wa_infra_tools

This package contains tooling for Wisconsin Autonomous infra.

Motivation and Goals

This package is meant to house any tools to assist in development for subteams. The initial motivation was to create a package to help manage data for the perception team. We've accomplished this by establishing a database on the CAE network and providing a python api (wa_infra_tools.database.PostgresDatabase) to interface with the database, allowing basic commands such as directly executing sql, inserting rows with images and downloading data.

In the future, we aim to continue improving on the database api by improving performance and adding more functionality (how do we sample data, asynchronous sending of files, etc). We also want to add a package to help the perception team optimize their GPU-trained models to be run on Intel CPUs using OpenVino. Finally, want to create an interface for recording and managing rosbags.

How To Use

Install

You can install the package locally by running pip install . in the root directory.

Components and Examples

• wa-mars CLI
• wa-git CLI
• PostgresDatabase
• SSHUtils

wa-mars

wa-mars is a CLI that allows the user to easily interface with our Model and Rosbag Storage (MARS) repository. Run wa-mars -h for more info on how to use it.

wa-git

wa-mars is a CLI that wraps git and prevents the user from downloading LFS files (only pointers). Run wa-git -h for more info on how to use it.

PostgresDatabase

PostgresDatabase interfaces with the database in the CAE storage.

from wa_infra_tools.database import PostgresDatabase

# create a connection to the remote database
db = PostgresDatabase('your_cae_username')

# view what tables are on the database
tables = db.tables

# get schema of a table
schema = db.get_schema('table_name')

# preview table
records = db.preview_table('table_name')

# execute a sql command and get the results
# NOTE: if you execute a mutation (insertion, creation, deletion, update) it will not be committed to the database until you call db.commit()
records = db.sql("some sql string")

# insert a row into a table
db.insert_row('table_name', {'col_name': col_val, ...})

# insert a row that contains an image
db.insert_row_with_image('table_name', {'col_name': col_val, ..., 'filepath': local_filepath_to_image})

# download data from a table (assumed you are downloading data in image/label pairs)
db.download_data('table_name', ['label_col_name_1', 'label_col_name_2', ...], 'output_path', 'format_type (only yolo so far)')

# if you want to join two tables and download the result (as it is common that images/labels are stored separately)
db.join_and_download_data('images_table_name', 'labels_table_name', ['label_col_name_1', 'label_col_name_2', ...], 'output_path', 'format_type (only yolo so far)')

SSHClient

SSHClient can create and manage ssh tunnels. It was made to support PostgresDatabase but can also be used independently.

from wa_infra_tools.ssh_utils import SSHClient

# create connection to a host
client = SSHClient('username', 'hostname')

# execute command via SSH on the remote machine
stdin, stdout, stderr = client.exec_command('command')

# get SFTP client to transfer  files
sftp = client.get_sftp_client()

# start a tunnel
client.start_tunnel(local_port, remote_host, remote_port)

# stop a tunnel
client.stop_tunnel(local_port, remote_host, remote_port)

# remove a tunnel
client.remove_tunnel(local_port, remote_host, remote_port)

# start all tunnels
client.start_tunnels()

# stop all tunnels
client.stop_tunnels()

# see what tunnels exist
tunnels = client.tunnels

Starting the Database

Since we don't currently have sudo access on the CAE servers, we have to start the database on each machine manually. If the machine restarts, we also have to restart the database manually. Currently, we are using tux-144.cae.wisc.edu to host our database. I have written a script under scripts/start_database.py to start the database. You can run it with python3 scripts/start_database.py your_username --hostname optional_hostname_if_needed.