This extension allows you to create native DTrace providers for your Node.js applications. That is, to create providers and probes which expose information specific to your application, rather than information about the node runtime.

You could use this to expose high-level information about the inner workings of your application, or to create a specific context in which to look at information from other runtime or system-level providers.

The provider is not created in the usual way, by declaring it and then changing the build process to include it, but instead dynamically at runtime. This is done entirely in-process, and there is no background compiler or dtrace(1M) invocation. The process creating the provider need not run as root.

$ npm install dtrace-provider

Here's a simple example of creating a provider:

var d = require('dtrace-provider');

var dtp = d.createDTraceProvider("nodeapp");
var p1 = dtp.addProbe("probe1", "int", "int");
var p2 = dtp.addProbe("probe2", "char *");
dtp.enable();

Probes may be fired via the provider object:

dtp.fire("probe1", function() {
    return [1, 2];
});
dtp.fire("probe2", function() {
    return ["hello, dtrace via provider", "foo"];
});

or via the probe objects themselves:

p1.fire(function() {
  return [1, 2, 3, 4, 5, 6];
});
p2.fire(function() {
  return ["hello, dtrace via probe", "foo"];
});

Note that .fire() takes a callback that returns the arguments to be provided when the DTrace probe actually fires. This allows you to call .fire() unconditionally when you want to fire the probe, but the callback will be invoked only when the DTrace probe is actually enabled. This allows you to create probes whose arguments might be expensive to construct, and only do any work when the probe is actually enabled. (Examples might include converting a large object to a string representation or gathering large amounts of information.)

In some cases, creating a new closure to pass to .fire() each time it's called may introduce unwanted overhead. For extremely CPU-intensive or memory-conscious workloads, you can avoid this by lifting the closures for your hot probes into an outer scope. You can then supply arguments to that function as additional arguments to .fire(). As an example, you can convert the following program:

function manipulateObj(largeObj) {
    var count = 0;
    var name = null;
    ...
    p1.fire(function () {
        return [count, keyToValue(name), JSON.stringify(largeObj)];
    });
}

Into this one:

function f(a, b, c) {
    return [a, keyToValue(b), JSON.stringify(c)];
}

function manipulateObj(largeObj) {
    var count = 0;
    var name = null;
    ...
    p1.fire(f, count, name, largeObj);
}

Be careful to avoid passing .fire() additional arguments that are themselves expensive to construct, as that undermines the design goal here: minimizing the effect of disabled probes.

This example creates a provider called "nodeapp", and adds two probes. It then enables the provider, at which point the provider becomes visible to DTrace.

The probes are then fired, which produces this output:

$ sudo dtrace -Z -n 'nodeapp*:::probe1{ trace(arg0); trace(arg1) }'  \
                 -n 'nodeapp*:::probe2{ trace(copyinstr(arg0));  }'
dtrace: description 'nodeapp*:::probe1' matched 0 probes
dtrace: description 'nodeapp*:::probe2' matched 0 probes
CPU     ID                    FUNCTION:NAME
  1 123562                      func:probe1                 1                2
  1 123563                      func:probe2   hello, dtrace                    

Arguments are captured by a callback only executed when the probe is enabled. This means you can do more expensive work to gather arguments.

The maximum number of arguments supported is 32.

Available argument types are "int", for integer numeric values, "char *" for strings, and "json" for objects rendered into JSON strings.

Arguments typed as "json" will be created as "char *" probes in DTrace, but objects passed to these probe arguments will be automatically serialized to JSON before being passed to DTrace. This feature is best used in conjunction with the json() D subroutine, but is available whether or not the platform supports it.

# create a json probe:

var dtp = d.createDTraceProvider("nodeapp");
var p1 = dtp.addProbe("j1", "json");
dtp.enable();
p1.fire(function() { return { "foo": "bar" }; });

# on a platform supporting json():

$ sudo dtrace -Z -n 'nodeapp*:::j1{ this->j = copyinstr(arg0); \
                                    trace(json(this->j, "foo")) }'
dtrace: description 'nodeapp$target:::j1' matched 0 probes
CPU     ID                    FUNCTION:NAME
  0  68712                            j1:j1   bar

This libusdt-based Node.JS module supports 64 and 32 bit processes on Mac OS X and Solaris-like systems such as illumos or SmartOS. As more platform support is added to libusdt, those platforms will be supported by this module. See libusdt's status at:

https://github.com/chrisa/libusdt#readme

When using Mac OS X, be aware that as of 10.11 (El Capitan), DTrace use is restricted, and you'll probably want to disable SIP to effectively use DTrace.

FreeBSD 10 and 11 are also supported, but you'll need to make sure that you have the DTrace headers installed in /usr/src otherwise libusdt won't be able to compile. You can clone them using SVN, or find the correct src.txz here and extract that. Also note that FreeBSD 10 is restricted to only 4 working arguments per probe.

Platforms not supporting DTrace (notably, Linux and Windows) may install this module without building libusdt, with a stub no-op implementation provided for compatibility. This allows cross-platform npm modules to embed probes and include a dependency on this module.

GNU Make is required to build libusdt; the build scripts will look for gmake in PATH first, and then for make.

If compilation fails during installation on platforms with DTrace, then the library will fall back to the stub implementation that does nothing. To force an installation failure when compiling fails, set the environment variable NODE_DTRACE_PROVIDER_REQUIRE to hard:

$ NODE_DTRACE_PROVIDER_REQUIRE=hard npm install

This will then show you the output of the build process so you can see at which point it's having an issue. Common issues are:

• Missing a C/C++ compiler toolchain for your platform.
• python is Python 3 instead of Python 2; run npm config set python python2.7 (or similar) to set the Python binary npm uses.
• On OS X you may need to agree to the XCode license if that's the compiler toolchain you're using. This will usually manifest with an error like Agreeing to the Xcode/iOS license requires admin privileges, please re-run as root via sudo. To accept the license, you can run sudo xcodebuild -license.

Once you've found and fixed the issue, you can run npm rebuild to rerun the lifecycle scripts.

There is some overhead to probes, even when disabled. Probes are already using the "is-enabled" feature of DTrace to control execution of the arguments-gathering callback, but some work still needs to be done before that's checked. This overhead should not be a problem unless probes are placed in particularly hot code paths.

To clone the project's source code:

$ git clone --recursive https://github.com/chrisa/node-dtrace-provider.git

For issues, please use the GitHub issue tracker linked to the repository. GitHub pull requests are very welcome.

$ npm install
$ sudo ./node_modules/.bin/tap --tap test/*.test.js

This node extension is derived from the ruby-dtrace gem, via the Perl module Devel::DTrace::Provider, both of which provide the same functionality to those languages.