10 times x16

Author: carderne

.github/workflows/unit-tests-webapp.yml

@@ -14,17 +14,18 @@ on:
 jobs:
   unitTests:
     name: "🧪 Unit Tests: Webapp"
-    # 3 shards on 32x machines instead of 10 on 8x: webapp tests run files in parallel
-    # (forks pool scales with cores) and are genuinely CPU-hungry, so keep aggregate
-    # compute similar while paying the setup cost (install, prisma generate, image pulls)
-    # 3 times instead of 10.
-    runs-on: warp-ubuntu-latest-x64-32x
+    # 10 shards on 16x machines: webapp test throughput is limited per-machine (one
+    # docker daemon + disk absorbing all the per-file Postgres/ClickHouse container
+    # spin-up), so many machines beats few big ones - fewer/bigger (3x32) measured
+    # SLOWER than 10x8. The 16x (vs 8x) gives the fork pool the CPU headroom the 8x
+    # runners lacked. Setup overhead per machine is ~1 min on warm runners.
+    runs-on: warp-ubuntu-latest-x64-16x
     strategy:
       # one flaky shard shouldn't cancel its siblings - lets us re-run only the failed shard
       fail-fast: false
       matrix:
-        shardIndex: [1, 2, 3]
-        shardTotal: [3]
+        shardIndex: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
+        shardTotal: [10]
     env:
       DOCKERHUB_USERNAME: ${{ secrets.DOCKERHUB_USERNAME }}
       SHARD_INDEX: ${{ matrix.shardIndex }}