User Guide


pip install topology

Topology requires at least one Platform Engine which is the engine responsible to build the topology. By default topology just includes a debug dummy platform only. We recommend to use the topology_docker engine which can be installed with:

pip install topology_docker

Further setup is required, in particular it is required to install docker and setup your environment to run some commands as root. See the topology_docker documentation for more information.

Describing topologies

The Topology framework supports a custom textual format that allows to quickly specify simple topologies that are only composed of simple nodes, ports and links between them.

The format for the textual description of a topology is similar to Graphviz syntax and allows to define nodes and ports with shared attributes and links between two endpoints with shared attributes too.

# Nodes
[type=switch attr1=1] sw1 sw2

# Ports
[speed=1000] sw1:3 sw2:3

# Links
[linkattr1=20] sw1: -- sw2:
[linkattr2=40] sw1:3 -- sw2:3

In the above example two nodes with the attributes type and attr1 are specified. Then a third node hs1 with no particular attributes is defined. Later, we specify some attributes (speed) for a couple of ports. In the same way, a link between endpoints MAY have attributes.

An endpoint is a combination of a node and a port name, but the port is optional. If the endpoint just specifies the node (sw1:) then the next available numeric port is implied.

Is up to the Platform Engine to determine what attributes are relevant to it. By default, the only ones interpreted by the framework is identifier which must be an unique UUID (and if missing it will be automatically assigned) and the name which is a human readable name used for plotting.

For a more programmatic format consider the metadict format or using the pynml.manager.ExtendedNMLManager directly.


Topology provides the --topology-log-dir pytest option that allows the user to specify where the logging output will be stored. The shells that extend PExpectShell will log their complete shell output there in files whose name identifies the node producing the output along with the shell.

Injecting attributes

It may be necessary to define some attributes for the nodes outside of the topology definition. Many cases exists for this, like testing a different image using a whole suite of tests, or to specify the IP of a machine to connect to.

These attributes can be injected using an injection file: a JSON file that defines which attributes are to be added to which nodes in which test topologies.

The injection file defines a list of injection specifications, JSON dictionaries with the following keys:

  • files a list of files where to look for nodes.
  • modifiers a list of dictionaries with the following keys:
    • nodes a list of nodes where to look for attributes.
    • attributes a dictionary with the attributes and values to inject.

This is an injection file example:

        "files": ["/path/to/directory/*", "/test/"],
        "modifiers": [
                "nodes": ["sw1", "type=host", "sw3"],
                "attributes": {
                    "image": "image_for_sw1_sw3_and_hosts",
                    "hardware": "hardware_for_sw1_sw3_and_hosts"
                "nodes": ["sw4"],
                "attributes": {
                    "image": "image_for_sw4"
        "files": ["/path/to/directory/"],
        "modifiers": [
                "nodes": ["sw1"],
                "attributes": {
                    "image": "special_image_for_sw1",

In order to avoid lengthy injection files, groups of files or nodes can be defined using shorthands:


The items in the files list are paths for test suites or SZN files.

  • test suites are files that match with test_*.py
  • SZN files are files that match with *.szn

A complete directory can be specified too with /path/to/directory/*.

Only test suites and SZN files are selected when injecting attributes.

Some examples:

  • and can both be selected with test_*
  • Any topology file can be selected with *.szn.
  • will never be selected, even if it is inside a directory specified with /path/to/directory/* because it is neither a test suite nor a configuration file.


Several nodes can be selected too:

  • * will select any node; in a similar fashion hs* will select any node whose name begins with hs.
  • some_attribute=some_value will select any node that has that specific pair of attribute and value already defined (either if that pair was defined in the topology definition or by attribute injection).

Overriding attributes

An attribute that was set in the topology definition will be overriden by another one with the same name in the injection file.

The order in which attributes are defined in the injection file matters. For example, in this example the image attribute for sw1 in /path/to/directory/* was set to image_for_sw1_sw3_and_hosts first, but after this, it was overriden to special_image_for_sw1 because of the second injection specification.

The low-level shell API

The test execution machine communicates with the topology nodes through their shells. This can be done by simply sending a command to the node like this:

ops1('some command', shell='the_node_shell')

This kind of communication is informally known as rapid fire.

Usually, this kind of command execution is not done directly in the test case, but done inside a communication library.

Now, rapid fire allows the user to send commands to the node without much typing but without much fine control also. For certain situations, is necessary to specify several other shell communication parameter besides of the command that is to be sent. For this situations, the low-level shell API comes handy.

