Monitoring and Production Guard

Monitoring and production guard

English · Русский

Host monitor

An SSH session opens a monitor screen of its own — a single exec round trip per interval, reading /proc, df and systemctl. The poll interval is adjustable and stated in the panel.

Card	What it shows
CPU	usage with history, core count, load average
Memory	used / total, swap
Disk	used / total / free per filesystem, as levels
Network	interface, ↑/↓ rates
Top processes	command, PID, CPU, RSS
Services and mounts	systemd units and their state, mount points
Containers	name, image, status
Alerts	thresholds crossed, and when they clear

Alerts are evaluated on the device — disk above the limit, memory or swap over a percentage, load above the core count — and clear themselves when the value returns.

A host that reports nothing usable says Metrics are unavailable on this host instead of drawing zeroes.

Production guard

Tag a host prod and the guard turns on for its sessions.

• Connecting to a #prod host confirms first, and says what the session will ask about.
• Every command is scored before it runs. Risky ones ask for confirmation and name the reason.
• Broadcast to several sessions asks when any of them is on a production host.
• Snippets ask before running on a production host.

The confirmation quotes the command as it will be sent: wrapped rather than scrolled sideways, cut at a stated length, with anything that does not draw as itself (bidi overrides, control characters) spelled out as its code point.

What counts as risky

Class	Examples of what is matched
Destruction	recursive force delete, delete of a broad path, mirror delete, bulk delete, truncating a file
Devices	writes straight to a disk device
Remote code	a download piped into a shell, output piped into an interpreter
Availability	power off or reboot, stopping or disabling a service, flushing firewall rules, killing processes
Privilege and permissions	elevated execution, recursive permission or ownership change on a system path, world-writable grants, overwriting a security-critical file
Packages and history	uninstalling packages, discarding local changes or rewriting git history
Resources	fork bombs

If part of the input exceeds what the guard can read, it says so (Exceeds what the guard reads) rather than passing it as safe.

The guard is a safety net for typos and pasted lines, not a security boundary: it reads the text you are about to send, and a determined obfuscation defeats it.