[Ganglia-developers] Latest on IRIX

[Steven W. <sw...@il...>]

FYI, on IRIX 6.5 I'm seeing the latest CVS build of gmond choke after 
running the pasted-in Linux mtu_func() ... but only on multi-processor systems.

What actually happens is that the data collection works OK - it walks the 
(four-interface-long) list of network interfaces, determines the lowest 
MTU, and passes it back to mtu_func.  mtu_func finishes the loop, populates 
val and apparently exits gracefully back to set_metric_value() ... but 
somewhere in there it dies.

I'd be more specific but I'm having about as much luck with gdb on these 
multi-proc SGI's as if I were using an antelope femur and some string. 
Can't even get an accurate stack trace - I found out the above with printf's...

Interestingly, when I add a line that forces val.uint32 to 0, it still 
bombs.  But when I clear out all the mtu_func code and put in two lines of 
dummy code, it works.

I guess it's possible that the intf_* stuff doesn't work well on multi-proc 
SGIs.  Any ideas?