rkt run

Image Addressing

Images can be run by either their name, their hash, an explicit transport address, or a Docker registry URL. rkt will automatically fetch them if they're not present in the local store.

Run By Name

# rkt run coreos.com/etcd:v2.0.0

Run By Hash

# rkt run sha512-fa1cb92dc276b0f9bedf87981e61ecde

Run By ACI Address

# rkt run https://github.com/coreos/etcd/releases/download/v2.0.0/etcd-v2.0.0-linux-amd64.aci

Run From a Docker Registry

# rkt --insecure-options=image run docker://quay.io/coreos/etcd:v2.0.0

Run multiple applications in the same pod

Multiple applications can be run in a pod by passing multiple images to the run command:

# rkt run example.com/app1 example.com/app2

Overriding the app's name

Be default, the image's name will be used as the app's name. It can be overridden by rkt using the --name flag. This comes handy when we want to run multiple apps using the same image:

# rkt --insecure-options=image run docker://busybox --name=busybox1 docker://busybox --name=busybox2

Overriding Executable to launch

Application images include an exec field that specifies the executable to launch. This executable can be overridden by rkt using the --exec flag:

# rkt --insecure-options=image run docker://busybox --exec /bin/date

Overriding Isolators

Application images can include per-app isolators and some of them can be overridden by rkt. The units come from the Kubernetes resource model. In the following example, the CPU isolator is defined to 750 milli-cores and the memory isolator limits the memory usage to 128MB.

# rkt run coreos.com/etcd:v2.0.0 --cpu=750m --memory=128M

Overriding User/Group

Application images must specify the username/group or the UID/GID the app is to be run as as specified in the Image Manifest Schema. The user/group can be overridden by rkt using the --user and --group flags:

# rkt --insecure-options=image run docker://busybox --user=1000 --group=100 --exec id

Passing Arguments

To pass additional arguments to images use the pattern of image1 -- [image1 flags] --- image2 -- [image2 flags]. For example:

# rkt run example.com/worker -- --loglevel verbose --- example.com/syncer -- --interval 30s

This can be combined with overridden executables:

# rkt run example.com/worker --exec /bin/ov -- --loglevel verbose --- example.com/syncer --exec /bin/syncer2 -- --interval 30s

Adding user annotations and user labels

Additional annotations and labels can be added to the app by using --user-annotation and --user-label flag. The annotations and labels will appear in the app's UserAnnotations and UserLabels field.

# rkt run example.com/example --user-annotation=foo=bar --user-label=hello=world

Influencing Environment Variables

To inherit all environment variables from the parent, use the --inherit-env flag.

To explicitly set environment variables for all apps, use the --set-env flag.

To explicitly set environment variables for all apps from a file, use the --set-env-file flag. Variables are expected to be in the format VAR_NAME=VALUE separated by the new line character \n. Lines starting with # or ; and empty ones will be ignored.

To explicitly set environment variables for each app individually, use the --environment flag.

The precedence is as follows with the last item replacing previous environment entries:

  • Parent environment
  • App image environment
  • Explicitly set environment variables for all apps from file (--set-env-file)
  • Explicitly set environment variables for all apps on command line (--set-env)
  • Explicitly set environment variables for each app on command line (--environment)
# export EXAMPLE_ENV=hello
# export EXAMPLE_OVERRIDE=under
# rkt run --inherit-env --set-env=FOO=bar --set-env=EXAMPLE_OVERRIDE=over example.com/env-printer

# export EXAMPLE_ENV=hello
# export EXAMPLE_OVERRIDE=under
# rkt run --inherit-env --set-env=FOO=bar --set-env=EXAMPLE_OVERRIDE=over example.com/env-printer --environment=EXAMPLE_OVERRIDE=ride

Disable Signature Verification

If desired, --insecure-options=image can be used to disable this security check:

# rkt --insecure-options=image run coreos.com/etcd:v2.0.0
rkt: searching for app image coreos.com/etcd:v2.0.0
rkt: fetching image from https://github.com/coreos/etcd/releases/download/v2.0.0/etcd-v2.0.0-linux-amd64.aci
rkt: warning: signature verification has been disabled

