Merge #66726 #66960

66726: sql: enhance join reader benchmarks r=rytaft a=cucaroach

Informs #51576


Introduce a "short" version of the benchmark

Add lookupExpr support

Allow lookupExpr to run w/o changing bench name to facilitate lookupExpr vs lookupColumn comparison paving way for lookupExpr only world.

Add a lookup expr stress test that sees how the costs of the lookupExpr varies as the expr gets more complicated and everything else stays constant.

The lookup expr stress case shows that we should definitely factor the number of columns involved in the lookupexpr into the cost model:

https://docs.google.com/spreadsheets/d/1mhhit52YZUOQmes6ttRryiZQtmMQzn_LVOg2kXCBnCU/edit?usp=sharing



66960: roachtest: maybe unbreak tpcc/weekly/headroom r=stevendanna a=tbg

The test had a four day ten minute timeout, but was attempting to run the
workload for four days. This left only ten minutes to import the dataset
and run the rebalancing ramp-up, which was orders of magnitude too
short.

Closes #66955.

Release note: None


Co-authored-by: Tommy Reilly <[email protected]>
Co-authored-by: Tobias Grieger <[email protected]>

pkg/cmd/roachtest/tpcc.go

@@ -425,7 +425,9 @@ func registerTPCC(r *testRegistry) {
 		MinVersion: "v19.1.0",
 		Tags:       []string{`weekly`},
 		Cluster:    r.makeClusterSpec(4, spec.CPU(16)),
-		Timeout:    time.Duration(4*24)*time.Hour + time.Duration(10)*time.Minute,
+		// Give the test a generous extra 10 hours to load the dataset and
+		// slowly ramp up the load.
+		Timeout: 4*24*time.Hour + 10*time.Hour,
 		Run: func(ctx context.Context, t test.Test, c cluster.Cluster) {
 			warehouses := 1000
 			runTPCC(ctx, t, c, tpccOptions{