Server connection errors mean binder sessions lasting for <= 10 minutes

I’m using the binder image at:

it starts up and runs fine - but after 5-10 minutes I start getting serve connection errors - not sure why, any advice much appreciated…



On, we cull images that have been idle for 5-10 mins in order to not waste money on unused pods/create resources for other users that may be waiting to start.

This happens on active sessions that I am currently using - it started happening a couple of weeks ago…

Are you running all the notebooks? Maybe you max out the memory? Each session only gets 2GB.

I ran your first 3 notebooks (00, 01, 02) and got close to the RAM limit, then started seeing the server connection error.

Screenshot 2021-12-06 at 14.46.59

1 Like

This happens even if I just run a single notebook, and then do some simple calculations after - here’s a screenshot showing that I seem to have spare memory - I’m typically getting 2-4 minutes before losing the connection at the moment…

I see it with that link you posted, too. Definitely wasn’t 10 minutes. I was running that notebook #10 and like the fourth pass, I saw what you report with the timeout. And I was just using another MyBinder-served environment a bunch with no such issues.
Bad node, perhaps?

Does it persist if you trigger an update and Binder rebuilds the image?

I suggest bad node because sometimes I’ve been able to bypass this same behavior by launching from a different Binder partner. It happened once to me in like the last month and in that case it was super non-demanding computationally. So definitely not common, if it is the same thing.

I think @sgibson91 is right, it’s probably a lack of memory.

Note that since the nodes are shared with multiple users you may not be able to use the full 2GB quota. You should also be aware that some mybinder federation members have a larger memory quota, but in general you can’t rely on this.


Thanks for everyone’s advice! Yes - looks like I’m running out of memory, I’ll see what I can do…