- Issue created by @mstrelan
- 🇦🇺Australia mstrelan
MR !11584 demonstrates how this would fail on a slow runner, where the loop took more than 10 seconds to update the element 1100 times.
The controller was added in #3211164: Random errors in Javascript Testing → when we had
WebDriverCurlServicebut that's not really relevant anymore. I'm also not sure why we need to replace it 1100 times, it seems we could simplify this by replacing an element once or twice, at an interval of around 500-1000 ms. - 🇦🇺Australia mstrelan
Switched to replacing 3 times with 500ms delay. The theory is that if it takes 4400ms for 1100 iterations, minus 1100ms in delays, that's 3300ms computation time, or around 3ms per iteration. A slower runner that is timing out might be taking 9ms per iteration, adding up to 11000 ms. But if the slow runner only has to do the work 3 times that's only ~27ms, on top of the 1500ms delay. Should be much less likely to time out.
- 🇺🇸United States smustgrave
Explanation in #5 makes sense. Lets give it a shot.
- 🇬🇧United Kingdom catch
The theory sounds good but we should run this a few hundred times before committing it.
- 🇦🇺Australia mstrelan
mstrelan → changed the visibility of the branch 3514924-repeat to hidden.
- 🇦🇺Australia mstrelan
mstrelan → changed the visibility of the branch 3514924-repeat-fix to hidden.
- 🇦🇺Australia mstrelan
Opened 2 MRs (both now closed) and ran the test 500 times in each. I think this one was more infrequent than 1/500 so unfortunately we didn't catch it. It's also probably not worth retrying, because even if we do catch it once in another 500 run, it doesn't really confirm that we've fixed it. I think this can go back to RTBC, but will leave that for someone else.
- 🇺🇸United States smustgrave
Seems feedback has been addressed
Personal question when testing for multiple loops why add - <<: *with-selenium-chrome ?
- 🇦🇺Australia mstrelan
Personal question when testing for multiple loops why add - <<: *with-selenium-chrome ?
The default config for repeating tests doesn't include a chrome service, which we need for JS tests. Perhaps it should by default, since a large percentage of random fails are in JS tests.