@Gargron We finally figured out. a bad relay was erroring out and causing our backlog to fill up which was already processing fairly close to its limit. disabled the relay and increased the parallelism on our end and its fixed now.
thanks anyway though.