Network stress test suite was created to catch flakes, it runs every test case 15 times https://github.com/openshift/origin/pull/24256

This job was always perma-failing, because the scheduler now will create 15 copies for every job and add them to the queue in random order, and sometimes two instances of the same test can be running at the same time. Even though there is a [Serial] label, it means don't run this test in parallel with the other tests, but not running in parallel with itself wasn't considered before. So a PR adding  [Serial:Self] label was created https://github.com/openshift/origin/pull/26359. Checking for every test if it can run with a copy of itself is too much work.

So the best fix would be to change queuing policy so that the same test won't run with another copy of itself - probably around these lines https://github.com/openshift/origin/blob/a946e2b9cd8acfdc73eac930b81623dc63f1a220/pkg/test/ginkgo/queue.go#L95-L133 (don't forget to remove  [Serial:Self] workaround if that's done).

This job should be disabled now, since its development was never finished really, and then once we fix the queuing policy and make sure the job consistently passes, it can be added back to regular jobs.