james et al- i *think* i've made gmond more bullet-proof to problems caused when a client prematurely closes its connection to the XML port. this will hopefully fix the problem where gstat can crash gmond. let me know otherwise.
see the latest CVS -matt Aug 7, James Braid wrote forth saying... > > that is the same problem. heartbeat messages are sent every > > 15 seconds so > > if a machine doesn't get a heartbeat message in 60 seconds (4 missed > > heartbeats) it assumes it is down. if you use the latest CVS > > source you > > should see the problem no longer is there. let me know otherwise. > > Just checked out the latest CVS (at 13:06 NZST). It looks like that > problem is fixed. The REPORTED value looks like its incrementing as it > should....sweet! > > BUT > > Gstat doesn't like talking to gmond (this happens whether I try to > connect using gstat on the local machine, or remotely... > > The following is from gmond running with debug = 6 > > <snip> > 4 pre_process_node() remote_ip=10.0.1.130 > pre_process_node() HOSTNAME =tycho.peace.co.nz > pre_process_node() TIMESTAMP=1028682333 > pre_process_node() HASHP =100128bc0 > pre_process_node() USER_HASHP=1001290e0 > pre_process_node() returning the ganglia internal hash pointer 100128bc0 > mcast_listen_thread() got internal hash 100128bc0 > mcast_listen_thread() built metricdata struct > mcast_listen_thread() attempting to hash_insert_data > mcast_listen_thread() inserted data into 100128bc0 > server_thread() 6 clientfd = 11 > > sent data to host 127.0.0.1 > Broken Pipe > </snip> > > However, telnetting to the port works fine (and I can do this as much as > I like and gmond stays alive, just running gstat kills it straight > away): > > lilo:~# telnet tycho 8649 > Trying 10.0.1.130... > Connected to tycho.peace.co.nz. > Escape character is '^]'. > <?xml version="1.0" encoding="ISO-8859-1" standalone="yes"?> > <!DOCTYPE GANGLIA_XML [ > <!ELEMENT GANGLIA_XML (CLUSTER)+> > <!ATTLIST GANGLIA_XML VERSION CDATA #REQUIRED > SOURCE CDATA #REQUIRED> > > Etc, etc, etc (all the stats are there)... > > Any ideas about this one? > > Thanks for the prompt and helpful answers > > James > >
