kv/bulk: add debug setting to discard SSTs

When debugging performance of various bulk operations, sometimes it is
useful to focus on the performance of the parts of the operation that produce
data to be ingested without the ingestion of said data complicating the picture
or applying a form of backpressure on the production process that changes its
performance.

This setting is strictly for debugging/development use and as such is locked with
the unsafe setting interlock.

Release note: none.
Epic: none.

Author: dt

pkg/kv/bulk/sst_batcher.go

@@ -628,6 +628,14 @@ func (b *SSTBatcher) syncFlush() error {
 	return flushErr
 }
 
+var debugDropSSTOnFlush = settings.RegisterBoolSetting(
+	settings.ApplicationLevel,
+	"bulkio.ingest.unsafe_debug.drop_sst_on_flush.enabled",
+	"if set, the SSTBatcher will simply discard data instead of flushing it (destroys data; for performance debugging experiments only)",
+	false,
+	settings.WithUnsafe,
+)
+
 // startFlush starts a flush of the current batch. If it encounters any errors
 // the errors are reported by the call to `syncFlush`.
 //
@@ -817,9 +825,13 @@ func (b *SSTBatcher) startFlush(ctx context.Context, reason int) {
 	b.asyncAddSSTs.GoCtx(func(ctx context.Context) error {
 		defer res.Release()
 		defer b.mem.Shrink(ctx, reserved)
-		results, err := b.adder.AddSSTable(ctx, batchTS, start, end, data, mvccStats, performanceStats)
-		if err != nil {
-			return err
+
+		var results []addSSTResult
+		if !debugDropSSTOnFlush.Get(&b.settings.SV) {
+			results, err = b.adder.AddSSTable(ctx, batchTS, start, end, data, mvccStats, performanceStats)
+			if err != nil {
+				return err
+			}
 		}
 
 		// Now that we have completed ingesting the SSTables we take a lock and