as reported in the linked URL and described in the following email thread : http://www.mail-archive.com/ganglia-developers@lists.sourceforge.net/msg04584.html gmetad will segfault when configured to pull metrics from another gmetad that reports extra metric description data (3.1.0) a fix has been developed and is currently queued for releasing upstream with 3.1.1 but has been catalogued as a critical showstopper for 3.1.0 as it won't allow a hierarchical gmetad setup to work (which is commonly used for scalability reasons) Reproducible: Always Steps to Reproduce: 1. install gmetad in "leaf" and start it to collect cluster metrics as a leaf 2. install gmetad in "head" and configure it to pull from the "leaf:8651" 3. Actual Results: gmetad in "head" segfaults after the fist pull Expected Results: gmetad collects data in "head" from all configured "leaf" servers and stores summarization from them visible through the web frontend. in amd64 with debugging enabled : #0 0x00007ffec0f82520 in hash_lookup (key=0x43e3d820, hash=0x0) at hash.c:304 #1 0x0000000000405860 in startElement_EXTRA_ELEMENT (data=0x43e3dbf0, el=<value optimized out>, attr=0x61ed10) at process_xml.c:731 #2 0x0000000000405e4c in start (data=0x43e3dbf0, el=0x61cd6c "EXTRA_ELEMENT", attr=0x61ed10) at process_xml.c:1010
Created attachment 162504 [details, diff] patch to avoid collecting extra data for not authoritative sources fixed in trunk and committed for backport for the 3.1 stable branch and expected to be released with 3.1.1
Pushed to cvs as ganglia-3.1.0-r1. Thanks for linking to the mailing list discussion and for the patch :)