Some glue around luigi.
Provides a base class, that autogenerates its output filenames based on
- some base path,
- a tag,
- the task id (the classname and the significant parameters)
Additionally, this package provides a few smaller utilities, like a TSV format, a benchmarking decorator and some task templates.
gluish.task.BaseTask
is intended to be used as a supertask.
from gluish.task import BaseTask
import datetime
import luigi
import tempfile
class DefaultTask(BaseTask):
""" Some default abstract task for your tasks. BASE and TAG determine
the paths, where the artefacts will be stored. """
BASE = tempfile.gettempdir()
TAG = 'just-a-test'
class RealTask(DefaultTask):
""" Note that this task has a `self.path()`, that figures out the full
path for this class' output. """
date = luigi.DateParameter(default=datetime.date(1970, 1, 1))
def run(self):
with self.output().open('w') as output:
output.write('Hello World!')
def output(self):
return luigi.LocalTarget(path=self.path())
When instantiating a RealTask
instance, it will automatically be assigned a
structured output path, consisting of BASE
, TAG
, task name and a slugified
version of the significant parameters.
task = RealTask()
task.output().path
# would be something like this on OS X:
# /var/folders/jy/g_b2kpwx0850/T/just-a-test/RealTask/date-1970-01-01.tsv
Was started on the
mailing list.
Continuing the example from above, lets create a task, that generates TSV
files, named TabularSource
.
from gluish.format import TSV
class TabularSource(DefaultTask):
date = luigi.DateParameter(default=datetime.date(1970, 1, 1))
def run(self):
with self.output().open('w') as output:
for i in range(10):
output.write_tsv(i, 'Hello', 'World')
def output(self):
return luigi.LocalTarget(path=self.path(), format=TSV)
Another class, TabularConsumer
can use iter_tsv
on the handle obtained
by opening the file. The row
will be a tuple, or - if cols
is specified -
a collections.namedtuple
.
class TabularConsumer(DefaultTask):
date = luigi.DateParameter(default=datetime.date(1970, 1, 1))
def requires(self):
return TabularSource()
def run(self):
with self.input().open() as handle:
for row in handle.iter_tsv(cols=('id', 'greeting', 'greetee'))
print('{0} {1}!'.format(row.greeting, row.greetee))
def complete(self):
return False
Log some running times. Mostly useful in interactive mode.
from gluish.benchmark import timed
class SomeWork(luigi.Task):
@timed
def run(self):
pass
def complete(self):
return False
Modeled after luigi.contrib.CopyToTable.
from gluish.esindex import CopyToIndex
import luigi
class ExampleIndex(CopyToIndex):
host = 'localhost'
port = 9200
index = 'example'
doc_type = 'default'
purge_existing_index = True
def docs(self):
return [{'_id': 1, 'title': 'An example document.'}]
if __name__ == '__main__':
task = ExampleIndex()
luigi.build([task], local_scheduler=True)
Elasticsearch support has been added to luigi as luigi.contrib.esindex.
Mirroring FTP shares. This example reuses the DefaultTask
from above. Uses the sophisticated lftp program under the hood, so it needs to be available on your system.
from gluish.common import FTPMirror
from gluish.utils import random_string
import luigi
class MirrorTask(DefaultTask):
""" Indicator makes this task run on each call. """
indicator = luigi.Parameter(default=random_string())
def requires(self):
return FTPMirror(host='ftp.cs.brown.edu',
username='anonymous', password='anonymous',
pattern='*pdf', base='/pub/techreports/00')
def run(self):
with self.input().open() as handle:
# FTPMirror output is in TSV
for row in handle.iter_tsv(cols=('path',)):
# do some useful things with the files here ...
def output(self):
return luigi.LocalTarget(path=self.path())
The output of FTPMirror
is a single file, that contains the paths to all mirrored files, one per line.
A short self contained example can be found in this gist.
To copy a single file from an FTP server, there is an FTPFile
template task.
Leverage command line tools with gluish.utils.shellout
. shellout
will
take a string argument and will format it according to the keyword arguments.
The {output}
placeholder is special, since it will be automatically assigned
a path to a temporary file, if it is not specified as a keyword argument.
The return value of shellout
is the path to the {output}
file.