Mount Volumes into a Pod

Each ACI can define a list of mount points that the app is expecting external data to be mounted into:

    "acKind": "ImageManifest",
    "name": "example.com/app1",
    "app": {
        "mountPoints": [
                "name": "data",
                "path": "/var/data",
                "readOnly": false,
                "recursive": true

To fulfill these mount points, volumes are used. A volume is assigned to a mount point if they both have the same name. There are today two kinds of volumes:

  • host volumes that can expose a directory or a file from the host to the pod.
  • empty volumes that initialize an empty storage to be accessed locally within the pod. When the pod is garbage collected, it will be removed.

Each volume can be selectively mounted into each application at differing mount points. Note that any volumes that are specified but do not have a matching mount point (or --mount flag) will be silently ignored.

If a mount point is specified in the image manifest but no matching volume is found, an implicit empty volume will be created automatically.

Mounting Volumes

Volumes are defined via the --volume flag, the volume is then mounted into each app running in the pod based on information defined in the ACI manifest.

There are two kinds of volumes, host and empty.

Host Volumes

For host volumes, the --volume flag allows you to specify the volume name, the location on the host, and whether the volume is read-only or not. The volume name and location on the host are mandatory. The read-only parameter is false by default.


--volume NAME,kind=host,source=SOURCE_PATH,readOnly=BOOL

In the following example, we make the host's /srv/data accessible to app1 on /var/data:

# rkt run --volume data,kind=host,source=/srv/data,readOnly=false example.com/app1

If you don't intend to persist the data and you just want to have a volume shared between all the apps in the pod, you can use an empty volume:

# rkt run --volume data,kind=empty,readOnly=false example.com/app1

Empty Volumes

For empty volumes, the --volume flag allows you to specify the volume name, and the mode, UID and GID of the generated volume. The volume name is mandatory. By default, mode is 0755, UID is 0 and GID is 0.


--volume NAME,kind=empty,mode=MODE,uid=UID,gid=GID

In the following example, we create an empty volume for app1's /var/data:

# rkt run --volume data,kind=empty,mode=0700,uid=0,gid=0

Mounting Volumes without Mount Points

If the ACI doesn't have any mount points defined in its manifest, you can still mount volumes using the --mount flag.

With --mount you define a mapping between volumes and a path in the app. This will supplement and override any mount points in the image manifest. In the following example, the --mount option is positioned after the app name; it defines the mount only in that app:

# rkt run --volume logs,kind=host,source=/srv/logs \
        example.com/app1 --mount volume=logs,target=/var/log \
        example.com/app2 --mount volume=logs,target=/opt/log

In the following example, the --mount option is positioned before the app names. It defines mounts on all apps: both app1 and app2 will have /srv/logs accessible on /var/log.

# rkt run --volume logs,kind=host,source=/srv/logs \
       --mount volume=data,target=/var/log \
        example.com/app1 example.com/app2

MapReduce Example

Let's say we want to read data from the host directory /opt/tenant1/work to power a MapReduce-style worker. We'll call this app example.com/reduce-worker.

We also want this data to be available to a backup application that runs alongside the worker (in the same pod). We'll call this app example.com/worker-backup. The backup application only needs read-only access to the data.

Below we show the abbreviated manifests for the respective applications (recall that the manifest is bundled into the application's ACI):

    "acKind": "ImageManifest",
    "name": "example.com/reduce-worker",
    "app": {
        "mountPoints": [
                "name": "work",
                "path": "/var/lib/work",
                "readOnly": false
    "acKind": "ImageManifest",
    "name": "example.com/worker-backup",
    "app": {
        "mountPoints": [
                "name": "work",
                "path": "/backup",
                "readOnly": true

In this case, both apps reference a volume they call "work", and expect it to be made available at /var/lib/work and /backup within their respective root filesystems.

Since they reference the volume using an abstract name rather than a specific source path, the same image can be used on a variety of different hosts without being coupled to the host's filesystem layout.

