Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
  • Apr 08 2019 14:17
    that wasn't a real error, just palying
  • Apr 08 2019 14:16
    oops
  • Sep 25 2018 09:50
    bot test
Min RK
@minrk
not that it's out of the question, but mainly: are you sure you want to use Binder for this?
Tim Head
@betatim
i liked your point about "the more we tighten the automatic cut off the more we will run into catching out innocent people"
but if we are Ok with automatic killing for stuff that is running for 9.5min out of 10minutes or so then lets do it at the process level no?
or even pod level
Min RK
@minrk
yeah, it's really tough. I suspect our 1 cpu-hour cutoff has mostly terminated miners we haven't identified yet, plus some students who left a busy kernel and moved on.
Tim Head
@betatim
then we don't need to tell the difference between a student leaving their for loop running and a miner
Min RK
@minrk
I think some non-Python kernels are also more cpu-intensive
yeah, I can't remember where minesweeper is right now on being able to stop processes
I think at least we don't have a 'should terminate' state differentiation for pod vs process
Tim Head
@betatim
https://github.com/jupyterhub/mybinder.org-deploy/runs/3671926155 is that because there is a closing parentheses missing on config_files.extend(sorted(glob.glob(os.path.join("secrets", "config", "common", "*.yaml"))) ?
squintingovermycoffeemug
Min RK
@minrk
I guess we should be running at least basic syntax checks on our scripts...
Tim Head
@betatim
not good enough to spot it before merging
Min RK
@minrk
it is 4 closing parentheses in a row
Tim Head
@betatim
would black have found it? then we get the benefit of formatting and basic syntax'ing
Min RK
@minrk
yeah, black will fail on any actually invalid files
Just about any check would have caught that one
Tim Head
@betatim
not the biggest of problems either
Min RK
@minrk
yeah. No real reason not to add pre-commit like we have everywhere else. Seems to be helping more than hurting
Tim Head
@betatim
by the way, i love the /test-this-pr :) it took me a moment to realise that sarah's comments were from a script and not human sarah
Min RK
@minrk
It's wonderful! So pleased. :cake: for @sgibson91
More static analysis is probably good on a repo like this one where we can't run a whole lot of tests on PRs before deploying them.
Tim Head
@betatim
good thing we see her in the team calls. otherwise i'd be starting to think she has it all automated and is living a life of free time while helping all these people on github
https://pydata.org/global2021/ - has someone been to a previous version of this? I miss the local (and small-ish) pydata conferences but somehow never felt the vibe of online only conferences :-/
i watch the recordings on youtube over a few weeks or so after the conference but somehow not sure what to do with myself when the conference is "on"
Sarah Gibson
@sgibson91
Glad y'all like /test-this-pr! And @betatim that is the ultimate plan 😉
I recently learned how to make commits over the GitHub API so I can make my bots more lightweight (no need for git to be installed, no need to locally clone repos). Got a new appreciation for what git add / git commit / git push does in the background
Tim Head
@betatim
oh wow
didnt know the GH API could do that
Sarah Gibson
@sgibson91
Min RK
@minrk
After deploying jupyterhub/mybinder.org-deploy#2043 the number of 404s served by hub.mybinder.org has dropped from 32,000 every 10 minutes, to 13 (not 13,000, just 13). These are replaced by 503 errors (424 after we bump jupyterlab to 3.1.17), but only ~1k, so a drop of about 75% of the total number of requests handled by the prod hub deployment.
1 reply
Screen Shot 2021-10-06 at 09.41.28.png
proxy cpu usage has also dropped a ton
Screen Shot 2021-10-06 at 09.45.31.png
mriduls
@mriduls:matrix.org
[m]
@minrk: could you also have a look at jupyterhub/binderhub#1402 It's preventing GESIS from taking in traffic from mybinder.org. Thanks!
Sarah Gibson
@sgibson91
Screenshot 2021-10-06 at 13.36.56.png
1 million pulls for the push-button binderhub on Azure!! :see_no_evil:
I should write another blog post...
Min RK
@minrk
:tada:
yuvipanda
@Yuvi:matrix.org
[m]
do we normally allow for requests like jupyterhub/mybinder.org-deploy#2050 The US Air force probably has money to pay for things
choldgraf
@choldgraf:matrix.org
[m]
it is kind of a judgment call each time, though for sure we generally are more inclined to give bumps to educatin/research/etc - basically groups that operate in the public interest. The air force is....complicated in that regard :-)
the problem with using "do they have money to pay or not?" as a rubric here is that there's no mechanism to allow anybody to pay even if they wanted to, so it's kind-of moot haha
Erik Sundell
@consideRatio
Omg so wonder with the /test-this-pr feature! Such a momentum boost to have that
update to docker-image-cleaner -> update to binderhub -> update to mybinder.org-deploy via /test-this-pr instead of merge -> debug at that point.
Erik Sundell
@consideRatio

I'm experiencing issues with the mybinder.org-deploy automation. I get 404 / 424 errors accessing some URLs that is tested against i think.

That makes the CI report a failure of the deployment on prod/turing, but practically the site is functional unless directly visiting hub.gke.mybinder.org for example.

Here is a failure: https://github.com/jupyterhub/mybinder.org-deploy/runs/3860610959

minrk
@minrk:matrix.org
[m]
Hm, can you change the test to /hub/api/health?
Min RK
@minrk
FWIW, that failure was caused by jupyterhub/mybinder.org-deploy#2044 where the custom failure route was moved to Hub config, instead of after-the-fact patches via the CHP API. I removed the explicit /hub/health route since it is redundant with /hub/api/health, but neglected the test.
Erik Sundell
@consideRatio
Ah thanks for sharing Min!