bulk: perform meta lookup on range cache miss during index backfill

Fixes #84290.

This commit addresses the sustained slowdown described in #84290 by replacing
the call in `SSTBatcher.flushIfNeeded` to `RangeCache.GetCached` with a call to
`RangeCache.Lookup`. The former method checks the cache but returns no range
descriptor if the cache currently has no overlapping key. This is possible if
the descriptor was recently evicted because it was stale. The later method
performs a meta lookup if the cache currently has no overlapping key, so it
should always return _some_ range descriptor.

There's a question of whether we should be logging a warning but proceeding if
this meta lookup fails. For now, I've decided not to change that behavior.

Release justification: None. Don't merge yet.

pkg/kv/bulk/sst_batcher.go

@@ -361,15 +361,14 @@ func (b *SSTBatcher) flushIfNeeded(ctx context.Context, nextKey roachpb.Key) err
 	if !b.flushKeyChecked && b.rc != nil {
 		b.flushKeyChecked = true
 		if k, err := keys.Addr(nextKey); err != nil {
-			log.Warningf(ctx, "failed to get RKey for flush key lookup")
+			log.Warningf(ctx, "failed to get RKey for flush key lookup: %v", err)
 		} else {
-			r := b.rc.GetCached(ctx, k, false /* inverted */)
-			if r != nil {
+			if r, err := b.rc.Lookup(ctx, k); err != nil {
+				log.Warningf(ctx, "failed to lookup range cache entry for key %v: %v", k, err)
+			} else {
 				k := r.Desc().EndKey.AsRawKey()
 				b.flushKey = k
 				log.VEventf(ctx, 3, "%s building sstable that will flush before %v", b.name, k)
-			} else {
-				log.VEventf(ctx, 2, "%s no cached range desc available to determine sst flush key", b.name)
 			}
 		}
 	}

pkg/kv/bulk/sst_batcher_test.go

@@ -277,27 +277,24 @@ func runTestImport(t *testing.T, batchSizeValue int64) {
 				return encoding.EncodeStringAscending(append([]byte{}, prefix...), fmt.Sprintf("k%d", i))
 			}
 
-			t.Logf("splitting at %s and %s", key(split1), key(split2))
+			t.Logf("splitting at %s", key(split1))
 			require.NoError(t, kvDB.AdminSplit(ctx, key(split1), hlc.MaxTimestamp /* expirationTime */))
-			require.NoError(t, kvDB.AdminSplit(ctx, key(split2), hlc.MaxTimestamp /* expirationTime */))
 
 			// We want to make sure our range-aware batching knows about one of our
-			// splits to exercise that codepath, but we also want to make sure we
+			// splits to exercise that code path, but we also want to make sure we
 			// still handle an unexpected split, so we make our own range cache and
-			// only populate it with one of our two splits.
-			mockCache := rangecache.NewRangeCache(s.ClusterSettings(), nil,
+			// populate it after the first split but before the second split.
+			ds := s.DistSenderI().(*kvcoord.DistSender)
+			mockCache := rangecache.NewRangeCache(s.ClusterSettings(), ds,
 				func() int64 { return 2 << 10 }, s.Stopper(), s.TracerI().(*tracing.Tracer))
-			addr, err := keys.Addr(key(0))
-			require.NoError(t, err)
-
-			tok, err := s.DistSenderI().(*kvcoord.DistSender).RangeDescriptorCache().LookupWithEvictionToken(
-				ctx, addr, rangecache.EvictionToken{}, false)
-			require.NoError(t, err)
-
-			r := roachpb.RangeInfo{
-				Desc: *tok.Desc(),
+			for _, k := range []int{0, split1} {
+				ent, err := ds.RangeDescriptorCache().Lookup(ctx, keys.MustAddr(key(k)))
+				require.NoError(t, err)
+				mockCache.Insert(ctx, roachpb.RangeInfo{Desc: *ent.Desc()})
 			}
-			mockCache.Insert(ctx, r)
+
+			t.Logf("splitting at %s", key(split2))
+			require.NoError(t, kvDB.AdminSplit(ctx, key(split2), hlc.MaxTimestamp /* expirationTime */))
 
 			ts := hlc.Timestamp{WallTime: 100}
 			mem := mon.NewUnlimitedMonitor(ctx, "lots", mon.MemoryResource, nil, nil, 0, nil)