Will Sorenson Machine Learning and Coding

Making working on Remote Boxes a Breeze

Up until recently, I rarely used the beefy workstation that I built out of frustration with the performance of my laptops’ performance on intensive tasks. This all changed last week and the catalyst was a tool called Mosh. Mosh is an alternative to ssh that has all the features I always wished it had. With mosh, there are no more broken pipes and no more fighting with tenuous internet connections. Here’s what my setup looks like now:

If you would normally type ssh do to ssh into your remote computer, instead you just do mosh do. It’s that simple! It uses .ssh/config that you already probably use for ssh. For me the main benefits of mosh are (1) decreased latency and (2) perpetual sessions. I can leave one of my laptops off for days and the moment I turn it back on, my remote tmux session is still there. Mosh allows me work on remote from one laptop during the day and then seamlessly pick up what I was working on in the evening from another laptop1.

The magic happens when I run a tool called ngrok from my mosh session. ngrok lets me expose and navigate to a port on a remote machine from anywhere without messing with firewalls or anything. If I’m using my workstation to develop a website, I’d like to be able to connect to that website locally. By the same token, if I’m wearing my Data Scientist hat, I’d like to be able to access Jupyter Lab. Ngrok makes both of these things possible. Here’s what it looks like:

Here, I’m forwarding my Jupyter Lab server to ngrok, which I can then access from any of my laptops:

ngrok has a bunch of security features to make this quite secure as the attack service is now quite small2.

Thanks to ngrok and mosh I can now work on the terminal and localhost on remote as if everything were locally hosted. That’s great but there’s still one problem – how can I work in an IDE like PyCharm on remote?3 The only seamless solution I could think of is keeping files syncronized in Dropbox. So far, this has been working fine for me. I use the IDE locally and then, within a minute of making the change, the change is propagated to remote. In the terminal I can then start the website/script/whatever. I’m still experimenting with this setup and it feels like this Dropbox syncronization is going to burn me one day.

So with all this, working on a powerful remote workstation is very nearly as convenient as working on my 12 inch, 2 pound laptop.

Footnotes

  1. Before I had to ssh into my cloud vps session and then type in another ssh command from there to get into my remote workscreen. Once I was there, typing on the terminal would respond rather slowly as my typing always had to make a round trip before showing up on my screen. ↩

  2. My VPS is completely locked down except for an ssh port being open to the world. For an intruder to get terminal access, they’d either have to find a 0 day in ssh or an exploit in both ngrok and whatever app I am serving using ngrok. My VPS also has tools built in to detect intrusions. Since I am using a reverse tunnel from my workstation to my VPS, it’s not possible to log in directly to my workstation. In the case of Jupyter Lab, I have two layers of authentication set up since compromising that allows shell access. ↩

  3. With pycharm in particular, there’s a better solution if you connect directly to remote without using an intermediary machine. PyCharm has the functionality to use remote python interpreters. See here for more details. ↩

Written on July 23rd , 2018 by Will Sorenson
Feel free to share!

You may also enjoy: