by removing the sleep() calls and reducing timeout time, tests remain
as accurate while taking less than 5% of the required time to run
fix T6914
stefan_schmidt | |
cedric |
by removing the sleep() calls and reducing timeout time, tests remain
as accurate while taking less than 5% of the required time to run
fix T6914
Automatic diff as part of commit; lint not applicable. |
Automatic diff as part of commit; unit tests not applicable. |
T6915 documents all remaining issues with the eio test suite that I'm aware of. None of the issues from this ticket are related to this change; they were either concealed by the extremely long timeouts or are rare enough that they have not been noticed until now.
I remember that there was a potential race condition between when a file is created and when a monitor is added and that is why we had this usleep. Basically, the file could some time popup in the monitor even if it was clearly created before. Did you see that problem or is it gone ? I would blame the kernel for such a behavior, but well, it impacts our tests.
I ran the updated tests hundreds of times. I ran them with forking enabled and disabled. I ran them under extreme cpu load and under no load. The tests succeeded in every case excepting the issues listed in T6915.
At no point could this be an issue since I added a check to ignore events on the directory which is created before starting the main loop. I don't know whether such events are received but it's irrelevant even if they are.
I am actually on it right now. Just giving it some testing and push it when it all holds together.