Parallel installs

Parallel installs enable you to run multiple instances of the same snap on the same system. Each instance is completely isolated from all other instances, including its name, configuration, interface connections, data locations, services, applications and aliases.

This feature is currently considered experimental. In particular, snap install will fail if the same snap is already installed. As a result, to experiment with parallel installs, an experimental feature-flag must first be enabled:

sudo snap set system experimental.parallel-instances=true

We recommend rebooting the system after toggling the experimental.parallel-instances flag state to avoid potential namespace problems with snap applications that have already been run.

Installing multiple instances

The process for installing multiple instances of a snap is identical to installing a single snap except you must provide a unique identifier, called an instance key, separated by an underscore (_) from the target snap name.

For example, the following command will install two instances of the hello-world snap with the second given an instance key of foo:

sudo snap install hello-world hello-world_foo

To specify a specific channel and track for each instance, install each instance separately.

For example, to install Juju 2.9 alongside Juju 3.0, run the following commands:

sudo snap install --channel 2.9/stable juju_29 --classic
sudo snap install --channel 3.0/stable juju_30 --classic

When installing from snap file, the instance key is set by passing --name=<snap>_<instance> explicitly:

sudo snap install --name hello-world_foo hello-world_27.snap

Instance key naming

The instance key needs to be manually appended to the snap name, and takes the following format: <snap>_<instance-key>

For example, the following are valid instance names, for example:

  • hello-world

  • hello-world_foo

  • hello-world_0123456789

Only lowercase letters or digits are valid, and the instance name can be up to 10 characters long. After being set during the installation of a snap, an instance name cannot be changed.

ⓘ The instance key must match the following regular expression: ^[a-z0-9]{1,10}$.

Instance management

When working with instances, the vast majority of snap commands function just as they would with a single snap. To remove an instance, for example, use remove:

$ snap remove hello-world_foo
hello-world_foo removed

Interfaces

Interfaces work across multiple snap instances just as they do from any one snap to another. For example, xkcd-webserver includes a network plug, as will all of its instances, any of which can be connected to the system’s :network slot:

$ snap connections xkcd-webserver_foo
Interface     Plug                             Slot           Notes
network       xkcd-webserver_foo:network       :network       -
network-bind  xkcd-webserver_foo:network-bind  :network-bind  -

$ sudo snap disconnect xkcd-webserver_foo:network

$ snap connections xkcd-webserver_foo
Interface     Plug                             Slot           Notes
network       xkcd-webserver_foo:network       -              -
network-bind  xkcd-webserver_foo:network-bind  :network-bind  -

$ sudo snap connect xkcd-webserver_foo:network

$ snap connections xkcd-webserver_foo
Interface     Plug                             Slot           Notes
network       xkcd-webserver_foo:network       :network       manual
network-bind  xkcd-webserver_foo:network-bind  :network-bind  -

Services

As with Interfaces, Services function the same with multiple instances of a snap as they do from any one snap to another.

However, with multiple instances, you’re more likely to run into port allocation issues, such as two web servers needing access to port 80. In such cases, only the service from one instance will be active.

$ snap services
Service                            Startup  Current
xkcd-webserver.xkcd-webserver      enabled  active
xkcd-webserver_foo.xkcd-webserver  enabled  inactive

You can see why xkcd-webserver_foo.xkcd-webserver is inactive by looking at its logs:

$ sudo snap logs xkcd-webserver_foo.xkcd-webserver
2018-10-03T12:31:59Z xkcd-webserver_foo.xkcd-webserver[1760]: OSError: [Errno 98] Address already in use
(...)

As with single snap service collisions, the solution is to stop the service on one instance and start the service on the other:

$ sudo snap stop xkcd-webserver.xkcd-webserver
Stopped.

$ sudo snap start xkcd-webserver_foo.xkcd-webserver
Started.

$ snap services
Service                            Startup  Current
xkcd-webserver.xkcd-webserver      enabled  inactive
xkcd-webserver_foo.xkcd-webserver  enabled  active

Application names and aliases

Snap application names for multiple instances are adjusted according to the following pattern:

<instance-name>.<app>

When an application name matches the snap name, a short alias is created to match the instance name.

For example, with hello-world and hello-world_foo installed, hello-world.env has an alias of hello-world_foo.env in the _foo instance:

$ hello-world.env  |grep SNAP_INSTANCE_NAME
SNAP_INSTANCE_NAME=hello-world
$ hello-world_foo.env  |grep SNAP_INSTANCE_NAME
SNAP_INSTANCE_NAME=hello-world_foo

As with regular snaps, aliases can be added separately:

$ sudo snap alias hello-world_bar bar_env
Added:
  - hello-world_bar as bar_env

$ sudo snap alias hello-world_foo foo_env
Added:
  - hello-world_foo as foo_env

$ snap aliases
Command          Alias    Notes
hello-world_bar  bar_env  manual
hello-world_foo  foo_env  manual

Aliases from instances generate conflict errors, just as they would with distinct snaps:

$ sudo snap alias hello-world_foo bar_env
error: cannot perform the following tasks:
- Setup manual alias "bar_env" => "hello-world" for snap "hello-world_foo"
(cannot enable alias "bar_env" for "hello-world_foo", already enabled for "hello-world_bar")

When aliases trigger a conflict during snap installation, try passing --unaliased with the command to disable automatic alias generation:

$ sudo snap install snap-with-conflicting-alias_foo --unaliased

Snap environment, data and namespace

When a snap application is run, its environment is populated with a number of SNAP* environment variables.

