Corrupted CDO size by pool manager

On miniHPC with verbs fabrics.

In a simple benchmark with MPI+Maestro with

1. Pool manager (PM)
2. One consumer that demands all CDOs (sink all)
3. Two producers, each producer one CDO

PM broadcasts its info to all components. 
MPI_barier 
The consumer declares, requires, demands, and disposes all CDOs (two in this case).
Producers declare and offer CDOs (one per producer)
MPI_barier 
Producers withdraw and dispose CDOs.

The PM reports a corrupted CDO size to the consumer. It seems to be a race condition in the PM, because it does not always happen. Also, I do not think it is related to offer/withdraw do not block correctly because I use MPI barrier between all offers/require/demand and withdraws, so blocking should not be the issue.

Please find the code and log attached.

core_benchmark.c

miniHPC_verbs_corrupted_CDO_info.log

omp_consumer.c

omp_consumer.h

omp_injector.c

omp_injector.h

run.sh