Juliabox currently provides the older versions of DataFrames (0.10.x) and related packages (prior to the Nullables -> Missings change). We are now ready to upgrade the entire set of built in packages to use the latest versions of old packages.
While we kept the old version of packages to ensure some stability, and allow the ecosystem to synchronise, we are now in a situation where almost all dependencies have moved to the new DataFrames. This has caused problems in installing new packages on Juliabox. As a result, we feel this is now the right time to make this change.
This upgrade will occur around midnight EST, Monday night, Tuesday morning. There will be a small downtime of upto 10 minutes, and any existing sessions might be reset.
This change might be slightly disruptive for some users. We would urge you to test your notebooks if you depend on DataFrames or Nullables. Also, please reset any packages you have installed yourself on Juliabox, otherwise you will see Kernel Died messages.
Paid customers who use custom images will continue to use those, and thus will see no changes.
I think a change like that needs way more time between an announcement and the actual switch, 24 hours seems very, very short.
This doesn’t affect me right now, but if I was teaching my course right now (which I did until two weeks ago), this kind of thing would have caused real chaos. It also makes me wonder whether I can use JuliaBox for teaching in the future, because I certainly would need something that is more predictable in terms of changes.
Maybe a good model would be something like a designated update day (the first Mon every month), and then say two weeks before that you announce what packages will be updated?
I understand that this can be disruptive, so this is not something we do without due consideration. We are pulled from many directions in this regard - from necessary bugfixes, to requests for additional packages. For Dataframes, we have had many, increasingly urgent, requests over the months for an upgrade.
I do take onboard your feedback on notice. But I’ll note that there is no way we can keep things absolutely stable over a 3-4 month class. Even with a designated update day, you might have many such disruptions during course of one class.
For classes that want an absolutely stable environment, we do build custom environments, but that takes some effort to implement. We already spend significant resources in keeping juliabox running, so you’ll appreciate that we cannot provide that particular service for free.