A good start to compile a list of all these custom solutions and requests to show that the community really needs this.
- pbs-workbench (this tool)
- The previous center of excellence for extreme weather used to have this tool.
- Discussion here, including manual steps.
- Requests from researchers:
I believe @sam.green creates an interactive job and sshs into it. We could gather “testimonials” of other researchers’ custom solutions.
Secondly , describe exactly what is missing from the ARE workflow. Some ideas are:
- IDEs limited to RStudio and jupyter notebooks.
- Virtual desktop is very slow compared with native/local GUIs.
- Harcoded modules loaded that cannot be changed after spinup (that is true for rstudio and jupyter, not sure about the VDI).
- If in the process of data analysis you realise you need a new module, you need to close everything, add the modules and spin the whole thing up again.
- This makes it hard to run the same code interactively in ARE and via terminal or in jobs in the queue.
- Relies on external service (what happens if you ARE is down?)
Third, specify what “endpoints” would we need.
- Start a workbench with particular resources/profile for a particular ammount of time
- List running workbenches.
- Get information from a workbench:
- Resources
- time spent
- time left
- node address
- SU cost?
- … Something else?
- Stop a workbench
- Save “profiles” for particular configurations (similar to ARE ability to save settings)
- Modify existing profiles.
- …Something else?
For troubleshooting, it might be useful to get easy access to the logs.