Only
/home/jovyan/workis kept. It's theworkfolder in your home directory
(~/work), and the file browser opens there. Everything else is erased each time your
server stops: the rest of your home directory,/tmp, and any packages you installed.
| Location | Kept when the server stops? | Disk limit |
|---|---|---|
~/work (/home/jovyan/work) |
Yes | 20 GB, and it can be raised |
Everything else: ~ outside work, /tmp, ~/.cache, ~/.local, installed packages |
No | 12.5 GB on a CPU node, 10 GB on a GPU node. It can't be raised |
work has a 20 GB quota. It belongs to you, not to a server, so it's the same on CPU
and GPU nodes and it carries over from one session to the next. If you need more,
ask.
Everything outside work has a fixed limit: 12.5 GB on a CPU node and 10 GB on a
GPU node. It covers all files your server writes outside work, and it starts at zero
each time the server starts. It can't be raised, because each node's disk is shared by
every notebook running on it.
work is fullwork fails, usually withwork (see Course folders). When there's noDelete files you no longer need, or ask for more space. If the
server won't start, ask first: you need a running server to delete files.
Once your server is past its limit outside work, ChemCompute stops it within a few
minutes and emails you:
Your Jupyter Notebook was stopped because it was over disk quota.
Stopping the server erases everything outside work, along with any notebook changes you
hadn't saved. Your files in work are safe. Start the server again, and move whatever
filled the disk into work (see the next section).
Psi4 writes its temporary files to /tmp. A large calculation can write several GB there,
and /tmp is outside work. To put them in work instead, add this after import psi4:
import os
os.makedirs("/home/jovyan/work/scratch", exist_ok=True)
psi4.core.IOManager.shared_object().set_default_path("/home/jovyan/work/scratch")
Those files then count against the 20 GB in work, so empty work/scratch when you're
done. Calculations too big for a notebook's 4 cores and 6 GB of memory usually run better
as a ChemCompute job.
Trajectories, cube files and downloaded data belong in work. When a program writes to the
folder it was started in, or to /tmp, point it at a folder in work.
pip install, conda install and mamba install write outside work, and so do their
download caches in ~/.cache. Big packages fill the disk quickly, and they have to be
installed again after every restart anyway. If you need a package every time,
ask for it to be added to the images.
Open a Terminal (New → Terminal in Notebook, or the Terminal tile in Lab's
launcher):
du -sh ~/work # all of work
du -sh ~/work/* | sort -h # each folder in work, largest last
du -sh ~/.cache ~/.local /tmp 2>/dev/null # the usual places outside work
For work: open a support ticket or use the
Contact page. Say how much you need and what for.
A raised quota stays raised.
Outside work: the limit can't be raised. Move scratch and output files into work,
or stop and restart your server, which empties everything outside work and resets the
count to zero.
workEach time your server starts, ChemCompute puts course folders in work. They count against
its 20 GB.
experiments and datasets are kept up to date with the originals. Their notebookswork, and keeppsi4education and SciCompforChemists get new files added, but files you'veDownload anything important at the end of a lab. To download a whole folder, make an
archive of it in a Terminal:
cd ~/work
tar czf myproject.tar.gz myproject
Then select myproject.tar.gz in the file browser and choose Download. Delete the
archive afterwards, since it counts against your quota too.
See also Notebook problems.