[Gluster-users] writev: Transport endpoint is not connected

[Lindolfo Meira]

Dear all,

I've been trying to benchmark a gluster file system using the MPIIO API of 
IOR. Almost all of the times I try to run the application with more than 6 
tasks performing I/O (mpirun -n N, for N > 6) I get the error: "writev: 
Transport endpoint is not connected". And then each one of the N tasks 
returns "ERROR: cannot open file to get file size, MPI MPI_ERR_FILE: 
invalid file, (aiori-MPIIO.c:488)".

Does anyone have any idea what's going on?

I'm writing from a single node, to a system configured for stripe over 6 
bricks. The volume is mounted with the options _netdev and transport=rdma. 
I'm using OpenMPI 2.1.2 (I tested version 4.0.0 and nothing changed). IOR 
arguments used: -B -E -F -q -w -k -z -i=1 -t=2m -b=1g -a=MPIIO. Running 
OpenSUSE Leap 15.0 and GlusterFS 5.3. Output of "gluster volume info" 
follows bellow:

Volume Name: gfs
Type: Stripe
Volume ID: ea159033-5f7f-40ac-bad0-6f46613a336b
Status: Started
Snapshot Count: 0
Number of Bricks: 1 x 6 = 6
Transport-type: rdma
Bricks:
Brick1: pfs01-ib:/mnt/data/gfs
Brick2: pfs02-ib:/mnt/data/gfs
Brick3: pfs03-ib:/mnt/data/gfs
Brick4: pfs04-ib:/mnt/data/gfs
Brick5: pfs05-ib:/mnt/data/gfs
Brick6: pfs06-ib:/mnt/data/gfs
Options Reconfigured:
nfs.disable: on


Thanks in advance,

Lindolfo Meira, MSc
Diretor Geral, Centro Nacional de Supercomputação
Universidade Federal do Rio Grande do Sul
+55 (51) 3308-3139