This has been around for a very long time (at least a year, if memory serves
correctly). The problem is that the system "hangs" while trying to flush the
io buffers through the RML because it loses connection to the head node
process (for 1.x, that's basically mpirun) - but the "flush" procedure
do
I used to use this code to get the number of nodes in a cluster /
machine / whatever:
int
get_num_nodes(void)
{
int rc;
size_t cnt;
orte_gpr_value_t **values;
rc = orte_gpr.get(ORTE_GPR_KEYS_OR|ORTE_GPR_TOKENS_OR,
ORTE_NODE_SEGMENT, NULL, NULL, &cnt,
Hi Nathan
Could you tell us which version of the code you are using, and print out the
rc value that was returned by the "get" call? I see nothing obviously wrong
with the code, but much depends on what happened prior to this call too.
BTW: you might want to release the memory stored in the retur
Open MPI: 1.0.2
Open MPI SVN revision: r9571
The rc value returned by the 'get' call is '0'.
All I'm doing is calling init with my own daemon name, it's coming up
fine, then I immediately call this to figure out how many nodes are
associated with this machine.
-- Nathan
Cor
Rc=0 indicates that the "get" function was successful, so this means that
there were no nodes on the NODE_SEGMENT. Were you running this in an
environment where nodes had been allocated to you? Or were you expecting to
find only "localhost" on the segment?
I'm not entirely sure, but I don't believ
I'm running this on my mac where I expected to only get back the
localhost. I upgraded to 1.0.2 a little while back, had been using one
of the alphas (I think it was alpha 9 but I can't be sure) up until that
point when this function returned '1' on my mac.
-- Nathan
Correspondence
--
I agree with Ralph, this code should work fine (we do this internally in
orte_ras_base_node_query()). You may try adding a 'dump' of the GPR to
make sure that the node segment has information on it. Add a call like the
following to your function:
orte_gpr.dup_segment(NULL);
or better yet
ort
