Recently, I’ve helped customer to migrate their service from single VM to 1 vm per-service. I am able to manage minimum downtime when migrating the web data and database data to each VM. All VM using latest OS and latest stack version in the repository’s.

All new VM and the dedicated service is working as expected. Then we found increase of throughput in the front-end. Further investigation shows that there a bottleneck in from the database vm, where we found an “Out of Memory”. From the log and monitoring statistic, i found that the memory usage increase dramatically once the mysql have high access but not releasing the memory until the mysqld service OOM. Even though on old environment, there is no such issue with the same configuration. By the way, old database use MariaDB 10.2 and the new one is using MariaDB 10.4.12

I have tried to tune the configuration to reduce the memory footprint, reduce the connection also tune the system kernel. But the same issue keep happening.

After searching on the web about this unexpected behavior, i got some pople that have the same issue with but there is no specific solution.
From the bug report Old Mariadb, 10.2-3 there is a memory leak issue with “Unresolved” status. Here’s the link.

Further reading result, this condition may caused by the memory allocation handler. Using this keyword i manage to found someone that have the same issue and able to resolved it by changing the memory allocation library of their MariaDB server. Please use the link to find the related infromation here.

Based on the server status, i found that the MariaDB 10.4 still using system default for their memory allocation.

Then i am install the jmalloc library to the system so we can use it as memory allocation library on MariaDB 10.4

Then update the env configuration to use new jmalloc library for MariaDB.

Validate the new memory allocation library.

Thank You.

Previous ArticleNext Article

Leave a Reply

Your email address will not be published. Required fields are marked *