OOM Killer Resolutions: Processes with High Risk of Memory Leak

Load Balancers

When a load-balancer process is terminated, the handled traffic may simply require more memory for normal operation.

Java Application Servers

Review the JVM memory settings and adjust them to match the application requirements.

java -Xmx2048m -Xms256m

• -Xmx defines the maximum JVM heap size.
• -Xms defines the initial heap size.
• Also review -Xmn when tuning the young-generation heap.
• Remember that a JVM uses additional memory outside the Java heap.
• Review garbage-collector settings when application-specific tuning is required.

PHP Application Servers

For Apache-based PHP servers, inspect the average memory consumed by each httpd process, disable automatic tuning when manual values are required, and reduce the process limits.

(Total_RAM - 5%) / Average_RAM

• Remove the automatic-configuration marker from /etc/httpd/httpd.conf.
• Use the calculated result to reduce ServerLimit and MaxClients.
• When memory per process keeps increasing, reduce MaxRequestsPerChild to approximately 1000–5000.

For NGINX PHP, connect through SSH and inspect the PHP-FPM process sizes using tools such as ps or top.

• If most PHP-FPM processes use about 50–100 MB, disable auto-configuration and reduce max_children.
• If process sizes vary significantly or exceed about 200–300 MB, inspect the application for a memory leak, optimise the code, or reduce max_requests_per_child.

Ruby Application Servers

Memory leaks are relatively common in Ruby applications. Start by inspecting and optimising the application code. Increasing the node RAM is an alternative when the workload legitimately requires more memory.

Python Application Servers

For Apache-based Python servers, use the same httpd process-sizing procedure described for PHP. When another Python process is killed, the application may simply need more RAM.

Node.js Application Servers

Restart the container to restore the terminated process. When the issue returns, increase the RAM limit because the application or build operation may require more memory.

Database Servers

MySQL

For InnoDB, inspect the current buffer usage:

SHOW ENGINE INNODB STATUS\G;

• If buffers use more than about 80% of the container RAM, reduce innodb_buffer_pool_size in /etc/my.cnf.
• When buffer usage is not excessive, allocate more RAM.
• Review MySQL logs for warnings and recommendations.

PostgreSQL

Redis

Common Processes for Different Stacks

What’s next?