Create Scheduler for Grouping Tests w/ Variable number of Workers #1130
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
TUR-21619 is a prototype method of executing groups of pytest test cases in series with each group running it's specific test cases in parallel with a variable number of workers. Tests can be marked with a custom mark. This custom mark specifies an arbitrary group name, and the number of processes to run tests in that group with, with the number of processes separated from the group name by an underscore. Due to limitations in pytest/execnet a shutdown signal must be sent to a node when one test case remains on the node (the test case is actually ran immediately, but we do not receive feedback through the channel until a shutdown signal is sent due to an old design decision in pytest). Because of this, in order to run additional tests after a group is nearly complete (when each worker has at most 1 test remaining), we must shutdown each node, recreate each node, and then schedule more tests. This adds overhead as we must teardown + startup nodes, however, this overhead is on the order of 1 second per teardown/startup. If this custom scheduling method allows us to save more than 1 second in test execution, it is worth the increased overhead cost. Note: The -n argument with --dist customgroup is the maximum number of worker processes. If -n specifies 4 but a specific group specifies 8, the specific group will be ran across only 4 nodes, we will never spin up more nodes than the -n argument allows. --------- Co-authored-by: Alvaro Barbeira <[email protected]>
* fix pytest error * clearer if
* update docstring * idk
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This pull request addresses Issue 1014. It adds a new scheduler that allows a user to add tests to arbitrary groups using the pytest mark
xdist_custom. Using the mark a group name and number of workers to use for that group is specified. Groups are ran sequentially but tests in each group are ran in parallel with the given number of workers.This pull request is a rough prototype. We aren't too familiar with pytest-xdist internals and we don't expect it is merge-able as is. However, we're hoping to open the conversation to see if there are improvements we can make to get this merged into pytest-xdist.
Since we wanted groups of tests to execute sequentially, we did not want to schedule tests from a pending group until after the currently running group completed all tests. We found that in order to obtain the status of the final test scheduled on a worker, we needed to send a shutdown signal to the worker. Because of this, between groups a shutdown signal is sent to each worker, and then workers are initialized again prior to scheduling the next group of tests.
We left the directory
xdist-testing-ntopin as an example but will remove that and add proper tests.Will add a
changelog/1014.featureif this PR gains traction.