You took the words right out of my mouth! :-) What the cf-monitord (new v3 cfenvd) was designed to do with its Nova extensions was to eliminate the need for SNMP by providing Unix-like sensors based on file streams and regular expressions (the two things that make Unix infinitely superior to anything else for management). Then, it does local (massively distributed) processing and leaks out highly compressed summaries that allow you to avoid the kinds of bottlenecks that other tools impose. We work hard on the design of this stuff to make sure the solutions scale.
M michoski wrote: > On 2/9/10 2:16 PM, "Justin Lloyd" <jll...@digitalglobe.com> wrote: >> Syslog-ng is what we've historically used for centralized logging and >> I'll be revamping it with v3 later this year. Combined with a >> logwatching tool, it makes a powerful combination with an snmp-based >> monitoring tool. Intuitively, I like better the idea of remote >> monitoring of Cfengine cf-execd processes and promise repairs via SNMP, >> though I seem to recall Mark not being the biggest fan of SNMP. If I'm >> recalling that correctly, I think there'd be a small chance of SNMP >> integration. > > I hate SNMP because it has "Simple" in the name and isn't, but I still admit > it's useful at times. > > It's possible a lot of "integration" may not be required to reap some > benefit. If you have a monitoring system and snmpd running on your hosts > (and cfengine pushing out snmpd.conf), SNMP can return the host's process > table... getting status (PID, CPU, mem, etc) for each cf* process remotely > from your monitoring hosts. This is straight-forward, but you could combine > a few of the discussed techniques with SNMP to go further... One example: > > -- Local process (cron, policy, part of monitoring) parses logs > > -- Local process houses logic to translate log events to monitoring speak > > -- snmpd.conf gets relevant "extend" > > e.g extend cfengine_stats /bin/cat /some/file > > (Keep logic out of the extends, so execution is fast.) > > -- Monitoring host does usual SNMP polling of cfengine_stats > > Our monitoring team does checks like this for other apps today, so I know > it's possible. I'm just not a SNMP guru, so can't provide more detail. > >> OTOH, perhaps centralized log monitoring is sufficient. It will take >> implementation, testing, and experience to determine this, I think. > > In general, having a few systems with some overlap isn't always bad (the > infamous onion)... One can fail, and you still get alerted. > > _______________________________________________ > Help-cfengine mailing list > Help-cfengine@cfengine.org > https://cfengine.org/mailman/listinfo/help-cfengine -- Mark Burgess ------------------------------------------------- Professor of Network and System Administration Oslo University College, Norway Personal Web: http://www.iu.hio.no/~mark Office Telf : +47 22453272 ------------------------------------------------- _______________________________________________ Help-cfengine mailing list Help-cfengine@cfengine.org https://cfengine.org/mailman/listinfo/help-cfengine