To tie it all together, we use the rkt run command-line to provide them with a volume by this name. Here's what it looks like:

# rkt run --volume=work,kind=host,source=/opt/tenant1/work \
  example.com/reduce-worker \

If the image didn't have any mount points, you can achieve a similar effect with the --mount flag (note that both would be read-write though):

# rkt run --volume=work,kind=host,source=/opt/tenant1/work \
  example.com/reduce-worker --mount volume=work,target=/var/lib/work \
  example.com/worker-backup --mount volume=work,target=/backup

Now when the pod is running, the two apps will see the host's /opt/tenant1/work directory made available at their expected locations.

Enabling metadata service registration

By default, rkt run will not register the pod with the metadata service. You can enable registration with the --mds-register command line option.

Pod Networking

The run subcommand features the --net argument which takes options to configure the pod's network.

Default contained networking

When the argument is not given, --net=default is automatically assumed and the default contained network will be loaded.

Host networking

Simplified, with --net=host the apps within the pod will share the network stack and the interfaces with the host machine.

# rkt run --net=host coreos.com/etcd:v2.0.0

Strictly seen, this is only true when rkt run is invoked on the host directly, because the network stack will be inherited from the process that is invoking the rkt run command.

Other Networking Examples

More details about rkt's networking options and examples can be found in the networking documentation.

Run rkt as a Daemon

rkt doesn't include any built-in support for running as a daemon. However, since it is a regular process, you can use your init system to achieve the same effect.

For example, if you use systemd, you can run rkt using systemd-run.

If you don't use systemd, you can use daemon as an alternative.

Use a Custom Stage1

rkt is designed and intended to be modular, using a staged architecture.

You can use a custom stage1 by using the --stage1-{url,path,name,hash,from-dir} flags.

# rkt --stage1-path=/tmp/stage1.aci run coreos.com/etcd:v2.0.0

rkt expects stage1 images to be signed except in the following cases:

  • it is the default stage1 image and it's in the same directory as the rkt binary
  • --stage1-{name,hash} is used and the image is already in the store
  • --stage1-{url,path,from-dir} is used and the image is in the default directory configured at build time

For more details see the hacking documentation.

Disabling overlay

rkt uses overlayfs by default when running application containers. This provides immense benefits to performance and efficiency: start times for large containers are much faster, and multiple pods using the same images will consume less disk space and can share page cache entries.

This feature will be disabled automatically if the underlying filesystem does not support overlay fs, see the prepare subcommand for details. This feature can also be explicitly disabled with the --no-overlay option:

