> ## Documentation Index
> Fetch the complete documentation index at: https://docs.axottle.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Flow Guard: Recovery

> Manual plans, auto-recovery policy, route reserves, and stop after failure.

Open **Configures → Flow → Recovery**. This is where **A autorestoration**, **Run manually** and **S Your plans**. By default, the changing recovery requires manual confirmation. The operator can allow automatic actions in advance by the policy of a particular Flow.

## Manual launch

1. Open “Run Manually” and select the appropriate plan and goal.
2. Perform a preliminary check. Examine the steps, effects, and causes of the blockage.
3. Confirm the plan shown and launch it.
4. Read the result of the actions and the final check of the route in history.

Successfully sending a task to an agent does not mean recovery. Flow Guard checks the result; if there is insufficient or conflicting data, further diagnosis is required.

## Turning on the automation

1. In Auto-Recovery, select **Restore Auto-**.
2. For the desired transit, WARP or proxy, switch on and mark allowable reserves.
3. Include authorized service recovery plans. The card shows the actions that are going into the resolution.
4. Set the number of failure confirmations, the pause between attempts, and the total hour limit.
5. Click **Save and enable**.

The policy is executed on the server; you don't need to keep the browser open. It can be prepared before the first use of Flow, but the launch will remain blocked until the running version appears.

Confirmations are different sequential checks of one failure and the active version. Reopening one result does not add confirmation. After trying, new checks are needed to re-act; pauses and limits continue to apply.

## Auto-route switching

Reserves are selected from the working version of Flow. First add and apply the desired outputs on the canvas, then allow them in the policy. Reserve, which depends on the faulty site or creates a cycle, can not be used.

| Method of selecting | What is counted by |
| - | - |
| Priority | Order of permitted reserves. |
| Minimum latency | Fresh measured RTT from CTRL to node. |
| Minimum boot | Fresh CPU boot from agent data. |

RTT from CTRL is not a delay from the client, the CPU does not show the bandwidth of the channel. In the absence of measurements, priority selection should be allowed separately. Unknown health is not considered a successful test.

**Direct Internet is used only as an expressly authorized bypass.** Before action, the output is re-checked. Changes to the pool directory not yet applied to Flow require manual application.

Switching creates a new working version from the image used, saves the conditions of incoming communications and the path of the selected reserve, and the current user draft does not unfold or overwrite this switch.

> [!WARNING]
>
> There is no automatic return to the old route. After switching, check the canvas before the next manual application: the old draft may return the faulty path.

## Your plans.

In the script card, select **What to do when a failure occurs · configure steps → Create your own version of this** plan or create a plan under Your plans. Set the event, goal and acceptable actions, perform validation and save.

A saved plan doesn't get permission to run automatically on its own. Include its version in the policy and keep the policy. Once you change the plan, you need to confirm the updated version. Plans with mandatory manual steps or actions that are not automatically triggered.

## Diagnosis of Support Events

In **HY plans → Events that launch the plan → Support and complaints** available fresh complaint about Flow in open appeal, expired obligation SLA urgent ticket, confirmed operator incident and escalation of open appeal. Terms use saved binding to a specific Flow.

Such a diagnosis does not automatically close the ticket or allow the restart of an arbitrary node; an old closed appeal complaint does not become a new failure after reopening without a new binding; retaining a plan requires a separate inclusion of its version in the policy.

The unsaved policy remains in the current browser tab during transitions and reboots. If you conflict with changes from another operator, download the current version and check your edits before saving.

## Stop and fail

Turning off the automation cancels the automatic actions that have not yet been sent; the task already transferred to the agent can be completed; turning off the policy does not roll back the completed steps.

An unconfirmed rerouting result suspends automation. Check the active version, nodes, and launch history, then explicitly resume it. The policy change does not nullify your spent hourly budget.

## Verification

Make sure you keep the correct version of the policy and plans, have a working version of Flow, and the interface shows the reason for the wait or readiness. Once launched, check the history, the steps actually completed, and a fresh route check.

## Rollback

First, disable automatic actions if they interfere with parsing. To return the configuration, use the [working version of Flow](https://docs.axottle.com/en/connection-units/lifecycle.md#откат). Restore the completed service actions according to the result of a specific plan; disabling automation does not constitute a rollback.

Check the availability of features in the installed version in the panel; this reconciliation does not confirm the deployment on the servers.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.