Details
-
Bug
-
Status: Closed (View Workflow)
-
Critical
-
Resolution: Fixed
-
13.0, 13.1
-
ARMv8, s390x, AMD64
-
Not for Release Notes
-
A redundant empty checkpoint could be written after an existing one.
-
Q4/2026 Server Maintenance
Description
The test innodb.log_archive that was added in MDEV-37949 is failing frequently:
|
13.0 a848493c6fe031f23606144420c1ca1e467cbd81 |
innodb.log_archive 'mmap' w2 [ fail ]
|
Test ended at 2026-08-11 12:37:41
|
|
|
CURRENT_TEST: innodb.log_archive
|
--- C:/a/server/server/mysql-test/suite/innodb/r/log_archive.result 2026-08-11 12:06:01.462876300 +0000
|
+++ C:\a\server\server\mysql-test\suite\innodb\r\log_archive.reject 2026-08-11 12:37:40.683306200 +0000
|
@@ -62,9 +62,9 @@
|
SELECT variable_name, variable_value FROM information_schema.global_status
|
WHERE variable_name LIKE 'INNODB_LSN%';
|
variable_name variable_value
|
-INNODB_LSN_CURRENT 18446744073705357295
|
-INNODB_LSN_FLUSHED 18446744073705357295
|
-INNODB_LSN_LAST_CHECKPOINT 18446744073705357279
|
+INNODB_LSN_CURRENT 18446744073705357311
|
+INNODB_LSN_FLUSHED 18446744073705357311
|
+INNODB_LSN_LAST_CHECKPOINT 18446744073705357295
|
INNODB_LSN_ARCHIVED 18446744073705357279
|
# restart: --innodb-log-recovery-start=0 --innodb-log-buffer-size=5m
|
SELECT variable_name, variable_value FROM information_schema.global_status
|
|
|
Result content mismatch
|
This failure has been observed both for the mmap and pread test variants. The difference of INNODB_LSN_CURRENT is 16, which corresponds to the size of a FILE_CHECKPOINT record. All test steps before this have been executed with either innodb_read_only or innodb_log_recovery_start being enabled. Therefore, writes to the log file should have been prevented. (Edit: Due to innodb_log_recovery_target=0 writes to the log are not blocked.)
Similar failures have also been observed on an ARMv8 based Apple macOS. On IBM System Z (s390x) and AMD64 the failure is much less frequent.
On any platform, the probability of the failure can be improved with the following patch, suggested by andrzej.jarzabek]:
diff --git a/mysql-test/suite/innodb/t/log_archive.test b/mysql-test/suite/innodb/t/log_archive.test
|
index 925108880ba..ee3fe8ad0f3 100644
|
--- a/mysql-test/suite/innodb/t/log_archive.test
|
+++ b/mysql-test/suite/innodb/t/log_archive.test
|
@@ -178,6 +178,7 @@ EOF
|
--let $restart_parameters= --innodb-log-recovery-start=18446744073705357279
|
--source include/start_mysqld.inc
|
|
+--sleep 2
|
SELECT variable_name, variable_value FROM information_schema.global_status
|
WHERE variable_name LIKE 'INNODB_LSN%';
|
|
|
If we view the LSNs in hexadecimal, it becomes clearer that the failure occurs due to a redundant write of a FILE_CHECKPOINT record. These failures started to occur on 2026-08-20, which is when the fix of MDEV-40791 was pushed. The extra write is caused by log_sys.circular_recovery_from_sequence_bit_0 being unnecessarily set in log_t::set_archive().
Attachments
Issue Links
- is caused by
-
MDEV-37949 Implement innodb_log_archive
-
- Closed
-
-
-
- Closed
-
