问题
I'm trying to find a solution that allows me to monitor resource consumption of a server. Preferably, the metrics I'm wanting to obtain are network utilisation IO, and if possible CPU usage/load average and disk IO.
The only other requirement I have is that this information be obtainable by Java so it can be manipulated, and at least work on Linux (Fedora).
I've heard about a few monitoring tools but I'm just not sure of the best way of going about this. I would probably want to be gathering the information about every 30 seconds.
Thanks
Update: Just to re-iterate, I am referring to system-wide monitoring NOT Java specific monitoring. I just want to use Java to access to these metrics
回答1:
You could choose to delegate the monitoring to a dedicated tool like Cacti, Centreon, or Zenoss but this might be a bit overkill for a single application.
For a simple solution, JMX might indeed be a better solution. As starting point, I suggest reading the following article: Monitoring Local and Remote Applications Using JMX 1.2 and JConsole. Then, have a look at Using JConsole to Monitor Applications, a very detailed article that shows how to use JConsole to access several core monitoring and management functionalities provided by the Java platform including:
- Detect low memory
- Enable or disable GC and class loading verbose tracing
- Detect deadlocks
- Control the log level of any loggers in an application
- Access OS resources—Sun's platform extension
- Manage an application's Managed Beans (MBeans)
But, AFAIK, JMX won't give you access to network IO so you might need a combination of these tools. Luckily, many tools (e.g. Cacti, SmokePing) use the RDD format that you can easily manipulate with Java APIs like JRobin or rdd4j.
回答2:
Just want to throw RHQ (http://rhq-project.org/ ) in the mix :-)
回答3:
You could look into using the Ganglia monitoring tool. It uses XML for its data representation and so I imagine it would be fairly straightforward to access the data from Java, and there are the added advantages that it's been designed to be highly scaleable, with the potential to record server metrics across a large number of machines.
回答4:
I'm growing fond of collectd, a modular C daemon focusing on monitoring (rather than graphing) with a multitude of plugins:
- Read - the usual CPU, I/O, network, load, memory and JMX as well as some more interesting ones like the ability to monitor DNS requests by sniffing packets, monitor latency/connectivity, and 'tail' a logfile to monitor things like SSH login attempts or incoming emails
- Write - send readings to a CSV file, remote host, RRDtool, Unix socket, webserver
- Notify - via libnotify or email
- Bindings - to Perl, Python and (most importantly for your purposes) Java
There would seem to be several options for getting metrics into your Java code:
- in-process, using the aforementioned Java plugin to register a write callback to receive data from the various other plugins
- over the network, by embedding jcollectd (a Java implementation of collectd's protocol) into your app
- indirectly, by writing to a CSV or RRD file and using one of the various RRD implementations for Java
回答5:
Your best bet would probably look in /proc for all your systems resource usage /proc/cpuinfo and /proc/net/ are a good place to start.
回答6:
What's wrong with Runtime.exec("some command")?
回答7:
I am not sure if this would assist in your query. IMHO Cacti and Centreon seem to be too complex to invest time into for a requirement as simple as yours. There are simpler tools out there for server monitoring such as New Relic and SeaLion. SeaLion is the my current favourite. It gives simple default commands which cover most basic requirements (will probably cover yours as well). It's also extremely simple to setup and is free. You could try these out.
来源：https://stackoverflow.com/questions/1404986/simple-server-monitoring-with-java