calendar warning only for rest.myria.cs...

[radion]

Should the "Myria cluster is reserved" message show up if you're not connected to rest.myria.cs.washington.edu?

[dhalperi]

please update from master for the fix I put in #243

[dhalperi]

Also note that we want the calendar warning present if the deployment uses any machines in the cluster. The reservation is for the hardware, not the rest.myria.cs.washington.edu deployment.

Not sure how to make that happen; this check is not sufficient.

[domoritz]

Do other deployments usually use the rest.myria.cs.washington.edu url or an ip?

[dhalperi]

.. or vega.cs.washington.edu, or dbserver02.cs.washington.edu, or ...

[domoritz]

Then let's just use a regex for those. I think that if somebody has their own deployment they may submit their queries manually anyway. I think the main use of the calendar is on demo.myria.cs.washington.edu anyway.

[dhalperi]

sure -- if the server name is localhost we can skip the check. Anything else, I'm happy making the developer comment out the line of JS.

This is unrelated to manual queries -- proposed change could only affect users that are using the website, anyway...

---

Daniel Halperin Director of Research for Scalable Data Analytics eScience Institute University of Washington

On Mon, Jan 5, 2015 at 12:50 PM, Dominik Moritz notifications@github.com wrote:

Then let's just use a regex for those. I think that if somebody has their own deployment they may submit their queries manually anyway. I think the main use of the calendar is on demo.myria.cs.washington.edu anyway.

— Reply to this email directly or view it on GitHub https://github.com/uwescience/myria-web/pull/242#issuecomment-68778813.

[radion]

So do we want to only filter out just localhost, regex the list of machines, use rest.myria..., or leave the code as it was? I'll update the code with the changes we want here.

[dhalperi]

Here's my calculus:

false positive: fraudulent warning, and to disable warning, developer has to figure out the right JS and disable

false negative: warning where there should be one. cluster reservations may be inadvertently violated through myria-web.

To me, cost of false negative is MUCH higher, false positive is minor inconvenience at most

• current state of the world: developer using a private deployment sees a calendar warning anyway
  • Downside: no true negatives
  • Upside: no false negatives
• Removing the check except for 'rest.cs.washington.edu' exactly:
  • Upside: no false positives
  • Downside: many false negatives
• Removing the check except for regex on *.cs.washington.edu:
  • similar to previous warning, does not solve either the false positive or false negative problem
  • false positive: my machine is dhm.dyn.cs.washington.edu...
  • false negative: e.g., deployment hosted from sol, not *.cs.washington.edu
• Adding an exception for specifically localhost
  • upside: most common "innocent dev / outsider dev" use case is probably people modifying deployment.cfg.local, so this would fix false positive in majority of cases.
  • downside: still may be some false positives
  • downside: false negatives, I guess, if someone ran myria-web from a cluster node.

I feel like the last one is optimal, but would hear other opinions.

[radion]

I agree, for now, we should go with the localhost option. That's the main reason I thought of this change.

The ideal solution, I think, is to create a set of reserved worker machines that the code checks against. But I don't know if that's a possibility.