You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Astropy data mirror download infrastructure is not robust #6061
@astrofrog as already mentioned somewhere that it's very unlikely that both the astropy data server and the github mirror fail at the same time, so this may be a symptom of a bug with the download infrastructure.
This test is set up to hit data.astropy.org twice and also www.astropy.org at the same time. Maybe that sometimes trigger some server-side panic? Will have to think about this some more...
I agree that there must be some bug hiding in the mirror infrastructure. Note that the easiest way to simulate an issue with the primary server is to edit your /etc/hosts file and add:
@pllim - I fear I have not looked carefully. It is just that regularly there is a failing travis tests, and scrolling to the bottom I see a URL timeout, so I restart. It would be very nice if this were not necessary.
One thing that is odd is that I only see this on travis - do the others not do any remote tests? Or does that mean it is the all-remote tests that usually fail.
See #7573 for a coupe of failures (including the one from @pllim in #6061 (comment)) that happen consistently if you 'turn off' one of the servers. The issue with that specific test is that it's flawed - it will only pass if both the main server and the mirror work.
Activity
bsipocz commentedon Jun 7, 2018
Here are another examples of the URL issues: https://circleci.com/gh/astropy/astropy/8682?utm_campaign=vcs-integration-link&utm_medium=referral&utm_source=github-build-link
https://travis-ci.org/astropy/astropy/jobs/389046646
@astrofrog as already mentioned somewhere that it's very unlikely that both the astropy data server and the github mirror fail at the same time, so this may be a symptom of a bug with the download infrastructure.
[-]Add more Astropy data mirrors?[/-][+]Astropy data mirror download infrastructure is not robust[/+]pllim commentedon Jun 7, 2018
Come to think of it, I wonder if it is the test itself that is flawed. Consider this:
This test is set up to hit
data.astropy.orgtwice and alsowww.astropy.orgat the same time. Maybe that sometimes trigger some server-side panic? Will have to think about this some more...mhvk commentedon Jun 13, 2018
Yes, have been seeing fairly regular timeouts again...
astrofrog commentedon Jun 13, 2018
I agree that there must be some bug hiding in the mirror infrastructure. Note that the easiest way to simulate an issue with the primary server is to edit your
/etc/hostsfile and add:pllim commentedon Jun 14, 2018
@mhvk do you mean just the
test_download_parallelor some other ones as well?mhvk commentedon Jun 14, 2018
@pllim - I fear I have not looked carefully. It is just that regularly there is a failing travis tests, and scrolling to the bottom I see a URL timeout, so I restart. It would be very nice if this were not necessary.
One thing that is odd is that I only see this on travis - do the others not do any remote tests? Or does that mean it is the all-remote tests that usually fail.
pllim commentedon Jun 14, 2018
Appveyor does not. CircleCI only does remote tests for
visualization.bsipocz commentedon Jun 15, 2018
And circleCI fails with those downloads regularly, too.
astrofrog commentedon Jun 18, 2018
I'll try and investigate this
astrofrog commentedon Jun 18, 2018
See #7573 for a coupe of failures (including the one from @pllim in #6061 (comment)) that happen consistently if you 'turn off' one of the servers. The issue with that specific test is that it's flawed - it will only pass if both the main server and the mirror work.
pllim commentedon Jun 18, 2018
Yeah... I think that is my bad (#6987)... I'll fix it...
pllim commentedon Jun 21, 2018
Everyone... Hopefully this is fixed. Please re-open if the problem resurfaces. Thanks for your patience!