Spaces in the given string are normalized, unless preserve_whitespace=True
is
passed. A literal curly brace can be inserted by {{
and }}
respectively.
An exception is raised, whenever the commands exit with a non-zero return value.
Note: If you want to make sure an executable is available on you system before the task runs,
you can use a gluish.common.Executable
task as requirement.
from gluish.common import Executable
from gluish.utils import shellout
import luigi
class GIFScreencast(DefaultTask):
""" Given a path to a screencast .mov, generate a GIF
which is funnier by definition. """
filename = luigi.Parameter(description='Path to a .mov screencast')
delay = luigi.IntParameter(default=3)
def requires(self):
return [Executable(name='ffmpg'),
Executable(name='gifsicle', message='http://www.lcdf.org/gifsicle/')]
def run(self):
output = shellout("""ffmpeg -i {infile} -s 600x400
-pix_fmt rgb24 -r 10 -f gif - |
gifsicle --optimize=3 --delay={delay} > {output} """,
infile=self.filename, delay=self.delay)
luigi.File(output).move(self.output().path)
def output(self):
return luigi.LocalTarget(path=self.path())
Sometimes the effective date for a task needs to be determined dynamically.
Consider for example a workflow involving an FTP server.
A data source is fetched from FTP, but it is not known, when updates are supplied. So the FTP server needs to be checked in regular intervals. Dependent tasks do not need to be updated as long as there is nothing new on the FTP server.
To map an arbitrary date to the closest date in the past, where an update
occured, you can use a gluish.parameter.ClosestDateParameter
, which is just an ordinary
DateParameter
but will invoke task.closest()
behind the scene, to
figure out the effective date.
from gluish.parameter import ClosestDateParameter
import datetime
import luigi
class SimpleTask(DefaultTask):
""" Reuse DefaultTask from above """
date = ClosestDateParameter(default=datetime.date.today())
def closest(self):
# invoke dynamic checks here ...
# for simplicity, map this task to the last monday
return self.date - datetime.timedelta(days=self.date.weekday())
def run(self):
with self.output().open('w') as output:
output.write("It's just another manic Monday!")
def output(self):
return luigi.LocalTarget(path=self.path())
A short, self contained example can be found in this gist.
If task.closest
is a relatively expensive operation (FTP mirror, rsync)
and the workflow uses a lot of ClosestDateParameter
type of parameters, it is
convenient to memoize the result of task.closest()
. A @memoize
decorator
makes caching the result simple:
from gluish.utils import memoize
...
@memoize
def closest(self):
return self.date - datetime.timedelta(days=self.date.weekday())
System package dependencies:
- Ubuntu: libmysqlclient-dev
- CentOS: mysql-devel
Setup:
$ git clone git@github.com:miku/gluish.git
$ cd gluish
$ mkvirtualenv gluish
$ pip install -r requirements.txt
$ nosetests
As of 4a7ec26ae6680d880ae20bebd514aba70d801687.
$ nosetests --verbose --with-coverage --cover-package gluish
Name Stmts Miss Cover Missing
------------------------------------------------
gluish 4 0 100%
gluish.benchmark 32 3 91% 59-62
gluish.colors 14 4 71% 12, 20, 24, 32
gluish.common 103 18 83% 68, 72, 80-81, 83, 86, ...
gluish.database 50 18 64% 12, 69-76, 79-88, 91-94
gluish.esindex 105 71 32% 40-43, 62-66, 70-71, 77-86, ...
gluish.format 27 2 93% 69, 72
gluish.intervals 16 0 100%
gluish.oai 58 14 76% 83-99
gluish.parameter 5 0 100%
gluish.path 82 14 83% 25, 39-48, 101-102, 165
gluish.task 45 2 96% 95, 114
gluish.utils 80 3 96% 35, 127, 166
------------------------------------------------
TOTAL 621 149 76%
----------------------------------------------------------------------
Ran 33 tests in 0.269s
$ pip install git-pylint-commit-hook
$ touch .git/hooks/pre-commit
$ chmod +x .git/hooks/pre-commit
$ echo '
#!/usr/bin/env bash
git-pylint-commit-hook
' > .git/hooks/pre-commit