Details
-
Bug
-
Status: Open (View Workflow)
-
Major
-
Resolution: Unresolved
-
10.2.26, 10.2.29, 10.3.17, 10.3.20, 10.4.7, 10.4.10
-
None
-
Encountered on Windows, debugged on Windows.
Description
When upgrading to 10.4.7 and continuing on through 10.4.10 (and equivalent version of 10.2 and 10.3), if I have the RocksDB plugin enabled:
[mysqld]
|
...
|
plugin_load_add = ha_rocksdb
|
Then as soon as I start that instance (a service in my case), the CPU usage spikes to 50% and does not let up until I kill the process.
It may dip to 48-49%.
If I start a second instance, they both end up taking ~49-50% each, and the CPU is pegged at 100%.
If I start a third instance, the 3 generally end up using about ~32-33% each, and the CPU is pegged at 100%.
The instances were only using about 32M RAM, and zero activity was occurring on them.
I started Windows profiler, and attached it (and CPU stats gatherer), allowed it to run for 50 seconds, and analyzed. It showed that the CPU usage came from ha_rocksdb.dll.
I then commented out the RocksDB plugin:
#plugin_load_add = ha_rocksdb
|
Restarted MariaDB.
And then CPU was back to normal, roughly 0% in this case, since there was 0 activity.
Comment: It is easy to remove if you enabled the plugin via the config file: simply remove or comment out the line, as above, and restart MariaDB.
If you enabled it via the command line, then you need to uninstall it via the command line:
UNINSTALL SONAME 'ha_rocksdb';
|
However, this command is just hanging for me in this situation (i.e., when CPU is pegged out). It is essentially stuck. It did finally complete after about 20 minutes