alpy 2.0.0

Library for testing network virtual appliances using Docker

Test network virtual appliance using Docker containers

Contents

• General information

  • The project

  • Author

  • License

  • Description

• Examples

• Features

  • The simplest docker to QEMU networking connection

  • Reliable packet capture

  • First-class Docker container support

  • Logging

  • No trash left behind

  • No root required

• API documentation

• Network design

• Building a network of nodes

• FAQ

  • How do I watch serial console?

  • How do I watch traffic on an interface?

  • Can I use Wireshark to watch traffic on an interface?

  • How do I debug my program?

  • How do I enter node network namespace?

• A note about GitLab Container Registry

• Related projects

General information

The project

This project is a Python library for testing network virtual appliances.

Author

Alexey Bogdanenko

License

Alpy is licensed under SPDX-License-Identifier: GPL-3.0-or-later. See COPYING for more details.

Description

Alpy manages containers via Docker Python API.

Alpy interacts with QEMU using Python API of the QEMU Monitor Protocol (QMP). QMP is a JSON-based protocol that allows applications to communicate with a QEMU instance.

Alpy gives user Pexpect object to interact with a serial console. The Pexpect object is configured to log console input and output via the standard logging module.

Alpy is packaged and deployed to PyPI. The package can be installed using pip.

There are unit tests (pytest) and integration tests in GitLab CI pipeline. Alpy is tested and works on the latest Ubuntu and the latest Ubuntu LTS release.

Examples

The alpy library repository includes scripts and modules to build a simple appliance called Rabbit. Rabbit is Alpine Linux with a few packages pre-installed. Having this simple DUT allows to demonstrate the library features and capabilities. The tests verify a few features of the network appliance, for example:

• IPv4 routing (see rabbit/tests/forward-ipv4/main.py)

• rate-limiting network traffic (see rabbit/tests/rate-limit/main.py)

• load-balancing HTTP requests (see rabbit/tests/load-balancing/main.py)

The tests are executed automatically in the GitLab CI pipeline.

Example network (test rate-limit):

+-------------------------------------+
|                                     |
|          Device under test          |
|          rate limit = 1mbps         |
+-------+--------------------+--------+
        |                    |
        |                    |
        |                    |
+-------+--------+   +-------+--------+
|                |   |                |
| 192.168.1.1/24 |   | 192.168.1.2/24 |
|                |   |                |
| node0          |   | node1          |
| iperf3 client  |   | iperf3 server  |
+----------------+   +----------------+

Example test output:

INFO     __main__               Test description: Check that rabbit rate-limits traffic.
INFO     alpy.node              Create tap interfaces...
INFO     alpy.node              Create tap interfaces... done
INFO     alpy.qemu              Initialize QMP monitor...
INFO     alpy.qemu              Initialize QMP monitor... done
INFO     alpy.qemu              Start QEMU...
INFO     alpy.qemu              Start QEMU... done
INFO     alpy.qemu              Accept connection from QEMU to QMP monitor...
INFO     alpy.qemu              Accept connection from QEMU to QMP monitor... done
INFO     alpy.node              Create nodes...
INFO     alpy.node              Create nodes... done
INFO     alpy.console           Connect to console...
INFO     alpy.console           Connect to console... done
INFO     alpy.utils             Enter test environment
INFO     __main__               Start iperf3 server on node 1...
INFO     __main__               Start iperf3 server on node 1... done
INFO     alpy.qemu              Start virtual CPU...
INFO     alpy.qemu              Start virtual CPU... done
INFO     alpine                 Wait for the system to boot...
INFO     alpine                 Wait for the system to boot... done
INFO     alpine                 Login to the system...
INFO     alpine                 Login to the system... done
INFO     alpy.remote_shell      Type in script configure-rabbit...
INFO     alpy.remote_shell      Type in script configure-rabbit... done
INFO     alpy.remote_shell      Run script configure-rabbit...
INFO     alpy.remote_shell      Run script configure-rabbit... done
INFO     __main__               Start iperf3 client on node 0...
INFO     __main__               Measure rate...
INFO     __main__               Measure rate... done
INFO     __main__               Parse iperf3 report...
INFO     __main__               Parse iperf3 report... done
INFO     __main__               Start iperf3 client on node 0... done
INFO     alpine                 Initiate system shutdown...
INFO     alpine                 Initiate system shutdown... done
INFO     alpy.qemu              Wait until the VM is powered down...
INFO     alpy.qemu              Wait until the VM is powered down... done
INFO     alpy.qemu              Wait until the VM is stopped...
INFO     alpy.qemu              Wait until the VM is stopped... done
INFO     __main__               Rate received, bits per second: 976321
INFO     __main__               Check rate...
INFO     __main__               Check rate... done
INFO     alpy.utils             Exit test environment with success
INFO     alpy.console           Close console...
INFO     alpy.console           Close console... done
INFO     alpy.qemu              Quit QEMU...
INFO     alpy.qemu              Quit QEMU... done
INFO     alpy.utils             Test passed

The tests for the Rabbit device share a lot of code so the code is organized as a library. The library is called carrot.

Features

The simplest docker to QEMU networking connection

Nothing in the middle. No bridges, no veth pairs, no NAT etc.

Each layer 2 frame emitted is delivered unmodified, reliably.

Reliable packet capture

Each frame is captured reliably thanks to the QEMU filter-dump feature.

First-class Docker container support

Alpy follows and encourages single process per container design.

Logging

Test logs are easy to configure and customize. Alpy consistently uses Python logging module.

Alpy collects serial console log in binary as well as text (escaped) form.

No trash left behind

Alpy cleans up after itself:

• processes stopped with error codes and logs collected,

• files, directories unmounted,

• temporary files removed,

• sockets closed,

• interfaces removed…

… reliably.

