commandRunner is yet another package created to handle running commands, scripts or programs on the command line targetted at thread safe execution for Celery. The simplest class lets you run anything locally on your machine. Other classes are targeted at Analytics and data processing platforms such as Grid Engine (and eventually HADOOP). The class attempts to run commands in a moderately threadsafe way by requiring that you provide it with sufficient information that it can build a uniquely labeled temporary directory for all input and output files. This means that this can play nicely with things like Celery workers.
You can install the required python modules with pip using the included requirements.txt file.
If using R you must have a recent >3.0.0 installation of R and its development libraries.
If you are using the DRMAA support you must be able to set the environment variables
DRMAA_LIBRARY_PATH SGE_ROOT
This release supports running commands on local unix shell and DRMAA compliant grid engine installs (ogs, soge and univa) and it can also run strings of python code. Commands are built/interpolated via some simple rules described below.
geRunner and localRunner are capable of executing arbitrary binaries as though run in a unix shell
This is the basic usage for running a binary using the local unix shell:
from commandRunner.localRunner import *
r = localRunner(tmp_id="ID_STRING", tmp_path=,/tmp/", out_glob=['file', ],
command="ls /tmp", input_data={DATA_DICT})
r.prepare()
exit_status = r.run_cmd(success_params=[0])
r.tidy()
print(r.output_data)
__init__ initalises all the class variables needed and performs the command string interpolation which allow you to parameterise the command.
Interpolation rules work the following way. The command string is split in to tokens. The tokens are serched for specific control flags the define inputs, output, paths and command parameters. Take the following call:
r = localRunner(tmp_id="ID_STRING",
tmp_path="/tmp/",
in_glob=['.in', ]
out_glob=['.file', ],
params=["-a","-l", "b"]
param_values={'b': {'value': 'this',
'spacing': True,
'switchless': False},
}
command="ls $P1 $P2 $P$ /tmp",
input_data={'input.in': 'some input data'},
std_out_str="file.stdout",
identifier="string"
env_vars={"str":"str"},
debug=False,)
This effectively builds the following command:
ls -a -l b this /tmp > file.stdout
Command string interpolation accepts a range of control sequences which begin with $.
If you provide a list of strings via the in_glob function variable these will be made availabe in the command string using $I[int]. tmp_id string and each sequential entry in the in_glob list are concatenated. For tmp_id="this" and in_glob=[".in", ".thing", ] two strings are created this.in and this.thing and they can be refered to in the command string as $I1 and $I2.
A near identical interpolation is carried out for the out_glob list providing $O[int]. With tmp_id="this" and out_glob=['.file', ] string this.file will be created and can be refered to in the command string as $O1.
Command line parameters are more subtle. These will be available as $P[int] where each integer refers to a successive entry in the params list. If param_values is not provided then the values in params will be interpolated as the appear in the params list. If params_values is provided then more sophisticated control of the param interpolation can be achieved. Providing a value (12) means that a param entry ("b") will be interpolated as "b 12". Setting spacing to false suppresses the space, "b12" and setting switchless to True suppresses the param name "12".
We also provide a couple of convenience strings $TMP, $ID and $VALUE. There will interpolate the contents of tmp_path, tmp_id and value_string respectively
Next it takes input_data. This is a dict of {Filename:Data_string} values. Iterating over, it writes the data to a file given the values in temp_path and temp_id. So given the following dict and the values above:
{ "test.file" : "THIS IS MY STRING OF DATA"}
would result in a file with the path /tmp/ID_STRING/test.file
env_vars is a dict of key:value strings which is used to set the unix environment variables for the running command.
debug takes a boolean. this controls if the tmp directory created is world writeable. Debug mode assumes you wish to leave behind the tmp dir to allow other process to access or analyse it.
Note that only tmp_id, tmp_path and command are required. Omitting input_data or out_glob assumes that there are respectively no input files to write or output files to gather and interpolations for $I[int], $O[int] and $P[int] will not reference anything.
The line r.run_cmd(success_params=[0]) runs the command string provided.
Once complete if out_globs have been provided and the files were output then the contents of those files can be found in the dict r.output_data. which has the same {Filename:Data_string} form as the input_data dict:
{ "output.file" : "THIS IS MY PROCESSED DATA"}
r.tidy() cleans up deleting any input and output files and the temporary working directory. Any data in the output file is available in to r.output_data
geRunner uses python DRMAA to submit jobs. A consequence of this that a command string is not constructed in quite the same way. The first portion of the command string is split off as a command. Subsequence portions are tokenised and added to a params array to be passed to DRMAA
The Options dict is flattened to a key:value list. You can include or omit as many of those as you'd like options as you like. Any instance of the string $I[int] and $O[int] in final args array will be interpolated as usual
If std_out_string is provided it will be used as a file where the Grid Engine thread STDOUT will be captured:
from commandRunner.geRunner import *
r = geRunner(tmp_id="ID_STRING", tmp_path="/tmp/", out_glob=['.file'],
command="ls -lah", input_data={"File.txt": "DATA"},
params = ["-file"]
param_values = {'-file': {'value': '$O1',
'spacing': True,
'switchless': False},
},
std_out_string="std.out")
r.prepare()
exit_status = r.run_cmd(success_params=[0])
r.tidy()
print(r.output_data)
Although DRMAA functions differently you can think of this as effectively run the following command (after following the interpolation rules):
ls -file out.file -lah > std.out
commandRunner classes can also call code natively, pythonRunner will take blocks of python code, rRunner will take blocks of R code. Both construct a temp directory and place the input data there. Any code passed will then execute as though is is running from the temp directory (via os.chdir).
