Just to chime in, I’ve seen similar issues, where newly added containers (or cloned, but with updated image tag, i.e. the image does not exist locally yet and has to be pulled) would fail to start. On a second attempt it would work.
It feels like the start procedure fails when the image is missing and has to be pulled, but the pull operation is triggered. So on the subsequent attempt, the image is in place and the start up goes through.
I’ve not posted anything about this earlier as I wasn’t sure were the issue was, or how repeatable it was (using a private registry, so thought it might be related to that). But reading of a very similar issue here, I thought it might be worth mentioning… 
EDIT: just to be clear, I don’t know why/when this happens, as it doesn’t always cause troubles (no issue when doing a rancher-compose up for instance). But, one of the things I would’ve looked into before reporting was if I could spot there actually being two start attempts for the container when it was launched by a service. The issues I have described above was when I was manually adding a container directly in infrastructure page using the Add Container button, so I guess there’s not as much retries and fixups going on as when using a Service.