Fix Cassandra metrics about dd-agent

[alq666]

To build the agent, simply edit config.py to bump up the version and git push. Jenkins will take care of the rest (building and deploying to staging).

tests/test_cassandra contains some basic tests (in particular for cfstats).

[clofresh]

Is this error related? Saw it in syslog:

float() argument must be a string or a number
Traceback (most recent call last):
  File "/usr/share/datadog/agent/checks/cassandra.py", line 296, in check
    self._parseInfo(pipe.read(), results)
  File "/usr/share/datadog/agent/checks/cassandra.py", line 76, in _parseInfo     
    results["token"] = float(Cassandra._find(lines, r"^Token[^:]+: ([0-9]+)$"))
TypeError: float() argument must be a string or a number

[alq666]

What to do:

1. move cassandra to the new checks.d interface
2. tag cfstats by "keyspace" and "cf", rather than creating a device, prefix metrics with "cassandra" to match the old metrics name without going through the flatten_cassandra bit in the agent-driver.

[alq666]

Time to move to jmx to get these stats. nodetool's format is just too brittle.

[alq666]

I've done some research on alternate ways to get jmx data from python. None are fundamentally more appealing than jmxterm (from calling java objects in python via some elaborate but very alpha method to running extra servlet to convert jmx into json), save implementing JRMP natively in python.

So jmxterm it is. What would be interesting is profiling the python code that traverses all the mbeans of a cassandra node (e.g. running 1.1.6).

[remh]

I'm afraid it can really takes a while to go through all mbeans, I'll write a small script that does that and profile it to be sure.

[alq666]

Ok, go ahead and time it. Else we can extend jmxterm to dump all the attributes in 1 command.