On Sun, May 9, 2010 at 01:01, Chandler Wilkerson <[email protected]> wrote: > I'm running a Power7 Linux cluster where we allow users to change the > number of active threads on each core on a per-job basis. I would like > to track these changes by allowing cpu_num to be a variable metric > rather than a check-once metric. I've found an if statement in the > cpu_num_func function in libmetrics/linux/metrics.c that seems to > redundantly prevent the function from working more than once. > > cpu_num_func ( void ) > { > static int cpu_num = 0; > g_val_t val; > > /* Only need to do this once */ > if (! cpu_num) { > /* We'll use _SC_NPROCESSORS_ONLN to get operating cpus */ > cpu_num = get_nprocs(); > } > val.uint16 = cpu_num; > > return val; > } > > > I've got a working setup monitoring cpu count changes on the fly > simply by commenting out the if statement above. I just wanted to put > a question to the list and developers whether there's a reason for the > if statement when the default configuration sets this metric as a > check_once. If not, I'd like to request a patch be accepted to remove > the if statement.
Is your change against trunk or 3.1.x, and can you please post it to the list? -- Jesse Becker Every cloud has a silver lining, except for the mushroom-shaped ones, which come lined with strontium-90. ------------------------------------------------------------------------------ _______________________________________________ Ganglia-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ganglia-general
