Marko Mäkelä added a comment - 2018-02-26 11:41 The test is nondeterministic. If the page containing the locked record is not in the buffer pool, NULL will be reported instead of the key of the record. In MariaDB 10.3.5 this failure might occur more easily, because a performance fix of MDEV-14756 removed the code that would attempt to load the page into the buffer pool some time before trying to dump the contents. Here is a recent failure : CURRENT_TEST: innodb.innodb_information_schema --- /usr/share/mysql/mysql-test/suite/innodb/r/innodb_information_schema.result 2018-02-25 22:27:15.000000000 -0500 +++ /dev/shm/var/4/log/innodb_information_schema.reject 2018-02-25 23:57:20.610548619 -0500 @@ -1,10 +1,10 @@ lock_mode lock_type lock_table lock_index lock_rec lock_data -X RECORD `test`.```t'\"_str` PRIMARY 2 '1', 'abc', '''abc', 'abc''', 'a''bc', 'a''bc''', '''abc''''' +X RECORD `test`.```t'\"_str` PRIMARY 2 NULL +X RECORD `test`.```t'\"_str` PRIMARY 4 NULL +X RECORD `test`.```t'\"_str` PRIMARY 4 NULL X RECORD `test`.```t'\"_str` PRIMARY 2 '1', 'abc', '''abc', 'abc''', 'a''bc', 'a''bc''', '''abc''''' X RECORD `test`.```t'\"_str` PRIMARY 3 '2', 'abc', '"abc', 'abc"', 'a"bc', 'a"bc"', '"abc""' X RECORD `test`.```t'\"_str` PRIMARY 3 '2', 'abc', '"abc', 'abc"', 'a"bc', 'a"bc"', '"abc""' -X RECORD `test`.```t'\"_str` PRIMARY 4 '3', 'abc', '\\abc', 'abc\\', 'a\\bc', 'a\\bc\\', '\\abc\\\\' -X RECORD `test`.```t'\"_str` PRIMARY 4 '3', 'abc', '\\abc', 'abc\\', 'a\\bc', 'a\\bc\\', '\\abc\\\\' X RECORD `test`.```t'\"_str` PRIMARY 5 '4', 'abc', '\0abc', 'abc\0', 'a\0bc', 'a\0bc\0', 'a\0bc\0\0' X RECORD `test`.```t'\"_str` PRIMARY 5 '4', 'abc', '\0abc', 'abc\0', 'a\0bc', 'a\0bc\0', 'a\0bc\0\0' X RECORD `test`.`t_min` PRIMARY 2 -128, 0, -32768, 0, -8388608, 0, -2147483648, 0, -9223372036854775808, 0   mysqltest: Result length mismatch Because this is the only test that covers the conversion of the record values, I think that we must include the column in the output. A simple fix might be to re-bootstrap the server before the test, so that no background operations (change buffer merge, purge) can cause the pages to be evicted. The test is only inserting a few records; it does not use much buffer pool by itself. Another possibility would be to restart the server with innodb_force_recovery=4 between the INSERT and SELECT . This could corrupt any secondary indexes, but the test itself is not creating any tables with secondary indexes.

Marko Mäkelä added a comment - 2020-08-26 05:42 It turns out that 10.5 is affected too, and because 10.5 includes MDEV-19514 , my earlier idea of using innodb_force_recovery=4 might not have worked. http://buildbot.askmonty.org/buildbot/builders/kvm-rpm-centos74-amd64-debug/builds/4542 10.5 a18639f1a913b446f32d7fbe531aa0d5782cf720 innodb.innodb_information_schema 'innodb' w2 [ fail ] Test ended at 2020-07-23 22:57:30   CURRENT_TEST: innodb.innodb_information_schema --- /usr/share/mysql-test/suite/innodb/r/innodb_information_schema.result 2020-07-23 18:55:27.000000000 +0000 +++ /dev/shm/var/2/log/innodb_information_schema.reject 2020-07-23 22:57:30.318387939 +0000 @@ -1,12 +1,12 @@ lock_mode lock_type lock_table lock_index lock_rec lock_data -X RECORD `test`.```t'\"_str` PRIMARY 2 '1', 'abc', '''abc', 'abc''', 'a''bc', 'a''bc''', '''abc''''' -X RECORD `test`.```t'\"_str` PRIMARY 2 '1', 'abc', '''abc', 'abc''', 'a''bc', 'a''bc''', '''abc''''' -X RECORD `test`.```t'\"_str` PRIMARY 3 '2', 'abc', '"abc', 'abc"', 'a"bc', 'a"bc"', '"abc""' -X RECORD `test`.```t'\"_str` PRIMARY 3 '2', 'abc', '"abc', 'abc"', 'a"bc', 'a"bc"', '"abc""' -X RECORD `test`.```t'\"_str` PRIMARY 4 '3', 'abc', '\\abc', 'abc\\', 'a\\bc', 'a\\bc\\', '\\abc\\\\' -X RECORD `test`.```t'\"_str` PRIMARY 4 '3', 'abc', '\\abc', 'abc\\', 'a\\bc', 'a\\bc\\', '\\abc\\\\' -X RECORD `test`.```t'\"_str` PRIMARY 5 '4', 'abc', '\0abc', 'abc\0', 'a\0bc', 'a\0bc\0', 'a\0bc\0\0' -X RECORD `test`.```t'\"_str` PRIMARY 5 '4', 'abc', '\0abc', 'abc\0', 'a\0bc', 'a\0bc\0', 'a\0bc\0\0' +X RECORD `test`.```t'\"_str` PRIMARY 5 NULL +X RECORD `test`.```t'\"_str` PRIMARY 3 NULL +X RECORD `test`.```t'\"_str` PRIMARY 5 NULL +X RECORD `test`.```t'\"_str` PRIMARY 2 NULL +X RECORD `test`.```t'\"_str` PRIMARY 4 NULL +X RECORD `test`.```t'\"_str` PRIMARY 2 NULL +X RECORD `test`.```t'\"_str` PRIMARY 3 NULL +X RECORD `test`.```t'\"_str` PRIMARY 4 NULL X RECORD `test`.`t_min` PRIMARY 2 -128, 0, -32768, 0, -8388608, 0, -2147483648, 0, -9223372036854775808, 0 X RECORD `test`.`t_min` PRIMARY 2 -128, 0, -32768, 0, -8388608, 0, -2147483648, 0, -9223372036854775808, 0 X RECORD `test`.`t_max` PRIMARY 2 127, 255, 32767, 65535, 8388607, 16777215, 2147483647, 4294967295, 9223372036854775807, 18446744073709551615   mysqltest: Result length mismatch

Marko Mäkelä added a comment - 2022-01-14 14:29 I thought that the easiest fix was to filter out the lock_data column from the output. It is not perfect, because our regression test will now be unable to catch real regressions in that functionality.