Details
-
Bug
-
Status: Open (View Workflow)
-
Major
-
Resolution: Unresolved
-
10.11.6, 10.11.7
-
None
-
None
-
None
-
CentOS7 VM, 16GB RAM, 1TB storage Mariadb 10.11.7
Description
I have a Galera Cluster running which was failing to insert data on large ingests. After removing Galera I discovered it was still occurring on a stand alone node. Trial and error found that innodb_autoinc_lock_mode=2 was the culprit. Whenever it is set to 2 and I perform a concurrent ingest of 100 or so files of data into a database I get multiple deadlocks and data is not reliably inserted. For the test to reproduce the issue
Create table 'loadtest' ('id' int(11) unsigned NOT NULL AUTO_INCREMENT,
'val' int(11),
PRIMARY KEY ('id'),
UNIQUE KEY ('val')
) ENGINE=InnoDB
Our test script uses a collection of 2000 files that contain the values. It spins up 100 threads running load data on each file then checks to verify that the data was inserted properly.
load data concurrent local infile 'filename1' ignore into table loadtest (val);
then
select from loadtest where val = 'value';
With innodb_autoinc_lock_mode=2 multiple deadlocks occur and there is dropped data without any errors. With innodb_autoinc_lock_mode=1 there are no issues at all.
This issue was observed in 10.11.6 and 10.11.7
My goal is to migrate to Galera, but this appears to be a roadblock.