consul/website/content/commands/exec.mdx

---
layout: commands
page_title: 'Commands: Exec'
description: >-
  The exec command provides a mechanism for remote execution. For example, this
  can be used to run the `uptime` command across all machines providing the
  `web` service.
---

# Consul Exec

Command: `consul exec`

The `exec` command provides a mechanism for remote execution. For example,
this can be used to run the `uptime` command across all machines providing
the `web` service.

Remote execution works by specifying a job, which is stored in the KV store.
Agents are informed about the new job using the [event system](/consul/commands/event),
which propagates messages via the [gossip protocol](/consul/docs/architecture/gossip).
As a result, delivery is best-effort, and there is **no guarantee** of execution.

While events are purely gossip driven, remote execution relies on the KV store
as a message broker. As a result, the `exec` command will not be able to
properly function during a Consul outage.

**Verbose output warning:** use care to make sure that your command does not
produce a large volume of output. Writes to the KV store for this output go
through the Consul servers and the Raft consensus algorithm, so having a large
number of nodes in the cluster flow a large amount of data through the KV store
could make the cluster unavailable.

The table below shows the [required ACLs](/consul/api-docs/api-structure#authentication) in order to
execute this command.

| ACL Required    | Scope             |
| --------------- | ----------------- |
| `agent:read`    | local agent       |
| `session:write` | local agent       |
| `key:write`     | `"_rexec"` prefix |
| `event:write`   | `"_rexec"` prefix |

In addition to the above, the policy associated with the [agent token](/consul/docs/security/acl/acl-tokens#acl-agent-token) should have `write` on `"_rexec"` key prefix. This is for the agents to read the `exec` command and write its output back to the KV store.

## Usage

Usage: `consul exec [options] [-|command...]`

The only required option is a command to execute. This is either given
as trailing arguments, or by specifying `-`; STDIN will be read to
completion as a script to evaluate.

#### Command Options

- `-prefix` - Key prefix in the KV store to use for storing request data.
  Defaults to `_rexec`.

- `-node` - Regular expression to filter nodes which should evaluate the event.

- `-service` - Regular expression to filter to only nodes with matching services.

- `-shell` - Optional, use a shell to run the command. The default value is true.

- `-tag` - Regular expression to filter to only nodes with a service that has
  a matching tag. This must be used with `-service`. As an example, you may
  do `-service mysql -tag secondary`.

- `-wait` - Specifies the period of time in which no agent's respond before considering
  the job finished. This is basically the quiescent time required to assume completion.
  This period is not a hard deadline, and the command will wait longer depending on
  various heuristics.

- `-wait-repl` - Period to wait after writing the job specification for replication.
  This is a heuristic value and enables agents to do a stale read of the job. Defaults
  to 200 msec.

- `-verbose` - Enables verbose output.

#### API Options

@include 'http_api_options_client.mdx'

@include 'http_api_options_server.mdx'