Supervisor Restart Strategies

Supervisors monitor child agents and restart them when they fail. The restart strategy determines how failures are handled.

Available Strategies

one-for-one

When one agent fails, only that agent is restarted. Other agents continue running.

Before:  [Agent A ✓] [Agent B ✗] [Agent C ✓]
After:   [Agent A ✓] [Agent B ✓] [Agent C ✓]
                          ↑
                     restarted

Use when: Agents are independent and don't share state.

naos supervisor create main --strategy one-for-one

one-for-all

When one agent fails, ALL agents are restarted.

Before:  [Agent A ✓] [Agent B ✗] [Agent C ✓]
After:   [Agent A ✓] [Agent B ✓] [Agent C ✓]
              ↑           ↑           ↑
           restarted   restarted   restarted

Use when: Agents share state and must be consistent.

naos supervisor create main --strategy one-for-all

rest-for-one

When one agent fails, that agent and all agents started AFTER it are restarted.

Start order: A → B → C → D
B fails:

Before:  [Agent A ✓] [Agent B ✗] [Agent C ✓] [Agent D ✓]
After:   [Agent A ✓] [Agent B ✓] [Agent C ✓] [Agent D ✓]
                          ↑           ↑           ↑
                     restarted   restarted   restarted

Use when: Later agents depend on earlier agents.

naos supervisor create main --strategy rest-for-one

Choosing a Strategy