Data scientists launch a Jupyter Notebook servers to tackle each machine learning task. Usually local computers are not enough to handle multiple machine learning tasks. And therefore data scientists do their experiments in servers launched in remote hosts such as EC2 instances.
To connect Jupyter Notebook servers in remote hosts, we use ssh port forwarding. Port forwarding is useful since we do not consume resources in local PC.
Unfortunately, when connecting servers in multiple remote hosts and ports numbers, we easily forget the port number or assign the local port number which is used in another task. Especially when there are multiple remote hosts and ssh servers as the following image, understanding the combinations of remote hosts and local ports are difficult.
pfm manages the remote hosts and port numbers used in port forwarding. Users understand which local ports are used and which ports are not. Once users register the port forwarding information, pfm generates ssh parameters any time specifying the task name.
We can install pfm with pip.
pip install pfm
pfm add
registers port forward settings.
The following is a sample of adding port forward settings with command line options.
$ pfm add -n image-classification --local-port 9999 --remote-port 8888 --ssh-server myml.aws.com --remote-host localhost
pfm add
provides the following options.
Usage: pfm add [OPTIONS] add port forwarding target Options: -n, --name TEXT name of port fowarding --forward-type TEXT port forwarding type [L (local) or R (remote)] --local-port INTEGER local port --remote-port INTEGER remote host port --ssh-server TEXT server to ssh login --server-port INTEGER server port --remote-host TEXT remote host for port forwarding --login-user TEXT login user of ssh server --help Show this message and exit.
Note that when local port number (--local-port
) is not specified, pfm add
automatically assigns the local port not to collide to other port forward settings.
If you think adding many options are tedious, pfm add
also provides argument style of registration.
The following is a sample of registration which forwards port 8888 in takahi-i-ml.aws.com to port 9999 of localhost with a local port forward setting.
$pfm add -n image-classification "9999:localhost:8888 takahi-i-ml.aws.com"
As we see the above example is simple and just uses -n
option. Note that default forward type is set to local (L). If you want to set forward type
to remote please specify the type with --forward_type
parameter.
After the registration of port forward settings with pfm add
, we can generate ssh parameters with pfm param
.
$ ssh `pfm param image-classification`
Or we can also use pfm connect
in order to do same (Note that this command includes -A ssh option).
$ pfm connect image-classification
After the registration of port forward settings with pfm add
, we can modify ssh parameters with pfm update
.
For example, the following command changes the port forwarding type of image-classification setting from local to remote.
$ pfm update --name image-classification --forward-type R
We can see the list of registered port forward settings.
$ pfm list +----------------------+------------+------------+--------------------------------+--------------+-----------------+--------------------------------+--------------+ | name | type | local_port | remote_host | remote_port | login_user | ssh_server | server_port | +======================+============+============+================================+==============+=================+================================+==============+ | image-processing | L | 9999 | localhost | 8888 | None | my-ml-instance.aws.com | | +----------------------+------------+------------+--------------------------------+--------------+-----------------+--------------------------------+--------------+ | text-processing | L | 7777 | localhost | 8888 | None | my-ml-instance-2.aws.com | | +----------------------+------------+------------+--------------------------------+--------------+-----------------+--------------------------------+--------------+
When a port forward settings is not needed, we can remove the setting with pfm delete
command
$ pfm delete image-process
- Free software: MIT license
We are welcome any contribution. For details, please see CONTRIBUTING.rst.