Improve least_duration algorithm by sorting durations #28
Conversation
|
@jerry-git could you please review? This algorithm is useful anyway because it creates better splits plus it gives users an option that is not affected by issue 25. |
|
How about instead of adding yet another algorithm, instead roll this functionality into least_duration, but also restore the original order within the groups? You can easily do this by assigning an index to each of the items before sorting, then after splitting, sort each group by this index. |
Good suggestion, will change it. |
By sorting the tests by their duration, we can deal with the test with largest duration first and balance out the groups in later assignments. At the end we sort each selected group by their original items order to maintain relative ordering.
|
This is ready for review now. |
| The algorithm sorts the items by their duration. Since the sorting algorithm is stable, ties will be broken by | ||
| maintaining the original order of items. It is therefore important that the order of items be identical on all nodes | ||
| that use this plugin. Due to issue #25 this might not always be the case. |
There was a problem hiding this comment.
Sorting by duration should make it less sensitive to the test order so long as the durations are unique, no? Seems to me like sorting by duration instead of not sorting at all would fix #25. Alternatively, what about sorting by the test name, then the duration to resolve any possible ties?
There was a problem hiding this comment.
Sorting by duration should make it less sensitive to the test order
Indeed, sorting makes it less sensitive to this problem. Only the collected items that have the same duration will now need to be collected in the right order. However, ties are likely to happen because missing durations get filled with the same average duration.
Alternatively, what about sorting by the test name, then the duration to resolve any possible ties?
I think we should discuss this further in that issue, and we can adapt this algorithm again later once we agree on a solution. As I mention in that issue, sorting by name would resolve the issue flagged there, but I think collection of tests parametrised with objects would still be problematic.
mbkroese
changed the title
|
@jerry-git just a reminder for this one now that you're back from holiday :) |
|
yeah thanks, will have a look 🙂 |
|
available in 0.3.2 |
Updates the
least_durationalgorithm to create higher quality splits by sorting by test durations first.This algorithm is less likely to be affected by #25