feat(benchmarks): update benchmark workflow with JSON output and caching for results comparison

.github/workflows/bench.yml

@@ -1,10 +1,15 @@
 # Benchmarks and integrity checks (push to main branches and workflow_dispatch).
+# Results are stored in a runner cache and compared on every run; the job fails
+# when any metric regresses beyond 150% of the stored baseline, and a commit
+# comment is posted with the offending numbers.
 #
 # Pinned first-party actions (bump tag and SHA together when upgrading):
 #   actions/checkout@v6.0.1            8e8c483db84b4bee98b60c0593521ed34d9990e8
 #   actions/setup-python@v6.2.0        a309ff8b426b58ec0e2a45f0f869d46889d02405
 #   actions/setup-node@v6.1.0          395ad3262231945c25e8478fd5baf05154b1d79f
 #   actions/cache@v4.2.0               1bd1e32a3bdc45362d1e726936510720a7c30a57
+#   benchmark-action/github-action-benchmark@v1.22.0
+#                                      a60cea5bc7b49e15c1f58f411161f99e0df48372
 
 name: Benchmarks
 
@@ -20,7 +25,7 @@ concurrency:
     cancel-in-progress: true
 
 permissions:
-    contents: read
+    contents: write
 
 env:
     FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true
@@ -71,12 +76,34 @@ jobs:
             - name: Setup Task
               run: sh scripts/ci/setup-task.sh
 
+            - name: Restore benchmark baseline cache
+              uses: actions/cache@1bd1e32a3bdc45362d1e726936510720a7c30a57
+              with:
+                  path: ./cache
+                  key: ${{ runner.os }}-bench-baseline-${{ github.ref_name }}
+                  restore-keys: |
+                      ${{ runner.os }}-bench-baseline-
+
             - name: Run benchmarks
               run: |
                   set -euo pipefail
-                  task bench 2>&1 | tee bench_results.txt
+                  poetry run python tests/backend/run_comprehensive_benchmarks.py \
+                      --json-output bench_results.json 2>&1 | tee bench_results.txt
 
             - name: Run integrity tests
               run: |
                   set -euo pipefail
                   task test:integrity 2>&1 | tee -a bench_results.txt
+
+            - name: Store and compare benchmark results
+              uses: benchmark-action/github-action-benchmark@a60cea5bc7b49e15c1f58f411161f99e0df48372
+              with:
+                  name: MeshChatX Backend Benchmarks
+                  tool: customSmallerIsBetter
+                  output-file-path: bench_results.json
+                  external-data-json-path: ./cache/benchmark-data.json
+                  github-token: ${{ secrets.GITHUB_TOKEN }}
+                  alert-threshold: "150%"
+                  fail-on-alert: true
+                  comment-on-alert: true
+                  summary-always: true