Alert Rules

MAC ships with a fixed set of alert rules covering node, hardware, container, software, and log conditions. Each rule can be enabled or disabled, and threshold-style rules expose their operator and threshold for editing. Writes are restricted to admin and operator; viewer sees the page in read-only mode.

Columns

Column	Notes
Enabled	Toggle the rule on/off (admin/operator only)
Name	Rule name and one-line description
Metric	The metric the rule watches (e.g. GPU Temperature, Node Offline)
Condition	Threshold expression (> 90 °C) for threshold rules, or Enabled / Disabled for toggle rules
Severity	critical / warning / info
Edit	Opens the edit modal (admin/operator only)

Built-in Rules

The seed set covers three categories:

Threshold rules (operator + threshold are editable):

• GPU Temperature Critical / GPU Temperature Warning
• CPU Overload — CPU utilization above the threshold
• Memory Overload — memory utilization above the threshold
• Storage Nearly Full — any filesystem above the threshold
• Storage SMART Unhealthy — SMART health check failure

Toggle rules (on/off only):

• Node Offline — agent unreachable
• Node Rebooting — node observed in a rebooting state
• Container Dead / Restarting / Unhealthy / Exited
• Software Version Drift — software versions differ across nodes
• Critical Error Logs — kernel panic, OOM, hardware errors, GPU Xid

Filter rule (edits a pattern list):

• Container Whitelist — not an alert itself, but a filter that scopes which containers the four Container alerts above monitor. When enabled, only containers whose names match one of its glob patterns (default mac-*) can raise container alerts; everything else is ignored. This keeps unrelated workloads on a node from firing container alerts. Its effect is also reflected on the Containers page.

Editing a Rule

Click the pencil icon to open the edit modal. Threshold rules expose three fields:

• Severity — critical / warning / info
• Operator — >, <, >=, <=, ==, !=
• Threshold — numeric value (with a unit hint such as °C, %, W)

Toggle rules expose only the severity. The Container Whitelist filter rule instead opens a pattern-list editor where you add or remove glob patterns (e.g. mac-*, gpu-*) that define which containers are monitored. Rule name, description, and metric type are not editable — they are part of the built-in catalog.

Page Actions

• Export CSV — download every rule (name, description, metric, operator, threshold, severity, enabled, category, scope, timestamps) for documentation and audits.
• Reset to Default — restore every rule to its factory state. A confirmation dialog protects against accidents. Admin/operator only.

The footer shows how many rules are currently disabled (e.g. 3 rules disabled) so you can quickly tell whether anything is muted before triaging an outage.