#system-administration

A developer successfully migrated a live server by freezing a TCP connection for 10 minutes without disrupting service. The technique involved maintaining an established connection while moving the underlying infrastructure.

The article argues that log messages should primarily serve the people operating software, not just developers. It emphasizes that log content should be meaningful and understandable to system administrators without requiring specialized knowledge. The author suggests developers should consider what operational value each log message provides to those running the software.

The author presents a taxonomy of text output methods from command-line tools, ranking them from best to worst. The preferred approach is plain text line-by-line output, while the most problematic involves full-screen terminal UI elements that disrupt normal command-line workflow. The author argues tools should provide options to disable fancy output and detect when it's inappropriate.

Fedora systems accumulate old GPG keys from RPM packages over time. The clean-rpm-gpg-pubkey tool helps remove obsolete keys, while manual removal may be needed for re-issued expired keys that cause DNF update issues.

A university computing group maintained duplicate systems for tracking disk usage over many years. They evolved from traditional Unix scripts to ZFS-based reporting, then to Prometheus metrics with a new command-line tool. Despite creating modern replacements, they haven't removed the old infrastructure due to cleanup effort and inertia.

The author describes how duplicate systems often linger when a new, improved system is built but doesn't fully replace the old one. Organizations keep the old system running for existing functionality while using the new system for new features, creating little motivation to eliminate the duplicate system until resource constraints or maintenance burdens arise.

The author describes switching libvirt virtual machines from traditional BIOS/MBR booting to UEFI booting by directly editing the libvirt XML configuration. This change became possible after libvirt gained support for UEFI snapshots, and the author provides technical details about the XML modifications required for the transition.

Due to escalating hardware costs, the University of Toronto's computer science department is retaining older systems rather than discarding them. They are slowing down retirement of old servers and acquiring used hardware from other university groups. This approach aims to maintain operational capacity amid rising expenses for new servers.

FreeBSD 15 now allows managing the entire system through the pkg package manager using freebsd-base (pkgbase), moving away from the traditional split where freebsd-update handled the base system. The author reports positive experiences with pkg-based management on multiple FreeBSD 15 installations, finding updates painless and the unified approach easier to manage.

The blog post introduces two small scripts for data processing: 'addup' sums values in a specified column using awk, while 'sumup' either counts occurrences or sums values grouped by another field, available in both awk and Python versions. These tools help with tasks like analyzing Apache log files or processing multi-column data.

The article explains two methods for identifying large RPM packages on Fedora systems. The first shows binary package sizes, while the second aggregates sizes by source package using custom scripts. Both approaches help diagnose disk space usage issues.

Systemd v258 introduces 'systemctl -v' to display unit log output during operations like restart. However, it stops following logs once systemd considers the service started, which may be before the service completes its actual startup process. This limitation means users may still need to use journalctl to see full startup logs for many services.

Maintaining accurate inventories of any kind is difficult because they require keeping records in sync with physical or software reality. Self-reporting systems help but have limitations, and inventory maintenance requires dedicated time and organizational support to be effective.

The article describes a declarative approach to installing NixOS over the network using nixos-anywhere. It explains how to build a customized NixOS installer ISO with pre-configured SSH access and preferred tools, enabling fully automated remote installation.

The article discusses SFTP sandboxing techniques in Unix systems, exploring methods to securely isolate file transfer operations within restricted environments.

The article stresses that backup planning should start with strategy before implementation. It covers key considerations like risk assessment, full disk versus file-level backups, and the importance of snapshots for data consistency. The author advocates for a pull-based architecture with external storage.