Chris, I disagree in some way with you ;)
Instrumentation is difficult because you (as a developer ?) want to go deep, but it is not necessary for a production team to go that deep to know when something is going wrong. It's a developer challenge, and he could choose to rely only on good logging. Whatever happens, there should be at least some logs for the production team, then they can 'grep' for SEVERE or Exceptions before asking more help to the N2 or N3 team. "1. Java doesn't directly support SNMP;" Sorry, but it does ( http://docs.oracle.com/javase/7/docs/technotes/guides/management/snmp.html ) and we use it with good results (deltas on GCs for example). Maybe are you confusing with the fact that Java has no built-in API for SNMP software developers. "2. If the JVM is braindead, the SNMP agent can't announce any state to the managers." The SNMP server will be unable to check the JVM, it will raise an alert. Easy, cheap, and in most cases enough ;) best regards A.T. 2015-10-24 14:18 GMT+02:00 Christopher Schultz <ch...@christopherschultz.net >: > Aurélien, > > On 10/23/15 6:47 PM, Aurélien Terrestris wrote: > > "Live monitoring is the only way to go, unless you want your customers > > to surprise you with performance problems on your own site. :)" > > > > Monitoring, ok, but alerting is critical for large scale production > > environments. > > Apologies... I already considered "alerting" to be a critical component > of "monitoring", so I didn't mention it specifically. Instrumentation is > the hard part; alerting is relatively easy once instrumentation has been > done. > > > You mention JMX callbacks for alerting, but it would be worth > > that you mention SNMP since it's very easy to set up and efficient. > > SNMP is great, but there are a few problems with using it with Java: > > 1. Java doesn't directly support SNMP; instead, you have to use > a 3rd-party library (which isn't difficult to find). This may be > a barrier for certain organizations who only want to use their > own code in their own applications. > > 2. If the JVM is braindead, the SNMP agent can't announce any state > to the managers. This requires the use of probe-based instrumentation > (active probing from the outside of the application), unless you want > to have an agent send a heartbeat for every status change. > > > We also can raise alerts on deltas with this method, and see a > > problem before the CPU says something bad is going to happen. > > There's nothing stopping you from doing deltas with other techniques. My > presentation even includes a tool that allows you to do deltas. > > -chris > > --------------------------------------------------------------------- > To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org > For additional commands, e-mail: users-h...@tomcat.apache.org > >