No root required

Run as a regular user.

API documentation

The documentation is published on GitLab Pages of your GitLab project (if GitLab Pages is enabled on your GitLab instance). For example, upstream project documentation lives at https://abogdanenko.gitlab.io/alpy.

Alpy API documentation is generated using Sphinx. To generate HTML API documentation locally, install Sphinx package and run the following command:

PYTHONPATH=. sphinx-build docs public

To view the generated documentation, open public/index.html in a browser.

Network design

The appliance being tested is referred to as a device under test or DUT.

The DUT communicates with containers attached to each of its network links.

Guest network adapters are connected to the host via tap devices (Figure 1):

+-----QEMU hypervisor------+
|                          |   +-------------+
| +-----Guest OS-----+     |   |             |
| |                  |     |   |  docker     |
| | +--------------+ |     |   |  container  |
| | |              | |     |   |  network    |
| | |  NIC driver  | |     |   |  namespace  |
| | |              | |     |   |             |
| +------------------+     |   |   +-----+   |
|   |              |       |   |   |     |   |
|   | NIC hardware +---+-----------+ tap |   |
|   |              |   |   |   |   |     |   |
|   +--------------+   |   |   |   +-----+   |
|                      |   |   |             |
+--------------------------+   +-------------+
                       |
                       |
                       v
                 +-----------+
                 |           |
                 | pcap file |
                 |           |
                 +-----------+

Figure 1. Network link between QEMU guest and a docker container.

Each tap device lives in its network namespace. This namespace belongs to a dedicated container - a node. The node’s purpose is to keep the namespace alive during the lifetime of a test.

For an application to be able to communicate with the DUT the application is containerized. The application container must be created in a special way: it must share network namespace with one of the nodes.

Figure 2 shows an example where application containers app0 and app1 share network namespace with node container node0. Application container app2 shares another network namespace with node2.

This sharing is supported by Docker. All we have to do is to create the application container with the --network=container:NODE_NAME Docker option. For example, if we want to send traffic to the DUT via its first link, we create a traffic generator container with Docker option --network=container:node0.

+----QEMU---+   +------shared network namespace-----+
|           |   |                                   |
|           |   |    eth0                           |
|   +---+   |   |   +---+   +-----+ +----+ +----+   |
|   |NIC+-----------+tap|   |node0| |app0| |app1|   |
|   +---+   |   |   +---+   +-----+ +----+ +----+   |
|           |   |                                   |
|           |   +-----------------------------------+
|           |
|           |
|           |
|           |   +------shared network namespace-----+
|           |   |                                   |
|           |   |    eth0                           |
|   +---+   |   |   +---+   +-----+                 |
|   |NIC+-----------+tap|   |node1|                 |
|   +---+   |   |   +---+   +-----+                 |
|           |   |                                   |
|           |   +-----------------------------------+
|           |
|           |
|           |
|           |   +------shared network namespace-----+
|           |   |                                   |
|           |   |    eth0                           |
|   +---+   |   |   +---+   +-----+ +----+          |
|   |NIC+-----------+tap|   |node2| |app2|          |
|   +---+   |   |   +---+   +-----+ +----+          |
|           |   |                                   |
+-----------+   +-----------------------------------+

Figure 2. Application containers attached to the DUT links.

Building a network of nodes

Network configuration operations are performed by temporary one-off Docker containers by calling ip commands inside the containers.

A distinction is made between a simplified version of the ip binary and the full version. The simplified version is a busybox applet. The full version is shipped in the iproute2 package.

Here is a list of features which alpy requires but which are missing from the simplified version:

1. Move a network interface to a different namespace (“ip link set netns …”)

2. Create a tap interface (“ip tuntap add mode tap …”)

The image which contains the simplified version is called busybox_image while the full image is called iproute2_image.

The images must be provided by the caller and must be present on the system. For example, set:

busybox_image = "busybox:latest"
iproute2_image = "registry.gitlab.com/abogdanenko/alpy/iproute2:latest"

FAQ

How do I watch serial console?

Use tail:

tail --follow name --retry console.log

The same command, but shorter:

tail -F console.log

How do I watch traffic on an interface?

Use tcpdump:

tail --bytes +0 --follow name --retry link0.pcap | tcpdump -n -r -

The same command, but shorter:

tail -Fc +0 link0.pcap | tcpdump -nr-

Can I use Wireshark to watch traffic on an interface?

Yes, you can:

tail --bytes +0 --follow name --retry link0.pcap | wireshark -k -i -

The same command, but shorter:

tail -Fc +0 link0.pcap | wireshark -ki-

How do I debug my program?

Use The Python Debugger.

How do I enter node network namespace?

1. Get node pid:

   docker inspect --format '{{.State.Pid}}' node0

2. Jump into node namespace using that pid:

   nsenter --net --target "$pid"

One-liner:

nsenter --net --target "$(docker inspect --format '{{.State.Pid}}' node0)"

A note about GitLab Container Registry

Many CI jobs use one of the custom images built on the “build-docker-images” stage. The images are stored in the GitLab Container Registry.

The images are pulled from locations specified by GitLab variables. By default, the variables point to the registry of the current GitLab project.

If you forked this project and GitLab Container Registry is disabled in your project, override the variables on a project level so that the images are pulled from some other registry.

For example, set IMAGE_UBUNTU_LTS=registry.gitlab.com/abogdanenko/alpy/ubuntu-lts:latest.

Related projects

• Containernet

• Kathara

• Netkit

• GNS3

• Virtual Networks over linuX (VNX)

• Pipework: Software-Defined Networking for Linux Containers

• Eve-NG