workload: use IMPORT by default, but warn against uniqueness violations

In 1041bf0 (#152979), we converted ycsb and kv workloads to use INSERT
statements for initial data loading when hashing keys.

Hashing keys can cause key collisions. At higher row counts, these collisions
are inevitable. Since IMPORTs require imported data to be unique, an IMPORT
with colliding keys would fail. Before #152979, only the ycsb workload was
susceptible to this failure mode, because the kv workload did not use the same
key generator for data loaded through --insert-count (#107874). Fixing #107874
made the kv workload susceptible to this type of failure.

The change to use INSERTs instead of IMPORTs ran afoul of KV message size
limits (see #153086). For now we revert the change to automatically use IMPORTs
and instead warn about the possibility of collisions.

Epic: none
Release note: none

Author: jbowens

pkg/ccl/workloadccl/fixture.go

@@ -15,6 +15,7 @@ import (
 	"math"
 	"net/url"
 	"path/filepath"
+	"slices"
 	"strconv"
 	"strings"
 	"sync/atomic"
@@ -310,6 +311,12 @@ func (l ImportDataLoader) InitialDataLoad(
 	if l.FilesPerNode == 0 {
 		l.FilesPerNode = 1
 	}
+	mayContainDuplicates := slices.ContainsFunc(gen.Tables(), func(tbl workload.Table) bool {
+		return tbl.InitialRows.MayContainDuplicates
+	})
+	if mayContainDuplicates {
+		log.Dev.Warningf(ctx, "importing fixture using a key generator that may contain duplicates; IMPORT may abort with a key uniqueness violation at higher row counts")
+	}
 
 	log.Dev.Infof(ctx, "starting import of %d tables", len(gen.Tables()))
 	start := timeutil.Now()

pkg/workload/workload.go

@@ -46,7 +46,7 @@ type Generator interface {
 func SupportsFixtures(gen Generator) bool {
 	tt := gen.Tables()
 	for _, t := range tt {
-		if t.InitialRows.FillBatch == nil || t.InitialRows.MayContainDuplicates {
+		if t.InitialRows.FillBatch == nil {
 			return false
 		}
 	}