Closed abouteiller closed 5 years ago
Original comment by Aurelien Bouteiller (Bitbucket: abouteiller, GitHub: abouteiller).
In pml_ob1_recvreq.c::mca_pml_ob1_put_completion
In pml_ob1_sendreq.c::mca_pml_ob1_rget_compeltion
Original comment by Aurelien Bouteiller (Bitbucket: abouteiller, GitHub: abouteiller).
This is related to errors in releasing UCT fragments, and needs to be addressed at the same time.
Original comment by Aurelien Bouteiller (Bitbucket: abouteiller, GitHub: abouteiller).
Code modified accordingly in #f8d53602
Original report by Aurelien Bouteiller (Bitbucket: abouteiller, GitHub: abouteiller).
In some cases, the status is not passed to the rdma callback.
At this point, we treat rdma_length == 0 as an indication of erroneous rdma operation, is this correct ?