InnoDB replace statement returns "can't find record" error during bulk insert operation.

origin/10.7 f3785f099c2f0f251f39632928e822328abe9a02 2022-09-13T08:48:40+03:00

Scenario in the RQG test:

Two concurrent sessions run a DDL/DML mix.

One or both sessions go set  SET foreign_key_checks = 0, unique_checks = 0 ;

After a few seconds with DDL/DML load one session catches

REPLACE INTO t6 (col1,col2, col_int, col_string, col_text) VALUES /* 839 */ (839,839,839,REPEAT(SUBSTR(CAST( 839 AS CHAR),1,1), 10),REPEAT(SUBSTR(CAST( 839 AS CHAR),1,1), @fill_amount) ), (839,839,839,REPEAT(SUBSTR(CAST( 839 AS CHAR),1,1), 10),REPEAT(SUBSTR(CAST( 839 AS CHAR),1,1), @fill_amount) )  /* E_R Thread1 QNO 27 CON_ID 17 */  failed: 1032 , errstr: Can't find record in 't6'

RQG reacts with killing the DB server with SIGABRT.

pluto:/data/results/1663181785/TBR-403$ _RR_TRACE_DIR=./1/rr/ rr replay

Maybe better rr trace

REPLACE INTO t3 (col1) VALUES /* 1894 */ (1894), (1894)  /* E_R Thread1 QNO 149 CON_ID 17 */  failed: 1032 , errstr: Can't find record in 't3'

pluto:/data/results/1663591301/MDEV-29545$ _RR_TRACE_DIR=./1/rr/ rr replay --mark-stdio

The problem is on 

origin/HEAD, origin/10.6 8ff10969996a5349e906fab44b45c6e3ec53eea0 2022-08-24T17:06:57+03:00

too.

No replay on 10.5.18

origin/10.5 23a8654cdbd84336fd2004bdd2b0cb4a0ecd3d4a 2022-09-15T12:20:50+04:00