Hi.

We are using Specflow 3.5 and XUnit as the test flavour to generate the tests. We are on NCrunch 4.3.0.13 currently, awaiting new licenses before we can upgrade...

I thought I understood the parallisation model used by NCrunch, but I'm seeing behaviour which conflicts with that understanding (which may be an issue in SpecFlow and not NCrunch)

Behaviour I'm seeing is this:

Start 3 tests with 2 test max parallelism.

Test 1 calls [BeforeTestRun] from process dotnet, with process Id 1000 and I initialise a static WebDriver
Test 2 calls [BeforeTestRun] from process dotnet, with process Id 2000 and I initialise a static WebDriver
Test 1 (or test 2) completes, and calls [AfterTestRun] which disposes the static webdriver.
Test 3 is started. My understanding is that NCrunch will reuse existing process to run test 3 to save creating a new test process etc
Test 3 calls [BeforeTestRun] from process dotnet, with process Id 1000 (or 2000) and things fall over because WebDriver is already initialised.

I would expect that [AfterTestRun] is not called because the test process has not been finished with, as it is reused
I would expect that [BeforeTestRun] is not called because the environment has already been initialised for this process and so static variables are set up already.

We create a single static WebDriver so when many tests are run in sequence the runtime is reduced (previously the tests were run without NCrunch, but now we want to distribute the tests across the NCrunch grid)

Is this an NCrunch issue? Or is it something that specflow needs to fix?

There was a similar issue with specflow and NUnit previously so I wonder if this is similar:

https://github.com/SpecF...OSS/SpecFlow/issues/638

If it is a specflow issue, I may be able to fix that. I'll try and look at that option over the weekend, but wanted to post this issue in case there is something on the NCRunch side which is causing this.

Thanks. To be clear we are using xunit, not nunit. I think you are probably right, and I'll look into the underlying implementation a bit more to see if there are options there.

Problem with the static flag is that then AfterTestRun can never be executed, as we can't tell when things actually finish, so we will never dispose the web driver instance. I've tested just not throwing it away and it allows me to reuse it, but then when the tests finish the chrome driver instances remain.

I suppose I can look into running the AfterTestRun code when the process is terminated instead.

Thanks for the reply.

yep all good points. We already do some cleanup of older chromedriver processes for the exact reasons you outline but maybe the better approach would be to use something like the build number in the name of the chromedriver process and then have the build clean up the processes it knows that its run created.

thanks for the input. And for an awesome tool (our test run time is down to ~30% of what is was now I've switched the integration tests to run on the grid :))