[libfabric-users] Trouble setting up a connection
Jörn Schumacher
joern.schumacher at cern.ch
Tue Feb 23 13:16:56 PST 2016
On 02/23/2016 08:17 PM, Hefty, Sean wrote:
>>> thanks for this list. I am trying to write a minimal example using
>>> libfabric in reliable connection mode. I extracted most of my approach
>>> from the msg.c example in fabtests.
>>>
>>> Here is my attempt: https://gist.github.com/joerns/bfe5d13969ce0bc008c4
>>>
>>> I got so far that the server listens and waits for incoming events on an
>>> EQ, and the client tries to connect. But when the client tries to
>>> connect and waits for a "connection established" event, the fi_eq_sread
>>> call fails and returns -259:
>>
>> I'd make one small tweak to your error checking: call fi_strerror()
>>>
>>>> error: reading from event queue failed (after connect): rd=-259
>>>
>>> Any ideas what could be the reason for that?
>>
>> 259 is "Error available" -- but yeah, what does that mean?
>
> It means that there is an error available on the EQ. You need to call fi_eq_readerr to retrieve it.
Indeed, I managed to read out the error from the EQ and it was
"connection timed out". I then changed the server side as Sean suggested
in his first E-Mail and added the FI_SOURCE flag to fi_getinfo and the
missing fi_accept. The connection is established now.
I updated the gist with the latest (now working) version.
Thanks and cheers,
Jörn
More information about the Libfabric-users
mailing list