In theory you can provide any arbitrarily large chunk of python or R code. In practice you probably want to keeps these to short single function scripts for less than 100 lines as debugging is quite tricky given the layer of abstraction.
It is also worth noting that accepted code forms a dialect of both python and R; the " character is not valid and you must use the single quote to bound strings.
Execution by pythonRunner is somewhat different to geRunner and localRunner. Instances of this class take a script arg and not a command arg and .prepare() and .run_cmd() function somewhat differently:
from commandRunner.pythonRunner import *
r = pythonRunner(tmp_id="ID_STRING",
tmp_path="/tmp/",
in_glob=['.in', ]
out_glob=['.file', ],
params=["-a","-l", "b"]
param_values={'b': {'value': 'this',
'spacing': True,
'switchless': False},
}
script="print(str(I1.read()))",
input_data={'input.in': 'some input data'},
std_out_str="file.stdout",
identifier="string"
env_vars={"str":"str"},
)
r.prepare()
exit_status = r.run_cmd()
r.tidy()
print(r.output_data)
As before input_data is a dict of 'file name': 'data' pairs which will be written to a directory specified by tmp_path+tmp_id+"/" (i.e. /tmp/ID_STRING/). in_glob and out_glob specify a set of file handles that will be opened for you so you do not have to open them in your provided script. in_globs should be matched to file names in input_data. In the example above the in_glob for '.in' will open the input.in data file and that will be available as a variable named I1. If there were more entries in in_glob they would be named in sequence I1, I2 I3 etc... out_glob functionas as a form of promise that your script will write to some output files. For each entry in out_glob a filehandle for writing is opened using the tmp_id as the file name. As above O1 would open a file called ID_STRING.file
Params are also created as variables, named P1, P2, P3, etc... These refer in order to the values in the params list. If there is not an entry for the param in param_values these variables are set to True. If there is an entry in the param_values arg then the variable will be a dict with a key value pair that gives you the name and the value. In the example above P3 is a dict of {'b': "this"}, In this way some runtime configuration can be passed in to the script.
Anything provided to env_vars will be add to the script environment using additions to os.environ[]
script is an argument that takes any valid python string. In the example above it reads the contents from the I1 filehandle ('some input data') and then echos that to stdout. In theory you can place any sized piece of python here but smaller scripts made up of a handful of lines are probably more ane/sensible. Note that escape characters will need to be double escaped (\n not n)
When .prepare() is called a temp directory is build and the input_data files are written to it. Next various filehandles and param variables are composed and appended to the provided script. Once the new script is prepared compile() is called on it to ensure the script is a valid python string. Assuming .prepare() is succesful you can then call .run_cmd().
run_cmd() creates a new python subprocess, runs the script in this child process (insulating it from the namespace of the parent process) and captures any writes to stdout and stderr.
Once complete you can find the outputs in the .output_data dict. There will be and entry for stdout with a key named for your std_out_str. There will also be a key for stderr named tmp_id+".err", in this example "ID_STRING.err". As per local runner there will be a key for every file that matched the provided out_glob list as long as the file has a non-zero size. If you do not write to one of the provided output file handles they will not be collected in output_data
rRunner makes use of rpy2 to execute R code. You may need to amend your LD_LIBRARY_PATH https://stats.stackexchange.com/questions/6056/problems-with-librblas-so-on-ubuntu-with-rpy2
The API and broad functioning is roughly similar to the pythonRunner. Unlike pythonRunner code is not checked for syntactic correctness before execution. So any errors will occur at runtime for the code you provide.
File handles (I1, I2, ... and O1, O2 etc...) are available as above. These are opened with R's base file() function. You may wish instead to override these with things like csv.reader() where it is more convenient. Params (P1, P2, etc...) also exist, name:value pairings are avaiable R lists() rather than python dicts.
Anything provided to env_vars will be add to the script environment using Sys.setenv()
Unlike the python case it is imperative you check the that the error data in output_data is empty before assuming your R code ran successfully. As above you can find the outputs from the stdout of your script in the output_data variable. We leave it to you
Best to run these 1 suite at a time, geRunner tests will fail if you do not have Grid Engine installed, DRMAA_LIBRARY_PATH set and SGE_ROOT set, for example:
export DRMAA_LIBRARY_PATH=/opt/ogs_src/GE2011.11/lib/linux-x64/libdrmaa.so export SGE_ROOT=/opt/ogs_src/GE2011.11/
Run tests with:
python setup.py test -s tests/test_commandRunner.py python setup.py test -s tests/test_localRunner.py python setup.py test -s tests/test_geRunner.py python setup.py test -s tests/test_pythonRunner.py python setup.py test -s tests/test_rRunner.py
- Implement octaveRunner for Octave backend
- matlab? mathematica?