Using the API

Each node object has one or more shells that can be used to communicate with it. These shells are closely related to the nature of the node. For example, a host node (basically, a node that represents a computer that runs Linux) has a bash shell, because these Linux hosts come with a bash shell.

These shells are represented by a shell object that can be obtained directly from the node in this way:

bash = hs1.get_shell('bash')

The get_shell method of the nodes returns the shell object specified in its argument. This object is the one who provides the low-level shell API to the user.

The complete API is defined in the methods for the BaseShell class of the module included in the platforms package of topology, found in


The low-level shell API is public and intended for advanced usage. This means that the user will have access to certain shell attributes like the prompt that it should be matching. Misuse of this low-level API can make the high-level API (rapid fire) to get out of sync if the prompt is not rematched properly in the low-level API. Use with caution.

Context manager for non-default shells

To avoid setting a non-default shell repeatedly in rapid fire calls, like in:

hs1('from bar import foo', shell='python')
hs1('foo.something()', shell='python')
hs1('foo.otherthing()', shell='python')

a context manager can be used. For example, this can be done for high-level shell usage:

with hs1.use_shell('python') as python:
    hs1('from bar import foo')

This can be done for low-level shell usage:

with hs1.use_shell('python') as python:
    python.send_command('from bar import foo')
    python.send_command('foo.otherthing()', timeout=99)

Overview of the API methods

There are two main methods that provide most of the functionality in the API:

  1. send_command
  2. get_response

The first one receives the following arguments:

  1. command: the command to be executed in the shell, mandatory
  2. matches: a list of strings that are to be matched by the shell (please see Some examples of send_command), optional, defaults to None
  3. newline: True if the shell should add a return character after the command, False otherwise, optional, defaults to True
  4. timeout: the amount of seconds to wait for the shell to return something that matches with the shell prompt (or with some element of matches if defined), optional, defaults to None
  5. connection: the shell connection to use (this will be explained in more detail in Using multiple shell connections), optional, defaults to None

Some examples of send_command


This is useful for situations where the execution of a command will not return the usual shell prompt. For example, if a command needs confirmation from the user, this can be handled like this:

bash = hs1.get_shell('bash')
    'rm -i some_file', matches=['rm: remove regular file ‘some_file’?']


If you have to handle a shell that performs some action immediately when a certain command is typed (some command that does not need a following return key press), then this argument can be useful because no extra return will be sent after the command:

bash = hs1.get_shell('bash')
bash.send_command('command that needs no following return', newline=False)


Shells will usually have a default timeout value that is used always when a command is executed. One example of such shell is the topology-provided PExpectShell that uses the pexpect package for its implementation. When a command takes more than this default time to execute, a timeout exception will be raised, to avoid this, you can use this argument:

bash = hs1.get_shell('bash')
bash.send_command('command that takes very long to return', timeout=900)

Some examples of get_response

Be aware that send_command returns no output, if a command is sent to the shell in this way, get_response is to be executed after it to get any output that the command execution may have produced:

bash = hs1.get_shell('bash')
bash.send_command('echo "something")
assert bash.get_response() == 'something'

Using multiple shell connections

Nodes have a close relationship with their shells, as mentioned before a Linux host will have one bash shell only, for example. Nevertheless, it may be possible to have more than one connection to the mentioned shell.

The shell API provides a connection argument in its methods to allow the user to specify the shell connection that is to be used to execute the command in the shell.

Shells will have a default connection, which is the connection that is used if connection is not specified (that means, using the default value of None for connection). Of course, before using any additional connection it must be created first. This can be done explicitly by calling the shell connect method, but is usually not necessary, by specifying a new connection in send_command a new connection is created automatically:

bash = hs1.get_shell('bash')

bash.send_command('echo "connection 1", connection='1')
bash.send_command('echo "connection 2", connection='2')

assert bash.get_response(connection='1') == 'connection 1'
assert bash.get_response(connection='2') == 'connection 2'

Managing the default connection

The shell objects have a default_connection attribute that returns the default connection that is set at that moment. By assigning it to other value, the default connection can be changed:

bash = hs1.get_shell('bash')

bash.send_command('echo "connection 0")
assert bash.get_response() == 'connection 0'

bash.send_command('echo "connection 1", connection='1')

assert '0' == bash.default_connection

bash.default_connection = '1'

# Notice how by not specifying the connection argument here, the default
# connection is used
assert bash.get_response() == 'connection 1'

Connecting and disconnecting multiple connections

The API provides the following methods too:

  1. connect
  2. disconnect
  3. is_connected

They are quite self-explanatory, and they all receive the connection argument that specifies on which connection the method operates. Please remember that the default connection is used if no value is set for the connection argument.

