What is interesting is that sometimes this can happen to only one of three pods. So the others would start in more reasonable time of like 6-10 mins. So I guess problem exists when queued writes queue grows too large.