Every now and again, we would get flakey failures on rpc.cracknames for unknown reasons. Cracking SAMBADOMAIN\torturetest410$ would fail. What I've discovered is that this is partly due to collisions in the test user name when using rand() % 1000. When the user is deleted, the sAMAccountName persists on the deleted object from a previous test (I'm unsure of the behaviour on Windows). Requesting GUID format then performs a SHOW_RECYCLED search which causes an error when it finds a non-unique search result.
Created attachment 13424 [details] patch for 4.7 cherry-picked from master
Pushed to autobuild-v4-7-test.
(In reply to Karolin Seeger from comment #2) Pushed to v4-7-test. Closing out bug report. Thanks!