commonpp is a multi purpose library easing very few operations like:
- Getting metrics out of your program (counter, gauge, statistical description of sample);
- Naming your threads (and have the custom name appearing in
- Have a really simple thread pool on top of
- Quick setup of
I chose the BSD 2-clause license, so the library can be used in any project, however I'd appreciate any bug fix, pull request and comment.
commonpp depends on Boost and TBB, a C++11 compiler and CMake to be built.
It builds on Mac OS X and Linux. It probably build on *BSD but I did not test
$> make cmake $> make $> make test
commonpp can be built and used on windows. The prefered way is to use
The required dependencies can be installed using the following command:
$> .\vcpkg.exe install boost:x64-windows tbb:x64-windows hwloc:x64-windows
Then you can generate a Visual Studio solution giving the vcpkg toolchain file.
Please note that only VS2017 has been tested. Also, static
cannot be done, so the DLL will need to be in the path to execute binaries
Although Boost and TBB are very complete low-level libraries, there are still
basic features missing, the biggest being getting out metrics out of your code.
commonpp will never replace Boost or TBB but instead adds some features that
The core part is mostly utility functions or class.
ExecuteOnScopeExit: executes a callable on scope exit. It can be canceled;
LoggingInterface: It is a header containing several logging function on top of
boost::log. Every record produced with
commonppis tagged so that it can be used in a project already using
init_loggingshould not be called).
RandomValuePicker: select a random value in a read only container;
FloatingArithmeticTools: Knuth's double comparison functions;
Options: an utility class working along with an enum to offer a simple interface to manage options, see the test;
- There are several string functions to stringify, encode, join, or get a formatted date.
The thread library is quite small:
Thread.hppcontains a function to get/set the current thread name;
ThreadPoolis a class managing several threads calling the
* It can schedule a callable to be called periodically; * It supports several `io_service`; * A function can be called on thread startup to setup any thread specific data;
Spinlock: should be obvious
ThreadTimer: it allows one to get the load of the current thread. This is experimental;
This is the main reason this
commonpp exists. I often found myself struggling
to get metrics out of my code to monitor performances, behavior, resource
usage, etc. Most of the library counters are based on the RRD ones, the
ExponentiallyDecayingReservoir is based on the implementation in Codahale as
well as the test.
Metrics.hppis the entry point of the metric library; one should use it to register counters;
MetricTag: This represents the metadata associated with a counter; for instance if you use Graphite, the tag will be converted in a path (concatenation of all the value separated by a point), if you use InfluxDB, it will be converted in a list of tag;
MetricValue: This holds the actual value(s) of a counter.
There are 3 types of counter:
SharedCountershould be used to represent a value that always increases and never goes back like a request counter. The actual value sent is the rate at which the counter increases. The SharedCounter can be shared among several thread safely.
Gauge: It represents an absolute value like a number of active connections.
DescStat: Associated with a reservoir, it will describe the distribution of the values, for instance, one can use a reservoir to push the time a function
Xtakes to run, and the
StatDescwill give several information like the min, max, variance, quantile, mean, etc.
There is also a helper:
TimeScopeis an helper to measure the time we spent in the current scope.
There is a complete example how to use the metric library here
Currently only one is implemented:
Documentation can be found
- Graphite: It sends a list of
MetricValuepair in a format understood by Graphite (actually Carbon). As stated before, the values are concatenated to form the metric path, therefore the order used to insert the tag is important.
- InfluxDB: It sends the measure name along with all the tags and the value
as described in the documentation. The
MetricValueclass were designed with InfluxDB primarely in mind.
This library for now consist in a HTTP request generator and parser and URL encode and decode functions.
- Migrate my other project HTTPP to use commonpp;
- Add tests.