I'm starting to see that tests start running too soon before the application started. Increasing the wait time to a bigger number.
It will increase the test runtime of each selfhost variation by a couple more seconds, but it will increase the reliability.
A couple of selfhost variations are failing on some of the new CI machines. While trying to reproduce I see that the klr.exe processes are lauching fine. But the tests fail for a couple of self-host variations. Only suspicion is this time out as the client gets a socket exception. It is possible that client is making a request before the server is ready.
This does not include the deployment helper on mono. So the test needs to point to a statically deployed application url to run.
There are some bugs in mono Httpclient like in case of 302's RequestMessage's Uri not modified, cookies not cleared in cookie container etc. I will file bugs in mono on those and this is a work around until then.
There is still more work and refactoring. This is a first step.
Adding some delay for the selfhost tests. With the recent change of merging the dbs selfhost based tests start even before the data is populated into the db & application start. Adding a slight delay to allow the setup finish.
And fixing a test bug that was not surfacing. Was checking with a fixed album id. This worked before as the id created was always 453.
Value of the KRE_DEFAULT_LIB of the 'k test' process was flowing into Helios resulting in an exception. As the test is always running on desktop clr, the value of this variable always points to the desktop KRE path resulting in the exception. This fix will fix the issue.
Praburaj
Suhas Joshi
David Fowler