-
Notifications
You must be signed in to change notification settings - Fork 937
Closed
Labels
Description
In a bunch of Cisco MTT runs, I'm getting error messages from PMIx at the end of the run. For example, in https://mtt.open-mpi.org/index.php?do_redir=2273, it looks like the test passed (this is a build from master), but then something apparently went wrong inside MPI_FINALIZE:
================ test_acc2 ========== Mon Mar 7 20:36:32 2016
Time per int accumulate 1.411261 microsecs
P31, Test No. 0, PASSED: multi-offset accumulate Mon Mar 7 20:36:32 2016
P0, Test No. 0, PASSED: accumulate performance Mon Mar 7 20:36:32 2016
================ test_acc2 ========== Mon Mar 7 20:36:32 2016
P0, Test No. 1, PASSED: accumulate self without permission Mon Mar 7 20:36:32 2016
P0, Test No. 2, PASSED: accumulate self Mon Mar 7 20:36:32 2016
P1, Test No. 0, PASSED: accumulate non-self Mon Mar 7 20:36:32 2016
[mpi018:03156] pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 61]
[mpi018:03156] pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 76][mpi018:03156]
pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 87]
[mpi018:03156] pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 79]
[mpi018:03156] pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 83]
[mpi018:03156] pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 56]
[mpi018:03156] pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 86][mpi018:03156]
pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 69]
[mpi018:03156] pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 85]
[mpi018:03156] pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 73][mpi018:03156]
pmix_usock_msg_send_bytes: write failed: Broken pipe (32) [sd = 64]
[mpi018:03156] mpirun: SIGPIPE detected on fd 13 - aborting
(there's other tests in that MTT URL that are failing because usNIC is complaining about lack of connectivity -- ignore those...)
I see similar pmix errors in "hello world" and "ring" tests, too.