Merge #145256 #145429

145256: roachtest/tlp: randomly force generic query plans r=mgartner a=mgartner

Generic query plans are now forced for 25% of TLP queries. Note that
only one type of TLP query uses placeholders, which are required for
"interesting" generic query plans, simple `SELECT .. WHERE ..` queries.
Joins and grouping do not yet use placeholders.

Fixes #128912

Release note: None


145429: roachtest: deflake lease-preferences roachtest r=wenyihu6 a=kvoli

`server.span_stats.span_batch_limit` no longer needs to be set after

`kv.enqueue_in_replicate_queue_on_span_config_update.enabled` defaults
to true in versions on or after `v24.1`.

Informs: #141359
Informs: #145055

---


The `lease-preferences/*` roachtests wait for lease preferences to be
initially conformed before purposefully forcing leases out of
conformance.

Its possible to the initial lease preference conformance to take up to
10 minutes (a replica scanner cycle). Allow this much time. Note the
non-initial conformance should be quicker, as leases are checked on
acquisition, to see whether they violate the lease preferences.

Resolves: #141359
Resolves: #145055
Release note: None

Co-authored-by: Marcus Gartner <[email protected]>
Co-authored-by: Austen McClernon <[email protected]>

Author: 3 people

pkg/cmd/roachtest/tests/lease_preferences.go

@@ -172,19 +172,16 @@ func runLeasePreferences(
 	// waitForLessPreferred=false). This duration is used to ensure the lease
 	// preference satisfaction is reasonably permanent.
 	const stableDuration = 30 * time.Second
+	// initialWaitDuration is how long the test will wait for initial lease
+	// preference conformance. This duration is approx. a replica scanner cycle.
+	const initialWaitDuration = 10 * time.Minute
 
 	numNodes := c.Spec().NodeCount
 	allNodes := make([]int, 0, numNodes)
 	for i := 1; i <= numNodes; i++ {
 		allNodes = append(allNodes, i)
 	}
 
-	// TODO(kvoli): temporary workaround for
-	// https://github.com/cockroachdb/cockroach/issues/105274
-	settings := install.MakeClusterSettings()
-	settings.ClusterSettings["server.span_stats.span_batch_limit"] = "4096"
-	settings.ClusterSettings["kv.enqueue_in_replicate_queue_on_span_config_update.enabled"] = "true"
-
 	startNodes := func(nodes ...int) {
 		for _, node := range nodes {
 			// Don't start a backup schedule because this test is timing sensitive.
@@ -198,7 +195,7 @@ func runLeasePreferences(
 				// dc=N: n2N-1 n2N
 				fmt.Sprintf("--locality=region=fake-region,zone=fake-zone,dc=%d", (node-1)/2+1),
 				"--vmodule=replica_proposal=2,lease_queue=3,lease=3")
-			c.Start(ctx, t.L(), opts, settings, c.Node(node))
+			c.Start(ctx, t.L(), opts, install.MakeClusterSettings(), c.Node(node))
 
 		}
 	}
@@ -215,9 +212,13 @@ func runLeasePreferences(
 	conn := c.Conn(ctx, t.L(), numNodes)
 	defer conn.Close()
 
-	checkLeasePreferenceConformance := func(ctx context.Context) {
+	checkLeasePreferenceConformance := func(ctx context.Context, initial bool) {
+		duration := spec.postEventWaitDuration
+		if initial {
+			duration = initialWaitDuration
+		}
 		result, err := waitForLeasePreferences(
-			ctx, t, c, spec.checkNodes, spec.waitForLessPreferred, stableDuration, spec.postEventWaitDuration)
+			ctx, t, c, spec.checkNodes, spec.waitForLessPreferred, stableDuration, duration)
 		require.NoError(t, err, result)
 		require.Truef(t, !result.violating(), "violating lease preferences %s", result)
 		if spec.waitForLessPreferred {
@@ -260,7 +261,7 @@ func runLeasePreferences(
 		leasePreference: spec.preferences,
 	})
 	t.L().Printf("waiting for initial lease preference conformance")
-	checkLeasePreferenceConformance(ctx)
+	checkLeasePreferenceConformance(ctx, true /* initial */)
 
 	// Run the spec event function. The event function will move leases to
 	// non-conforming localities.
@@ -273,7 +274,7 @@ func runLeasePreferences(
 	// Wait for the preference conformance with some leases in non-conforming
 	// localities.
 	t.L().Printf("waiting for post-event lease preference conformance")
-	checkLeasePreferenceConformance(ctx)
+	checkLeasePreferenceConformance(ctx, false /* initial */)
 }
 
 type leasePreferencesResult struct {

pkg/cmd/roachtest/tests/tlp.go

@@ -9,6 +9,7 @@ import (
 	"context"
 	gosql "database/sql"
 	"fmt"
+	"math/rand"
 	"os"
 	"path/filepath"
 	"strings"
@@ -172,7 +173,7 @@ func runOneTLP(
 			continue
 		}
 
-		if err := runTLPQuery(t, conn, tlpSmither, logStmt); err != nil {
+		if err := runTLPQuery(t, conn, rnd, tlpSmither, logStmt); err != nil {
 			t.Fatal(err)
 		}
 	}
@@ -209,15 +210,29 @@ func runMutationStatement(
 // partitioned query. See GenerateTLP for more information on TLP and the
 // generated queries.
 func runTLPQuery(
-	t task.Tasker, conn *gosql.DB, smither *sqlsmith.Smither, logStmt func(string),
-) error {
+	t task.Tasker, conn *gosql.DB, rnd *rand.Rand, smither *sqlsmith.Smither, logStmt func(string),
+) (err error) {
 	// Ignore panics from GenerateTLP.
 	defer func() {
 		if r := recover(); r != nil {
 			return
 		}
 	}()
 
+	// Force generic query plans for 25% of queries.
+	generic := false
+	const forceGeneric = "SET plan_cache_mode = force_generic_plan"
+	if rnd.Intn(4) == 0 {
+		generic = true
+		_, err = conn.Exec(forceGeneric)
+		if err != nil {
+			return err
+		}
+		defer func() {
+			_, err = conn.Exec("RESET plan_cache_mode")
+		}()
+	}
+
 	unpartitioned, partitioned, args := smither.GenerateTLP()
 	combined := sqlsmith.CombinedTLP(unpartitioned, partitioned)
 
@@ -263,6 +278,9 @@ func runTLPQuery(
 		}
 
 		diff := unsortedMatricesDiff(unpartitionedRows, partitionedRows)
+		if generic {
+			logStmt(forceGeneric)
+		}
 		logStmt(unpartitioned)
 		logStmt(partitioned)
 		return errors.Newf(