Using the topology executable

The topology executable allows to build topologies on demand and interact with them. The topology program allows to launch a topology from a textual description or from a test (see below). The topology program is installed as part of the Topology framework.

$ topology --help
usage: topology [-h] [-v] [--version] [--platform {}] [--non-interactive]
                [--show-build-commands] [--plot-dir PLOT_DIR]
                [--plot-format PLOT_FORMAT] [--nml-dir NML_DIR]
                [--inject INJECT]

Network Topology Framework using NML, with support for pytest.

positional arguments:
  topology              File with the topology description to build

optional arguments:
  -h, --help            show this help message and exit
  -v, --verbose         Increase verbosity level
  --version             show program's version number and exit
  --platform {}         Platform engine to build the topology with
  --non-interactive     Just build the topology and exit
                        Show commands executed in nodes during build
  --plot-dir PLOT_DIR   Directory to auto-plot topologies
  --plot-format PLOT_FORMAT
                        Format for plotting topologies
  --nml-dir NML_DIR     Directory to export topologies as NML XML
  --inject INJECT       Path to an attributes injection file

You can run a topology and interact with their nodes:

$ cat my_topology.szn
# +-------+                                 +-------+
# |       |     +-------+     +-------+     |       |
# |  hs1  <----->  sw1  <----->  sw2  <----->  hs2  |
# |       |     +-------+     +-------+     |       |
# +-------+                                 +-------+

# Nodes
[type=openvswitch name="Switch 1"] sw1
[type=openvswitch name="Switch 2"] sw2
[type=host name="Host 1"] hs1
[type=host name="Host 2"] hs2

# Links
hs1:1 -- sw1:3
sw1:4 -- sw2:3
sw2:4 -- hs2:1

This topology can be run and interacted with like this:

$ topology --platform=docker my_topology.szn
Starting Network Topology Framework v0.1.0
Building topology, please wait...
Engine nodes available for communication:
    sw1, sw2, hs1, hs2
>>> response = hs1('uname -r', shell='bash')
[hs1].send_command(uname -r) ::
>>> response

Using with pytest

The topology framework install a pytest plugin that provides, among other things the topology fixture. This fixture is a module level fixture that will try to find a global variable called TOPOLOGY with the textual description of the topology. If found, it will parse it and build it with the platform engine specified with the new pytest --topology-platform parameter.

Test will be able to interact with the topology nodes by calling topology.get('nodeid') function. When all test are done, the fixture will unbuild the topology automatically. In other words, all tests in a module share the same topology.

# +-------+                                 +-------+
# |       |     +-------+     +-------+     |       |
# |  hs1  <----->  sw1  <----->  sw2  <----->  hs2  |
# |       |     +-------+     +-------+     |       |
# +-------+                                 +-------+

# Nodes
[type=openvswitch name="Switch 1"] sw1
[type=openvswitch name="Switch 2"] sw2
[type=host name="Host 1"] hs1
[type=host name="Host 2"] hs2

# Links
hs1:1 -- sw1:3
sw1:4 -- sw2:3
sw2:4 -- hs2:1

def test_example(topology):
    Example text for topology.
    sw1 = topology.get('sw1')
    sw2 = topology.get('sw2')
    hs1 = topology.get('hs1')
    hs2 = topology.get('hs2')

    assert sw1 is not None
    assert sw2 is not None
    assert hs1 is not None
    assert hs2 is not None

    # Assert that Linux kernel version >= 3
    version = hs1('uname -r').split('.')
    assert version[0] >= 3

Added pytest arguments

Platform Engine to build the topology with.
Path to an attributes injection file, See the Attribute Injection section above.
Directory to auto-plot topologies. All topologies found will be plotted to this directory automatically.
Format for ploting topologies (default 'svg') This option requires Graphviz installed.
Directory to export topologies as NML XML. All topologies found will be exported to NML XML to this directory automatically.

Added pytest markers


Assign a test identifier to the test. The test id will __always__ be added to the jUnit XML report.

For example:

from pytest import mark
def test_example(topology):
platform_incompatible(platforms, reason=None)

Mark a test as incompatible with a list of platform engines. The test will be skipped automatically if it is attempted to be run with an incompatible platform. Optionally specify a reason for better reporting.

For example:

from pytest import marl
@mark.platform_incompatible(['debug'], reason='Not ready yet')
def test_example(topology):