# rkt run --no-overlay=true --insecure-options=image coreos.com/etcd:v2.0.0


Flag Default Options Description
--caps-remove none capability to remove (e.g. --caps-remove=CAP_SYS_CHROOT,CAP_MKNOD) Capabilities to remove from the process's capabilities bounding set; all others from the default set will be included.
--caps-retain none capability to retain (e.g. --caps-retain=CAP_SYS_ADMIN,CAP_NET_ADMIN) Capabilities to retain in the process's capabilities bounding set; all others will be removed.
--cpu none CPU units (e.g. --cpu=500m) CPU limit for the preceding image in Kubernetes resource model format.
--dns none IP Address Name server to write in /etc/resolv.conf. It can be specified several times.
--dns-domain none DNS domain (e.g., --dns-domain=example.com) DNS domain to write in /etc/resolv.conf.
--dns-opt none DNS option DNS option from resolv.conf(5) to write in /etc/resolv.conf. It can be specified several times.
--dns-search none Domain name DNS search domain to write in /etc/resolv.conf. It can be specified several times.
--environment none environment variables add to the app's environment variables Set the app's environment variables (example: '--environment=foo=bar').
--exec none Path to executable Override the exec command for the preceding image.
--group root gid, groupname or file path (e.g. --group=core) Group override for the preceding image.
--hosts-entry none an /etc/hosts entry within the container (e.g., --hosts-entry= Entries to add to the pod-wide /etc/hosts. Pass 'host' to use the host's /etc/hosts.
--hostname rkt-$PODUUID A host name Set pod's host name.
--inherit-env false true or false Inherit all environment variables not set by apps.
--interactive false true or false Run pod interactively. If true, only one image may be supplied.
--ipc auto auto, private or parent Whether to stay in the host IPC namespace.
--mds-register false true or false Register pod with metadata service. It needs network connectivity to the host (--net as default, default-restricted, or host).
--memory none Memory units (e.g. --memory=50M) Memory limit for the preceding image in Kubernetes resource model format.
--mount none Mount syntax (e.g. --mount volume=NAME,target=PATH) Mount point binding a volume to a path within an app. See Mounting Volumes without Mount Points.
--name none Name of the app Set the name of the app (example: '--name=foo'). If not set, then the app name default to the image's name
--net default A comma-separated list of networks. (e.g. --net[=n[:args], ...]) Configure the pod's networking. Optionally, pass a list of user-configured networks to load and set arguments to pass to each network, respectively.
--no-overlay false true or false Disable the overlay filesystem.
--oom-score-adjust none adjust /proc/$pid/oom_score_adj oom-score-adj isolator override.
--pod-manifest none A path The path to the pod manifest. If it's non-empty, then only --net, --no-overlay and --interactive will have effect.
--port none A port name and number pair Container port name to expose through host port number. Requires contained network. Syntax: --port=NAME:HOSTPORT The NAME is that given in the ACI. By convention, Docker containers' EXPOSEd ports are given a name formed from the port number, a hyphen, and the protocol, e.g., 80-tcp, giving something like --port=80-tcp:8080.
--private-users false true or false Run within user namespaces.
--pull-policy new never, new, or update Sets the policy for when to fetch an image. See image fetching behavior
--readonly-rootfs none set root filesystem readonly (e.g., --readonly-rootfs=true) if set, the app's rootfs will be mounted read-only
--seccomp none filter override (e.g., --seccomp mode=retain,errno=EPERM,chmod,chown) seccomp filter override
--set-env none An environment variable (e.g. --set-env=NAME=VALUE) An environment variable to set for apps.
--set-env-file none Path of an environment variables file (e.g. --set-env-file=/path/to/env/file) Environment variables to set for apps.
--signature none A file path Local signature file to use in validating the preceding image.
--stage1-from-dir none Image name (e.g. --stage1-name=coreos.com/rkt/stage1-coreos) A stage1 image file name to search for inside the default stage1 images directory.
--stage1-hash none Image hash (e.g. --stage1-hash=sha512-dedce9f5ea50) A hash of a stage1 image. The image must exist in the store.
--stage1-name none Image name (e.g. --stage1-name=coreos.com/rkt/stage1-coreos) A name of a stage1 image. Will perform a discovery if the image is not in the store.
--stage1-path none Absolute or relative path A path to a stage1 image.
--stage1-url none URL with protocol A URL to a stage1 image. HTTP/HTTPS/File/Docker URLs are supported.
--supplementary-gids none supplementary group IDs (e.g., --supplementary-gids=1024,2048) supplementary group IDs override for the preceding image
--user none uid, username or file path (e.g. --user=core) User override for the preceding image.
--user-annotation none annotation add to the app's UserAnnotations field Set the app's annotations (example: '--user-annotation=foo=bar').
--user-label none label add to the apps' UserLabels field Set the app's labels (example: '--user-label=foo=bar').
--uuid-file-save none A file path Write out the pod UUID to a file.
--volume none Volume syntax (e.g. --volume NAME,kind=KIND,source=PATH,readOnly=BOOL) Volumes to make available in the pod. See Mount Volumes into a Pod.
--working-dir none working directory override (e.g. --working-dir=/tmp/bar) Override the working directory in the preceding image.

Per-application options

Flag Default Options Description
--stdin "null" "null", "tty", "stream" Mode for this application stdin.
--stdout "log" "null", "tty", "stream", "log" Mode for this application stdout.
--stderr "log" "null", "tty", "stream", "log" Mode for this application stderr.

Global options

See the table with global options in general commands documentation.