Dear Ganglia Developers,
Gmetad 2.5.x never had a cleaup thread, which caused a type of memory
leak if there were lots of gmetrics from monitored clusters. This
version 2.5.7 fixes this problem: a gmetad cleanup thread trims the
metric hash tables of metrics whose DMAX (delete time) has been
reached.
I have tested this with a large monitoring tree of 1500 hosts (3
monitoring nodes), and it looks stable. My test machines were Linux
2.4, P4 and Athlon.
I respectfully ask anyone interested to compile and test this version
of Ganglia monitoring code. Gmond has remained unchanged, only gmetad
has new code.
The tarball is available here:
http://heron.sdsc.edu/ganglia-software/ganglia-monitor-core-2.5.7.tar.gz
Federico
Rocks Cluster Group, San Diego Supercomputer Center, CA
