ompi/runtime: add instead of set opal_progress_event_flags

[wzamazon]

Currently, ompi_mpi_init() call

 opal_progress_set_event_flag(OPAL_EVLOOP_NONBLOCK),

whith the intention to ensure OPAL_EVLOOP_NONBLOCK is set
in opal_progress_event_flag.

However, this call will remove other existing flags
(like OPAL_EVLOOP_ONCE) in opal_progress_event_flag,
which can cause deadlock.

This patch address the issue by adding OPAL_EVLOOP_NONBLOCK
to that flag.

Signed-off-by: Wei Zhang [email protected]
(cherry picked from commit f22d897)

[wzamazon]

gcc9 build failed. However, the error message is vague

 ...
 CC       pbsend_init_f.lo
  CC       pbuffer_attach_f.lo
  CC       pbuffer_detach_f.lo
  CC       pcancel_f.lo
  CC       pcart_coords_f.lo
  CC       pcart_create_f.lo
  CC       pcartdim_get_f.lo
  CC       pcart_get_f.lo
  CC       pcart_map_f.lo
  CC       pcart_rank_f.lo
  CC       pcart_shift_f.lo
  CC       pcart_sub_f.lo
  CC       pclose_port_f.lo
FATAL: command execution failed
Command Close created at
	at hudson.remoting.Command.<init>(Command.java:70)
	at hudson.remoting.Channel$CloseCommand.<init>(Channel.java:1312)
	at hudson.remoting.Channel$CloseCommand.<init>(Channel.java:1310)
	at hudson.remoting.Channel.close(Channel.java:1486)
	at hudson.remoting.Channel.close(Channel.java:1453)
	at hudson.remoting.Channel$CloseCommand.execute(Channel.java:1318)
	at hudson.remoting.Channel$1.handle(Channel.java:607)
	at hudson.remoting.SynchronousCommandTransport$ReaderThread.run(SynchronousCommandTransport.java:81)
...

[jsquyres]

That usually means a Jenkins worker died abnormally. Just try running the test again.

[jsquyres]

bot:ompi:retest

[gpaulsen]

bot:aws:retest

[wzamazon]

All CI tests passed. Ready for review.