cf_worker
A Cuneiform worker implementation.
cf_worker is a worker implementation for the common runtime environment (CRE). The worker uses the Erlang foreign function interface (Effi) to execute tasks and interacts with the (distributed) filesystem via a Posix interface.
Usage
Adding the Cuneiform worker to a Project
Although the Cuneiform worker application can be imported also directly from GitHub, we recommend adding a dependency via hex.pm. Here, we show how this can be done using the build tools rebar3 or mix.
rebar3
To integrate the Cuneiform worker application into a rebar3-managed project change the deps entry in your application's rebar.config file to include the tuple {cf_worker, "0.1.5"}.
{deps, [{cf_worker, "0.1.5"}]}.
mix
{:cf_worker, "~> 0.1.5"}
Compiling
Having rebar3 available on your system, compile the project as an Erlang project by entering
rebar3 compile
If you want to drive the project from the command line please compile the project by entering
rebar3 escriptize
Starting the Cuneiform Worker
The Cuneiform worker application can be started in several different ways. It can be started from the command line, as an Erlang application, as a supervision tree hosting workers under a single supervisor, or directly as a process. In all cases there has to be a way for the workers to find the CRE instance and connect to it.
Starting from the Command Line
Compiling the Cuneiform client using escriptize creates an Erlang script file cf_worker which allows starting the Cuneiform client via the command line.
To display a help text enter
./cf_worker --help
This will show the command line synopsis, which looks like the following:
Usage: cf_worker [-v] [-h] [-s <suppl_file>] [-c <cre_node>] [-n <n_wrk>]
[-w <wrk_dir>] [-r <repo_dir>] [-d <data_dir>]
-v, --version Show cf_worker version.
-h, --help Show command line options.
-s, --suppl_file Supplementary configuration file.
-c, --cre_node Erlang node running the CRE application (must be
specified).
-n, --n_wrk Number of worker processes to start. 0 means
auto-detect available processors.
-w, --wrk_dir Working directory in which workers store temporary
files.
-r, --repo_dir Repository directory for intermediate and output data.
-d, --data_dir Data directory where input data is located.
To start the worker application from the command line and connect with a running CRE instance enter
./cf_worker -c cre@my_node
Here, we assume that the CRE runs on an Erlang node identified as cre@my_node.
Starting as an Erlang Application
If a CRE instance is already running on the same Erlang node you can start the Cuneiform worker application by calling
cf_worker:start().
Which is exactly the same as calling
application:start( cf_worker ).
Starting Under the Default Supervisor
To start the Cuneiform worker default supervisor under a custom supervision tree enter
CreNode = node().
NWrk = 4.
WrkDir = "./_cuneiform/wrk".
RepoDir = "./_cuneiform/repo".
DataDir = "./".
cf_client_sup:start_link( CreNode, NWrk, WrkDir, RepoDir, DataDir ).
This starts a worker supervisor with four workers using WrkDir to store temporal data, RepoDir for intermediate and output data, and DataDir to look up input data. Also, we expect a CRE to be running on the same node.
Starting Directly
The Cuneiform client process can be started directly. There are several ways to do this. The first is to start the process with a function that allows it to locate the CRE:
CreNode = node().
F = fun() -> cre:pid( CreNode ) end.
WrkDir = "./_cuneiform/wrk".
RepoDir = "./_cuneiform/repo".
DataDir = "./".
{ok, WorkerPid} = cf_worker_process:start_link( F, WrkDir, RepoDir, DataDir ).
Giving a function instead of a plain CRE process identifier has the advantage, that if the CRE crashes, taking the Cuneiform worker with it, the restarted worker instance uses the output of the function, which offers the possibility of locating the CRE under its new process identifier.
If this is too tedious, one can start it giving the CRE process identifier directly:
CrePid = cre:pid( node() ).
WrkDir = "./_cuneiform/wrk".
RepoDir = "./_cuneiform/repo".
DataDir = "./".
{ok, WorkerPid} = cf_worker_process:start_link( CrePid ).
Both previous direct starting methods do not register the Cuneiform client with any registry service.
System Requirements
Resources
- joergen7/cre. A common runtime environment (CRE) for distributed workflow languages.
- joergen7/cuneiform. A functional language for large-scale data analysis whose distributed execution environment is implemented on top of the CRE.
- joergen7/effi. Erlang foreign function interface.
Authors
- Jörgen Brandt (@joergen7) joergen.brandt@onlinehome.de