In an InnoDB record cursor, there is a counter row_prebuilt_t::n_rows_fetched that counts records that have been fetched in a key range scan. The main purpose of this counter is to control whether to trigger a prefetch of a cache that is slated for removal in MDEV-16232.

On 64-bit systems, this counter cannot overflow. If we overestimate that InnoDB could read 10¹² or 2⁴⁰ records per second in a range scan, it would take more than half a year of such a range scan before the counter would wrap around. An even more rigid limitation would be that InnoDB only supports 2³² pages per tablespace, and you could fit at most a few thousand or 2¹² records per page, that is, there can't be more than some 2⁴⁴ records per index.

I would propose to revise an overflow check as follows, disabling it on 64-bit systems:

diff --git a/storage/innobase/row/row0sel.cc b/storage/innobase/row/row0sel.cc

index 4504fabd4e0..b217f6d20fe 100644

--- a/storage/innobase/row/row0sel.cc

+++ b/storage/innobase/row/row0sel.cc

@@ -4406,13 +4406,11 @@ row_search_mvcc(

 			goto func_exit;

 		}

+#if SIZEOF_SIZE_T < 8

+		if (UNIV_LIKELY(~prebuilt->n_rows_fetched))

+#endif

 		prebuilt->n_rows_fetched++;

-		if (prebuilt->n_rows_fetched > 1000000000) {

-			/* Prevent wrap-over */

-			prebuilt->n_rows_fetched = 500000000;

-		}

-

 		mode = pcur->search_mode;

 	}