Starting from this push in 10.10
some tests take many times longer, and almost inevitably fail with timeout. E.g. rocksdb.concurrent_alter fails on nearly every build/builder in buildbot, while before the change it was only taking a few seconds.
On my machine the ratio is not as big, but still very visible (it's the command that the test runs inside):
Probably there are other affected tests, the rocksdb is only most visible because it actually fails.
we probably need to change MTR default config to turn off SSL for mysqlslap, but first it needs to be confirmed that such a difference in default client performance is expected.