgres-perf-shape-summary/main_test.go
@@ -0,0 +1,33 @@
+package main
+
+import (
+	"bytes"
+	"errors"
+	"strings"
+	"testing"
+)
+
+func TestRunRequiresArtifactsDirectory(t *testing.T) {
+	for _, args := range [][]string{nil, {"--artifacts-dir", "test", "extra"}} {
+		if err := run(args, &bytes.Buffer{}); err == nil {
+			t.Fatal("expected usage error")
+		}
+	}
+}
+
+func TestRunPublishesIncompleteReportBeforeFailing(t *testing.T) {
+	var out bytes.Buffer
+	err := run([]string{"--artifacts-dir", t.TempDir()}, &out)
+	if err == nil || !strings.Contains(out.String(), "Comparison incomplete") || !strings.Contains(out.String(), "missing artifact") {
+		t.Fatalf("report=%s, error=%v", out.String(), err)
+	}
+}
+
+type failedWriter struct{}
+
+func (failedWriter) Write([]byte) (int, error) { return 0, errors.New("write failed") }
+func TestRunReturnsWriteFailure(t *testing.T) {
+	if err := run([]string{"--artifacts-dir", t.TempDir()}, failedWriter{}); err == nil || err.Error() != "write failed" {
+		t.Fatalf("got %v", err)
+	}
+}
diff --git a/justfile b/justfile
@@ -301,6 +301,7 @@ test:
 test-unit:
     go test -v -p 1 . ./configresolve/... ./duckdbservice/... ./server/... ./transpiler/... ./internal/... ./tests/manifests/...
     go test -v -count=1 ./tests/mw-dev/...
+    go test -v -count=1 ./tests/perf/shapecompare ./cmd/duckgres-perf-shape-summary
 
 # Run scenario runner unit tests
 [group('test')]

diff --git a/scripts/scenario_run.sh b/scripts/scenario_run.sh
@@ -21,6 +21,7 @@ Optional environment:
   DUCKGRES_SCENARIO_DBT_BIN
   DUCKGRES_SCENARIO_MAX_RUNTIME
   DUCKGRES_SCENARIO_GO_TEST_TIMEOUT
+  DUCKGRES_SCENARIO_PERF_MODE      (default: full; trino-only for posthog-frozen-perf)
 
 Scenario-specific required environment:
   DUCKGRES_SCENARIO_ORG_ID       (required by successful provisioning scenarios)
@@ -83,6 +84,29 @@ root_relative_path() {
 scenario_file="$(root_relative_path "$scenario_file")"
 output_base="$(root_relative_path "$output_base")"
 
+# Keep preflight aligned with the Go scenario loader: Trino-only removes the
+# Athena target and its environment requirements before template resolution.
+perf_mode="${DUCKGRES_SCENARIO_PERF_MODE:-full}"
+case "$perf_mode" in
+  full) ;;
+  trino-only)
+    scenario_name="$(awk '/^name:[[:space:]]*/ {
+      sub(/^name:[[:space:]]*/, "")
+      sub(/[[:space:]]*$/, "")
+      gsub(/^["'\'']|["'\'']$/, "")
+      print; exit
+    }' "$scenario_file")"
+    if [ "$scenario_name" != posthog-frozen-perf ]; then
+      echo "DUCKGRES_SCENARIO_PERF_MODE=trino-only requires the posthog-frozen-perf scenario." >&2
+      exit 2
+    fi
+    ;;
+  *)
+    echo "DUCKGRES_SCENARIO_PERF_MODE must be full or trino-only." >&2
+    exit 2
+    ;;
+esac
+
 scenario_required_env() {
   awk '
     /^[^[:space:]]/ { in_required = 0 }
@@ -104,6 +128,11 @@ required=(
 )
 if [ -f "$scenario_file" ]; then
   while IFS= read -r name; do
+    if [ "$perf_mode" = trino-only ]; then
+      case "$name" in
+        DUCKGRES_SCENARIO_ATHENA_REGION|DUCKGRES_SCENARIO_ATHENA_WORKGROUP|DUCKGRES_SCENARIO_ATHENA_DATABASE|DUCKGRES_SCENARIO_ATHENA_RESULTS_S3_URI) continue ;;
+      esac
+    fi
     required+=("$name")
   done < <(scenario_required_env "$scenario_file")
 elif [ "$check_env_only" -eq 1 ]; then

diff --git a/scripts/scenario_trino_shapes.sh b/scripts/scenario_trino_shapes.sh
@@ -0,0 +1,24 @@
+#!/usr/bin/env bash
+# Print the scenario workflow matrix. Single-shape runs retain the original run
+# ID; all-shape runs append distinct numeric suffixes for harness isolation.
+set -euo pipefail
+
+shape="${TRINO_PERF_SHAPE:-baseline}"
+case "$shape" in
+  baseline|large|scaleout|large-scaleout|all) ;;
+  *)
+    printf '%s\n' 'Invalid TRINO_PERF_SHAPE: expected baseline, large, scaleout, large-scaleout, or all' >&2
+    exit 1
+    ;;
+esac
+
+if [[ "$shape" != baseline && "${SCENARIO_NAME:-}" != posthog_frozen_perf ]]; then
+  printf '%s\n' 'Trino shape experiments require posthog_frozen_perf' >&2
+  exit 1
+fi
+
+if [[ "$shape" == all ]]; then
+  printf '%s\n' '{"include":[{"shape":"baseline","suffix":"1"},{"shape":"large","suffix":"2"},{"shape":"scaleout","suffix":"3"},{"shape":"large-scaleout","suffix":"4"}]}'
+else
+  printf '{"include":[{"shape":"%s","suffix":""}]}\n' "$shape"
+fi
diff --git a/tests/mw-dev/README.md b/tests/mw-dev/README.md
@@ -57,6 +57,20 @@ The isolated control plane's default worker request is configurable through
 `scenario-dev.yml` explicitly overrides them to 3 CPU and 12Gi for the frozen
 perf workload. Direct `run.sh` callers can make the same explicit override.
 
+Manual frozen-perf runs can select `TRINO_PERF_SHAPE=baseline|large|scaleout|large-scaleout`
+(`trino_perf_shape` in workflow dispatch). The default remains `baseline` for
+scheduled runs and the Trino E2E lane. Workflow dispatch also accepts `all` to
+run all four shapes sequentially with shared image builds and one comparison
+summary. Each shape receives a separate temporary stack and cleanup. Baseline
+runs the full cross-engine benchmark; nonbaseline shapes measure only Trino,
+with identical SQL, warmup, and measured iterations, while retaining setup and
+validation. The harness derives `DUCKGRES_SCENARIO_PERF_MODE=full` or
+`trino-only` from the shape and records it in provenance; an inherited value
+cannot override it. Nonbaseline shapes need no Athena configuration or scenario
+Pod Identity role. See the
+[Trino experiment runbook](../perf/README.md#trino-worker-shape-experiments)
+for resource budgets, result provenance, sequential comparison, and recovery.
+
 ### Scenario Trino readiness
 
 Scenarios that opt an org into Trino in their `provision_warehouse` request can