Page 12 - Linux

A monitoring alert reports a Load Average of 12 on a server with four cores, yet CPU utilization barely reaches 20 percent. Anyone who reflexively buys more cores here often fixes the wrong problem. This workflow shows how to trace Load Average layer by layer back to its actual cause using uptime, vmstat, pidstat and iostat.

The PHP-FPM process manager decides how many parallel requests a Magento server can handle before requests start queuing up or too many workers exhaust available memory. Setting pm.max_children by gut feeling risks either wasted capacity or an out of memory crash under load.

MySQL already caches database pages in the InnoDB buffer pool in RAM. If the Linux kernel simultaneously uses its own page cache for the same data, double buffering occurs: the same memory content sits twice in RAM without any extra benefit. O_DIRECT deliberately bypasses the page cache and hands full caching control to InnoDB, while buffered I/O remains the more pragmatic choice in certain scenarios.

Standard logs like syslog or journald only show what applications choose to log themselves, not who changed /etc/passwd or how a process elevated its privileges. auditd hooks directly into the Linux kernel audit subsystem, logs syscalls and file access without gaps, and makes changes forensically traceable with ausearch and aureport, including rules for sensitive configuration files and privilege escalation.

Anyone who sees free -h for the first time often mistakes their system for being short on memory, because buff/cache appears as used. This article explains how the page cache puts idle RAM to work for performance, why swap is an overflow mechanism rather than primary memory, and how to read proc meminfo for an accurate picture of the real memory state.

A server can appear completely full even though df -h still shows plenty of free disk space, because it is not the disk blocks but the number of available inodes that has run out. This article explains what an inode stores, how to reliably diagnose inode exhaustion with df -i, and how PHP applications generating millions of small cache and session files can avoid hitting the limit in the first place.

Anyone hosting several client projects on one server eventually runs into different PHP version requirements, since Magento, WordPress, and legacy applications rarely demand the same version. Versioned FPM packages, update-alternatives, and cleanly separated pools per vhost let you run multiple PHP versions side by side without conflicts and safely in production.

SSH tunneling turns a single encrypted connection into a secure channel for any TCP service that would otherwise only be reachable locally on a server. This article explains local, remote and dynamic port forwarding through concrete examples, shows a tunnel into a database only reachable from the server itself, covers autossh for permanently stable connections, and the security risks of tunnels left open.

A database snapshot taken while writes are actively in flight can be internally inconsistent, even when the backup completes without any error message. Only during the actual restore, often months later, does it become clear that tables and indexes no longer match. This article shows how MariaDB, MySQL, and PostgreSQL on Linux can produce truly consistent snapshots, using locking, Mariabackup, LVM, and WAL archiving.

A cron job that fails silently leaves no error message on a screen anyone reads. This guide explains how MAILTO actually works, how to evaluate exit codes in cron scripts, how healthcheck pings act as a dead man switch to catch missing executions, and how all of this grows into a reliable alerting chain for cron jobs.