The following environmental variables are potentially affected when installing multiple instances of a snap:

  • SNAP_NAME: the name of the snap, eg. hello-world

  • SNAP_INSTANCE_NAME: the name of the instance, eg. hello-world_foo

  • SNAP_INSTANCE_KEY: instance key, eg, foo for hello-world_foo

  • SNAP: location of files for the snap

  • SNAP_DATA: data for this particular snap revision

  • SNAP_COMMON: data shared between revisions of the same snap

  • SNAP_USER_DATA: user data for this particular snap revision

  • SNAP_USER_COMMON: user data shared between revisions of the same snap

  • HOME: user’s home, equivalent to SNAP_USER_DATA

  • XDG_RUNTIME_DIR: user’s XDG runtime directory

For example, if hello-world and its instance, hello-world_foo, are installed, the following name variables would be set:

snap

hello-world

hello-world_foo

SNAP_NAME

hello-world

hello-world

SNAP_INSTANCE_NAME

hello-world

hello-world_foo

SNAP_INSTANCE_KEY

foo

The data and mount points of parallel installed snaps are kept separate on the host filesystem. Assuming the snap mount directory is /snap, the following locations are used:

snap

hello-world

hello-world_foo

mount point

/snap/hello-world/27

/snap/hello-world_foo/27

system data

/var/snap/hello-world/

/var/snap/hello-world_foo/

user data

$HOME/snap/hello-world/

$HOME/snap/hello-world_foo/

Within the mount namespace of a snap, adjustments are made to map instance-specific locations to the snap locations. Taking hello-world as an example, system data locations are set to the following:

snap

hello-world

hello-world_foo

SNAP

/snap/hello-world/27

/snap/hello-world/27

SNAP_DATA

/var/snap/hello-world/27

/var/snap/hello-world/27

SNAP_COMMON

/var/snap/hello-world/common

/var/snap/hello-world/common

However, due to security concerns, user data locations are not similarly adjusted:

snap

hello-world

hello-world_foo

SNAP_USER_DATA

/home//snap/hello-world/27

/home//snap/hello-world_foo/27

SNAP_USER_COMMON

/home//snap/hello-world/common

/home//snap/hello-world_foo/common

HOME

/home//snap/hello-world/27

/home//snap/hello-world_foo/27

XDG_RUNTIME_DIR

/run/user//snap.hello-world

/run/user//snap.hello-world_foo

Current limitations

There are a few limitations with parallel installs that will be addressed in subsequent snapd releases.

User data and runtime locations

The user data locations are kept separate across multiple instances. Applications that hard code paths to either user data directories or the XDG runtime directory may not function correctly as the AppArmor profile has been updated to allow accessing instance specific paths only.

Applications need to use SNAP_USER_DATA, SNAP_USER_COMMON or fallback to using HOME.

Applications built using most popular frameworks, such as Gtk/glib or Qt, and that are already capable of observing both HOME and XDG_RUNTIME_DIR, should continue to work.

Ports, DBus names, shared memory and socket activation

Snaps may provide APIs consumed by other snaps or the host system at some well known locations or addresses, such as DBus service names, /dev/shm objects or semaphores, abstract socket addresses. While the default AppArmor template has been updated to allow only instance specific access, connecting interfaces may allow a wider access capabilities that can lead to conflicts between instances of the same snap.

Ports

When snapd installs a snap with services, the services are automatically started. Services trying to bind to the same port may fail or function incorrectly. Such snaps should deliver means for configuration of the service via configure hooks.

A demo snap https://github.com/bboozzoo/parallel-installs-demo provides an example of reconfiguration via configure hooks.

Repeated failed starts of a service due to a port already in use can trigger systemd start-rate limiting and suppress further restarts. To avoid restart suppression:

  1. Identify the service name:

snap services <snap-instance-name>
  1. Clear the failed state before applying the new port configuration:

sudo systemctl reset-failed snap.<snap-instance-name>.<service-name>.service
  1. Apply the port configuration (this restarts the service through the configure hook):

sudo snap set <snap-instance-name> <key>=<port>

DBus names

Services exporting the API on DBus under a well known name may conflict with other instances of the same snap. Those may require fixes from application or snap developers.

/dev/shm

The default AppArmor profile allows instance specific access only. Applications hardcoding paths can break. Note that some interfaces such as mir allow a wider access to /dev/shm. Such snap would need to be updated to made instance aware.

Socket activation

Snaps can use listen-stream socket definitions of three kinds: a network address, or an AF_UNIX socket addressed by either a filesystem path or an abstract name. Each behaves differently with parallel installs:

  • Network sockets (for example a bare <port>) are not adjusted for parallel installs. Instances trying to bind to the same address and port collide in the same way described above under Ports. Since the address is hard-coded, using AF_INET/AF_INET6 socket activation is not recommended for snaps that want to support parallel installs.

  • AF_UNIX filesystem-path sockets using $SNAP_DATA, $SNAP_COMMON or $XDG_RUNTIME_DIR for system daemons, or $SNAP_USER_DATA, $SNAP_USER_COMMON or $XDG_RUNTIME_DIR for user daemons, are automatically expanded to instance-specific locations, so the clients can use the appropriate path, for example /var/snap/hello-world_foo/common/name.socket.

  • AF_UNIX abstract sockets must be declared with a @snap.<snap-name>. prefix. When generating the systemd socket unit for a snap instance, snapd automatically remaps this prefix to @snap.<instance-name>., so each instance listens on its own abstract address without colliding with other instances. Clients must connect using the instance-qualified address, for example @snap.hello-world_foo.socket rather than @snap.hello-world.socket.

If, after any variable expansion or instance remapping, the resulting socket address exceeds the maximum length of 108 bytes allowed for a Unix domain socket address, snapd fails to generate the socket unit and reports an error.