I’ve experienced the same as you, I’ve lost many hours to fixing different protocol time out errors. We’re running it in ECS now and it’s pretty stable, every now and then one pops up but is fixed by retrying the job.

Do you have a good record of these sites/configurations? We suspect it's related to the host environment in some way that we can't reproduce on our machines. It'd be useful information for debugging if you can share it. Thanks!