Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Incomplete
-
10.1.38, 10.2.27, 10.2.28, 10.3.20
-
It's running in a 2 nodes galera cluster on mariadb 10.2.27 and debian 9. Both nodes have 16 cores (32 threads) and 196GB ram with 4TB of disk space. One node is the primary node that handles all the connections and the other is the hot standby.
UPDATE: This problem is also happening on 10.3.20 and various server setupsIt's running in a 2 nodes galera cluster on mariadb 10.2.27 and debian 9. Both nodes have 16 cores (32 threads) and 196GB ram with 4TB of disk space. One node is the primary node that handles all the connections and the other is the hot standby. UPDATE: This problem is also happening on 10.3.20 and various server setups
Description
In our situation with galera we have a problem that on random moments the `AUTO_INCREMENT` gets set to a lower value than the maximum value of the primary_key of that table. We have a total of ~700 databases with a total size of ~180GB and in about 24 hours of handling requests 20 tables change to having this problem.
For example:
Database testdb1 has two module tables for car parts: module_cars and module_car_parts. The table module_cars has 100 records and has a AUTO_INCREMENT of 102 (+2 cause of the offset in a node 2 cluster). When I check this table 24 hours later the table got changed to still having 100 records but having a AUTO_INCREMENT of 89. This value can vary from 90% to 10% lower than the max value of the primary_key.
I can't find why this is happening. I found https://jira.mariadb.org/browse/MDEV-6076 where there was a problem with handling AUTO_INCREMENT's but that should've be fixed a long time ago.
See attachment for the `my.cnf`
UPDATE: I have upgraded mariadb to 10.3.20 and removed galera from our setup but without any change. Even our other servers without replication and with master-slave replication keep getting this issue where the AUTO_INCREMENT gets set lower then the max value of the primary_key with AUTO_INCREMENT. This causes major data corruption if this is not contained. Galera seems not to be the only issue here.
Attachments
Issue Links
- relates to
-
MDEV-12123 Page contains nonzero PAGE_MAX_TRX_ID
- Closed
-
MDEV-13094 SHOW CREATE TABLE can report non-persistent AUTO_INCREMENT value before server restart
- Confirmed
-
MDEV-6076 Persistent AUTO_INCREMENT for InnoDB
- Closed
-
MDEV-15400 Upgrading to 10.2 breaks auto increment
- Closed
-
MDEV-33277 In-place migration from MySQL 5.7 causes invalid AUTO_INCREMENT values
- Closed