Additional information -- looks like the issue is tied to the glibc version installed, specifically in regards to epoll_create() (which apr_pollset_create() calls).
With glibc < 2.9, epoll_create(0) errors out. However, with glibc 2.9, epoll_create(0) is totally valid. Cheers, Bernard On Fri, Mar 19, 2010 at 3:03 PM, Bernard Li <bern...@vanhpc.org> wrote: > Dear all: > > Looks like we have a bug in setup_listen_channels_pollset() in gmond.c. > > If your gmond.conf has no udp_recv_channel or tcp_accept_channel > defined, gmond will fail to run with the error message: > > apr_pollset_create failed: Invalid argument > > The error checking for apr_pollset_create() was recently implemented > since r2041. > > The issue seems to be that on certain platform, apr_pollset_create() > will fail if "total_listen_channels" = 0 (this is the "size" argument > according to the apr_pollset_create definition). > > Previously, since there was no error checking, the code would continue > merrily without erroring out. listen_channels will still be NULL and > thus would set deaf = 1 in main(). Now since we have error checking, > it actually bombs out. > > One fix is basically to check whether total_listen_channels is 0 prior > to the apr_pollset_create() call and if so just return. This should > give the same behaviour as before. > > So far I have been able to reproduce it on CentOS 5 x86_64. However, > there has been conflicting reports regarding whether this fails on > Ubuntu 9.04 or not. So if you guys could test this out and report > back what platforms you encounter this bug, that would be great. > > To reproduce the bug, simply comment out the udp_recv_channel and > tcp_accept_channel clauses and run gmond. It should fail with the > error message mentioned. > > Thanks to BarnacleBob on IRC for reporting/troubleshooting the bug. > > Cheers, > > Bernard > ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev _______________________________________________ Ganglia-developers mailing list Ganglia-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-developers
