operations

Approving TIaaS Requests

1. Go to the unprocessed trainings.
2. Click on one training: this brings you to the edit-tiaas-request page.
3. Estimate the resources:
   • Get the individual resources for each tool needed in the training. If a tool is not on the list, you can assume {mem: 4, cores: 1}.
   • Get the max_resources based on the max mem and max cores for all the requested tools.
   • estimated_resources = max_resources * (#users + 2)
4. Change the status of the request: at the bottom, change Processed to Approved.
5. Save the changes.
6. Script for the resource allocation:
   • Clone https://github.com/usegalaxy-eu/vgcn-infrastructure
   • Go into vgcn-infrastructure, and run ./add-training.sh with the left parameters:

    Usage:
      ./add-training.sh <training-identifier> <vm-size (e.g. c1.c28m225d50)> <vm-count> <start in YYYY-mm-dd> <end in YYYY-mm-dd> [-- donotautocommitpush]
   

   • The script will add an entry to resources.yaml with the information for the training.
   • If the option donotautocommitpush was not used, the script will commit and push.
   • A template email will be printed as a result of the script.
   • NOTE: There are 9 training nodes. If two trainings overlap, the number of total nodes shouldn’t exceed 9.
7. Check in the calendar that the training has been booked.
8. Send the email to the user.

What happens in the background

• The next time the galaxy playbook runs, it deploys the tiaas training. Then the URL becomes active.
• Whenever todays_date >= start date, vgcn-infrastructure will automatically try and launch the VM.

---

FAQ

Question	Answer
What if there are multiple trainings?	It has not happened yet. If it does, it is with long-running trainings and we usually give them fewer / smaller machines.
What is the recommended machine?	c1.c28m225d50, generally used for training, so no changes to the main queue needed.
What is the format for the date?	YYYY-mm-dd
How many machines can be assigned?	There is a maximum of 7 machines.
How do I estimate the resources accurately?	It’s hard to estimate the number correctly without more information that isn’t so easy to collect. Usually there’s no information about the time limitations they have, maybe they expect a step to run in 1 minute, but we don’t know that. We don’t know either if they’ll run a dataset collection.