Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.6.17, 10.11.7
-
None
-
Os release: Debian 12 "Bookworm"
Kernel: 6.1.0-17-amd64
Hardware CPU: Intel(R) Xeon(R) CPU E5-2630 v3 @ 2.40GHz
Memory: 128G
mariadb-server: 1:10.11.6+maria~deb12
Description
We are facing serious CPU performance issue (on system usage) after upgrading our database from 10.4 to 10.11.
While not 100% sure, we identified that the issue appeared on release 10.6.5 (10.6.4 seems not affected).
How to reproduce:
Create a database named "neutron":
$ mysql -e 'create database neutron' |
Inject the values from the two attached sql files:
$ mysql neutron < networkrbacs2.sql
|
$ mysql neutron < subnets2.sql
|
Execute 150 requests in parallel:
for i in $(seq 1 3); do for j in $(seq 1 50); do (cat get_subnet3 | mysql 2>&1> /dev/null ) & done; sleep 0.5; done |
Take a look at your CPU system usage (it's very high).
Here is a perf.data file that we captured during the requests:
https://plik.ovh/file/UMgOCcPerrToULdK/mnc0kBKJocKmeyTK/perf.data
You will see using:
perf report -i perf.data
|
that it's mostly wasting system CPU in native_queued_spin_lock_slowpath