/quic-interop-runner

QUIC interop runner

Primary LanguagePythonOtherNOASSERTION

Interop Test Runner

The Interop Test Runner aims to automatically generate an interop matrix by running multiple test cases using different QUIC implementations.

Requirements

The Interop Runner is written in Python 3. You'll need to install the following softwares to run the interop test:

  • Python3 modules. Run the following command:
poetry install

Running the Interop Runner

Run the interop tests:

./run.py

IPv6 support

To enable IPv6 support for the simulator on Linux, the ip6table_filter kernel module needs to be loaded on the host. If it isn't loaded on your machine, you'll need to run sudo modprobe ip6table_filter.

Building a QUIC endpoint

To include your QUIC implementation in the Interop Runner, create a Docker image following the instructions for setting up an endpoint in the quic-network-simulator, publish it on Docker Hub and add it to implementations.json. Once your implementation is ready to interop, please send us a PR with this addition. Read on for more instructions on what to do within the Docker image.

Typically, a test case will require a server to serve files from a directory, and a client to download files. Different test cases will specify the behavior to be tested. For example, the Retry test case expects the server to use a Retry before accepting the connection from the client. All configuration information from the test framework to your implementation is fed into the Docker image using environment variables. The test case is passed into your Docker container using the TESTCASE environment variable. If your implementation doesn't support a test case, it MUST exit with status code 127. This will allow us to add new test cases in the future, and correctly report test failures und successes, even if some implementations have not yet implemented support for this new test case.

The Interop Runner mounts the directory /www into your server Docker container. This directory will contain one or more randomly generated files. Your server implementation is expected to run on port 443 and serve files from this directory. Equivalently, the Interop Runner mounts /downloads into your client Docker container. The directory is initially empty, and your client implementation is expected to store downloaded files into this directory. The URLs of the files to download are passed to the client using the environment variable REQUESTS, which contains one or more URLs, separated by a space.

After the transfer is completed, the client container is expected to exit with exit status 0. If an error occurred during the transfer, the client is expected to exit with exit status 1. After completion of the test case, the Interop Runner will verify that the client downloaded the files it was expected to transfer, and that the file contents match. Additionally, for certain test cases, the Interop Runner will use the pcap of the transfer to verify that the implementations fulfilled the requirements of the test (for example, for the Retry test case, the pcap should show that a Retry packet was sent, and that the client used the Token provided in that packet).

The Interop Runner generates a key and a certificate chain and mounts it into /certs. The server needs to load its private key from priv.key, and the certificate chain from cert.pem.

Examples

If you're not familiar with Docker, it might be helpful to have a look at the Dockerfiles and scripts that other implementations use:

Implementers: Please feel free to add links to your implementation here!

Logs

To facilitate debugging, the Interop Runner saves the log files to the logs directory. This directory is overwritten every time the Interop Runner is executed.

The log files are saved to a directory named #server_#client/#testcase. output.txt contains the console output of the interop test runner (which might contain information why a test case failed). The server and client logs are saved in the server and client directory, respectively. The sim directory contains pcaps recorded by the simulator.

If implementations wish to export the TLS secrets, they are encouraged to do so in the format in the NSS Key Log format. The interop runner sets the SSLKEYLOGFILE environment variable to a file in the logs directory. In the future, the interop runner might use those files to decode the traces.

Implementations that implement qlog should export the log files to the directory specified by the QLOGDIR environment variable.

Test cases

The Interop Runner implements the following test cases. Unless noted otherwise, test cases use HTTP/0.9 for file transfers. More test cases will be added in the future, to test more protocol features. The name in parentheses is the value of the TESTCASE environment variable passed into your Docker container.

  • Version Negotiation (versionnegotiation): Tests that a server sends a Version Negotiation packet in response to an unknown QUIC version number. The client should start a connection using an unsupported version number (it can use a reserved version number to do so), and should abort the connection attempt when receiving the Version Negotiation packet. Currently disabled due to #20.

  • Handshake (handshake): Tests the successful completion of the handshake. The client is expected to establish a single QUIC connection to the server and download one or multiple small files. Servers should not send a Retry packet in this test case.

  • Transfer (transfer): Tests both flow control and stream multiplexing. The client should use small initial flow control windows for both stream- and connection-level flow control, such that the during the transfer of files on the order of 1 MB the flow control window needs to be increased. The client is expected to establish a single QUIC connection, and use multiple streams to concurrently download the files.

  • ChaCha20 (chacha20): In this test, client and server are expected to offer only ChaCha20 as a ciphersuite. The client then downloads the files.

  • KeyUpdate (keyupdate, only for the client): The client is expected to make sure that a key update happens early in the connection (during the first MB transferred). It doesn't matter which peer actually initiated the update.

  • Retry (retry): Tests that the server can generate a Retry, and that the client can act upon it (i.e. use the Token provided in the Retry packet in the Initial packet).

  • Resumption (resumption): Tests QUIC session resumption (without 0-RTT). The client is expected to establish a connection and download the first file. The server is expected to provide the client with a session ticket that allows it to resume the connection. After downloading the first file, the client has to close the connection, establish a resumed connection using the session ticket, and use this connection to download the remaining file(s).

  • 0-RTT (zerortt): Tests QUIC 0-RTT. The client is expected to establish a connection and download the first file. The server is expected to provide the client with a session ticket that allows it establish a 0-RTT connection on the next connection attempt. After downloading the first file, the client has to close the connection, establish and request the remaining file(s) in 0-RTT.

  • HTTP3 (http3): Tests a simple HTTP/3 connection. The client is expected to download multiple files using HTTP/3. Files should be requested and transfered in parallel.

  • Handshake Loss (multiconnect): Tests resilience of the handshake to high loss. The client is expected to establish multiple connections, sequential or in parallel, and use each connection to download a single file.

  • V2 (v2): In this test, client starts connecting server in QUIC v1 with version_information transport parameter that includes QUIC v2 (0x6b3343cf) in other_versions field. Server should select QUIC v2 in compatible version negotiation. Client is expected to download one small file in QUIC v2.

Measurements

See descriptions in ./testcases.py.

Real Links

For measurements with real links instead of ns-3 links between client and server, the client is deployed on a different machine as the server. The IP addresses are determined automatically (which is sometimes somewhat hacky and maybe broken). Ensure, that the default route of the client PC is configured to use the link that should be used for the measurement. For SL, AST, and EUT the SSH connection link can be configured in config.yml.

Plotting

After running the measurements, use

./post_process_logs.py <path/to/result.json>

to prepare the logs.

After that, use

./plot_all.py -j $(($(nproc) - 1)) --mode='offset-number' <path/to/result.json>

to plot the offset plots for all measurements.

Interactive help and more description should be available for all scripts using the --help flag.