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
Testing: pytest should be running multiple nodes in one single process #19796
Sorry if I missed an existing feature request with my searches I couldn't find one.
Hello,
When selecting multiple nodes to run tests, one would/could expect all of them to be run in one single process instantiation, for the following reasons:
performances: preparing some resources (data, loading modules on a big codebase...) may take some time and it is possible to have this time done once for all instead of once for each
reporting: one may want to get the coverage report generated by a selection of test nodes and coverage can be consolidated automatically with python-coverage instead of having to manually set coverage file option and then consolidate several files into one
isolation: it happens, despite efforts, that some tests don't behave the same when run together instead of one by one. To reproduce isolation issues that may be encountered later in CI, it'd be useful to be able to run in same conditions.
I've specified pytest in this issue although I think it may be valid for other runners, because I'm only using pytest and I know it's possible to specify several tests in command-line invocation.
Thinking about the case where people may expect parallel runs and consider current state as a feature, I think they should rather consider options like pytest-xdist instead.
I dug in the code, and, basically, it would mean that instead of having runTests do an await of delegated calls to runTest for each node, it may directly invoke pytest with all nodes in arguments.
Maybe even, if a parent and a child node are selected together only parent node could be passed to invocation, to avoid user mistakes leading to confusion.
In case this feature is accepted and needs external contribution, I may be able to code it
Hello! We are currently working on a rewrite of how we do pytest discovery/execution which will fix this issue. Here is the issue which outlines that work #17242 and the currently issue which is being worked on to prototype that functionality #19790.
Yes, we will keep this ticket open until the issue is fixed. Thank you for your offer to help, we will let you know if anything comes up especially related to verifying this bug is resolved once the fix is pushed.
Not sure if this is the exact same issue and if this will be fixed with the linked ticket. I noticed some strange behavior with unexpected parallel test execution.
Running this with the editor icon beside the line number it spawns two Python processes for the 2 tests (you can verify this with Process Explorer for example or just by checking the Python test output). This behavior with parallel execution is unexpected as this might break tests if shared resources are involved (in my case access to shared memory).
Running the test in the testing panel it spawns only one Python test process as expected.
Yes I get the same because explorer sees 2 tests at the same line (happens also when a test is a method of a class that gets inherited)
And so 2 nodes, and so running 2 parallel runs...
Hello! I have reviewed this issue and have tested a similar scenario successfully on our testing rewrite. Therefore I am going to close this issue but please comment or open a new issue if you are still seeing a problem when you try this yourself!
To use the rewrite yourself just add ”python.experiments.optInto": ["pythonTestAdapter"] to your user settings. You can confirm you have the rewrite enabled by setting your log level to trace, via the Developer: Set Log Level command in the command palette. Then check to see if Experiment 'pythonTestAdapter' is active is in your python logs.
Activity
eleanorjboyd commentedon Sep 8, 2022
Hello! We are currently working on a rewrite of how we do pytest discovery/execution which will fix this issue. Here is the issue which outlines that work #17242 and the currently issue which is being worked on to prototype that functionality #19790.
lachaib commentedon Sep 12, 2022
Hello Eleanor Boyd (@eleanorjboyd), so I understand this ticket is added to the meta-issue, and we can keep it open until done?
Again if you need a contribution I may be able to give it a try, otherwise I let you reach back to me when it's done.
Thanks
eleanorjboyd commentedon Sep 12, 2022
Yes, we will keep this ticket open until the issue is fixed. Thank you for your offer to help, we will let you know if anything comes up especially related to verifying this bug is resolved once the fix is pushed.
schlamar commentedon Sep 21, 2022
Not sure if this is the exact same issue and if this will be fixed with the linked ticket. I noticed some strange behavior with unexpected parallel test execution.
Minimal pytest example:
Running this with the editor icon beside the line number it spawns two Python processes for the 2 tests (you can verify this with Process Explorer for example or just by checking the Python test output). This behavior with parallel execution is unexpected as this might break tests if shared resources are involved (in my case access to shared memory).
Running the test in the testing panel it spawns only one Python test process as expected.
lachaib commentedon Sep 21, 2022
Yes I get the same because explorer sees 2 tests at the same line (happens also when a test is a method of a class that gets inherited)
And so 2 nodes, and so running 2 parallel runs...
eleanorjboyd commentedon Oct 30, 2023
Hello! I have reviewed this issue and have tested a similar scenario successfully on our testing rewrite. Therefore I am going to close this issue but please comment or open a new issue if you are still seeing a problem when you try this yourself!
To use the rewrite yourself just add
”python.experiments.optInto": ["pythonTestAdapter"]to your user settings. You can confirm you have the rewrite enabled by setting your log level to trace, via theDeveloper: Set Log Levelcommand in the command palette. Then check to see ifExperiment 'pythonTestAdapter' is activeis in your python logs.Thanks!