open-nomad/website/content/docs/job-specification/restart.mdx

---
layout: docs
page_title: restart Block - Job Specification
description: The "restart" block configures a group's behavior on task failure.
---

# `restart` Block

<Placement
  groups={[
    ['job', 'group', 'restart'],
    ['job', 'group', 'task', 'restart'],
  ]}
/>

The `restart` block configures a task's behavior on task failure. Restarts
happen on the client that is running the task.

```hcl
job "docs" {
  group "example" {
    restart {
      attempts = 3
      delay    = "30s"
    }
  }
}
```

If specified at the group level, the configuration is inherited by all
tasks in the group, including any [sidecar tasks][sidecar_task]. If
also present on the task, the policy is merged with the restart policy
from the encapsulating task group.

For example, assuming that the task group restart policy is:

```hcl
restart {
  interval = "30m"
  attempts = 2
  delay    = "15s"
  mode     = "fail"
}
```

and the individual task restart policy is:

```hcl
restart {
  attempts = 5
}
```

then the effective restart policy for the task will be:

```hcl
restart {
  interval = "30m"
  attempts = 5
  delay    = "15s"
  mode     = "fail"
}
```

Because sidecar tasks don't accept a `restart` block, it's recommended
that you set the `restart` for jobs with sidecar tasks at the task
level, so that the Connect sidecar can inherit the default `restart`.

## `restart` Parameters

- `attempts` `(int: <varies>)` - Specifies the number of restarts allowed in the
  configured interval. Defaults vary by job type, see below for more
  information.

- `delay` `(string: "15s")` - Specifies the duration to wait before restarting a
  task. This is specified using a label suffix like "30s" or "1h". A random
  jitter of up to 25% is added to the delay.

- `interval` `(string: <varies>)` - Specifies the duration which begins when the
  first task starts and ensures that only `attempts` number of restarts happens
  within it. If more than `attempts` number of failures happen, behavior is
  controlled by `mode`. This is specified using a label suffix like "30s" or
  "1h". Defaults vary by job type, see below for more information.

- `mode` `(string: "fail")` - Controls the behavior when the task fails more
  than `attempts` times in an interval. For a detailed explanation of these
  values and their behavior, please see the [mode values section](#mode-values).

### `restart` Parameter Defaults

The values for many of the `restart` parameters vary by job type. Here are the
defaults by job type:

- The default batch restart policy is:

  ```hcl
  restart {
    attempts = 3
    delay    = "15s"
    interval = "24h"
    mode     = "fail"
  }
  ```

- The default service and system job restart policy is:

  ```hcl
  restart {
    interval = "30m"
    attempts = 2
    delay    = "15s"
    mode     = "fail"
  }
  ```

### `mode` Values

This section details the specific values for the "mode" parameter in the Nomad
job specification for constraints. The mode is always specified as a string:

```hcl
restart {
  mode = "..."
}
```

- `"delay"` - Instructs the client to wait until another `interval`
  before restarting the task.

- `"fail"` - Instructs the client not to attempt to restart the task
  once the number of `attempts` have been used. This is the default
  behavior. This mode is useful for non-idempotent jobs which are
  unlikely to succeed after a few failures. The allocation will be
  marked as failed and the scheduler will attempt to reschedule the
  allocation according to the
  [`reschedule`] block.

### `restart` Examples

With the following `restart` block, a failing task will restart 3
times with 15 seconds between attempts, and then wait 10 minutes
before attempting another 3 attempts. The task restart will never fail
the entire allocation.

```hcl
restart {
  attempts = 3
  delay    = "15s"
  interval = "10m"
  mode     = "delay"
}
```

With the following `restart` block, a task that fails after 1
minute, after 2 minutes, and after 3 minutes will be restarted each
time. If it fails again before 10 minutes, the entire allocation will
be marked as failed and the scheduler will follow the group's
[`reschedule`] specification, possibly resulting in a new evaluation.

```hcl
restart {
  attempts = 3
  delay    = "15s"
  interval = "10m"
  mode     = "fail"
}
```

[sidecar_task]: /nomad/docs/job-specification/sidecar_task
[`reschedule`]: /nomad/docs/job-specification/reschedule