I found in all the publish tasks, the test and development repo will take significantly longer time than the regular repo and they are very likely timeout. In such failing scenarios, we will restart the whole publish task for that platform.

Talking with Zakhar, I realized the barque can only run 1 publishing task for the same repo at the same time. Test and development repo are used for regular server builds as well so they are more likely timeout.  Also if we restart everything when it's timed out, it will add unnecessary load on the barque service as well. 

A good idea is to split out the test and development repo from releasing or have the ability to restart them separately.