microbench-ci: benchmark suite tuning

Previously, one node local KV microbenchmarks were run as part of CI. These
tests are sensitive to compiler induced variance. Even though the regressions /
improvements are reproducible, often the code change is unrelated to the
performance change. It happens often enough with these benchmarks that it could
be wasteful for engineers to investigate these slight differences.

This change updates the benchmark suite to run the 3 node KV benchmarks instead
which is less likely to be triggered due to compiler induced variance.

Author: herkolategan

pkg/cmd/microbench-ci/config/pull-request-suite.yml

@@ -13,25 +13,25 @@ benchmarks:
       - "allocs/op"
 
   - display_name: Sysbench
-    labels: ["KV", "1node", "local", "oltp_read_only"]
-    name: "BenchmarkSysbench/KV/1node_local/oltp_read_only"
+    labels: ["KV", "3node", "oltp_read_only"]
+    name: "BenchmarkSysbench/KV/3node/oltp_read_only"
     package: "pkg/sql/tests"
     runner_group: 2
-    count: 20
-    iterations: 6000
+    count: 15
+    iterations: 3000
     compare_alpha: 0.025
     retries: 3
     metrics:
       - "sec/op"
       - "allocs/op"
 
   - display_name: Sysbench
-    labels: ["KV", "1node", "local", "oltp_write_only"]
-    name: "BenchmarkSysbench/KV/1node_local/oltp_write_only"
+    labels: ["KV", "3node", "oltp_write_only"]
+    name: "BenchmarkSysbench/KV/3node/oltp_write_only"
     package: "pkg/sql/tests"
     runner_group: 2
-    count: 20
-    iterations: 6000
+    count: 15
+    iterations: 3000
     compare_alpha: 0.025
     retries: 3
     metrics: