source: main/waeup.kofa/trunk/src/waeup/kofa/utils/batching.txt @ 7933

:mod:`waeup.kofa.utils.batching` -- Batch processing
****************************************************

Batch processing is much more than pure data import.

Overview
========

Basically, it means processing CSV files in order to mass-create,
mass-remove, or mass-update data.

So you can feed CSV files to processors, that are part of
the batch-processing mechanism.

Processors
----------

Each CSV file processor

* accepts a single data type identified by an interface.

* knows about the places inside a site (University) where to store,
  remove or update the data.

* can check headers before processing data.

* supports the mode 'create', 'update', 'remove'.

* creates log entries (optional)

* creates csv files containing successful and not-successful processed
  data respectively.

Output
------

The results of processing are written to loggers, if a logger was
given. Beside this new CSV files are created during processing:

* a pending CSV file, containing datasets that could not be processed

* a finished CSV file, containing datasets successfully processed.

The pending file is not created if everything works fine. The
respective path returned in that case is ``None``.

The pending file (if created) is a CSV file that contains the failed
rows appended by a column ``--ERRROR--`` in which the reasons for
processing failures are listed.

The complete paths of these files are returned. They will be in a
temporary directory created only for this purpose. It is the caller's
responsibility to remove the temporay directories afterwards (the
datacenters distProcessedFiles() method takes care for that).

It looks like this::
 
     -----+      +---------+
    /     |      |         |              +------+
   | .csv +----->|Batch-   |              |      |
   |      |      |processor+----changes-->| ZODB |
   |  +------+   |         |              |      |
   +--|      |   |         +              +------+
      | Mode +-->|         |                 -------+
      |      |   |         +----outputs-+-> /       |
      |  +----+->+---------+            |  |.pending|
      +--|Log |  ^                      |  |        |
         +----+  |                      |  +--------+
           +-----++                     v
           |Inter-|                  ----------+
           |face  |                 /          |
           +------+                | .finished |
                                   |           |
                                   +-----------+


Creating a batch processor
==========================

We create an own batch processor for an own datatype. This datatype
must be based on an interface that the batcher can use for converting
data.

Founding Stoneville
-------------------

We start with the interface:

    >>> from zope.interface import Interface
    >>> from zope import schema
    >>> class ICave(Interface):
    ...   """A cave."""
    ...   name = schema.TextLine(
    ...     title = u'Cave name',
    ...     default = u'Unnamed',
    ...     required = True)
    ...   dinoports = schema.Int(
    ...     title = u'Number of DinoPorts (tm)',
    ...     required = False,
    ...     default = 1)
    ...   owner = schema.TextLine(
    ...     title = u'Owner name',
    ...     required = True,
    ...     missing_value = 'Fred Estates Inc.')
    ...   taxpayer = schema.Bool(
    ...     title = u'Payes taxes',
    ...     required = True,
    ...     default = False)

Now a class that implements this interface:

    >>> import grok
    >>> class Cave(object):
    ...   grok.implements(ICave)
    ...   def __init__(self, name=u'Unnamed', dinoports=2,
    ...                owner='Fred Estates Inc.', taxpayer=False):
    ...     self.name = name
    ...     self.dinoports = 2
    ...     self.owner = owner
    ...     self.taxpayer = taxpayer

We also provide a factory for caves. Strictly speaking, this not
necessary but makes the batch processor we create afterwards, better
understandable.

    >>> from zope.component import getGlobalSiteManager
    >>> from zope.component.factory import Factory
    >>> from zope.component.interfaces import IFactory
    >>> gsm = getGlobalSiteManager()
    >>> cave_maker = Factory(Cave, 'A cave', 'Buy caves here!')
    >>> gsm.registerUtility(cave_maker, IFactory, 'Lovely Cave')

Now we can create caves using a factory:

    >>> from zope.component import createObject
    >>> createObject('Lovely Cave')
    <Cave object at 0x...>

This is nice, but we still lack a place, where we can place all the
lovely caves we want to sell.

Furthermore, as a replacement for a real site, we define a place where
all caves can be stored: Stoneville! This is a lovely place for
upperclass cavemen (which are the only ones that can afford more than
one dinoport).

We found Stoneville:

    >>> stoneville = dict()

Everything in place.

Now, to improve local health conditions, imagine we want to populate
Stoneville with lots of new happy dino-hunting natives that slept on
the bare ground in former times and had no idea of
bathrooms. Disgusting, isn't it?

Lots of cavemen need lots of caves.

Of course we can do something like:

    >>> cave1 = createObject('Lovely Cave')
    >>> cave1.name = "Fred's home"
    >>> cave1.owner = "Fred"
    >>> stoneville[cave1.name] = cave1

and Stoneville has exactly

    >>> len(stoneville)
    1

inhabitant. But we don't want to do this for hundreds or thousands of
citizens-to-be, do we?

It is much easier to create a simple CSV list, where we put in all the
data and let a batch processor do the job.

The list is already here:

    >>> open('newcomers.csv', 'wb').write(
    ... """name,dinoports,owner,taxpayer
    ... Barneys Home,2,Barney,1
    ... Wilmas Asylum,1,Wilma,1
    ... Freds Dinoburgers,10,Fred,0
    ... Joeys Drive-in,110,Joey,0
    ... """)

All we need, is a batch processor now.

    >>> from waeup.kofa.utils.batching import BatchProcessor
    >>> class CaveProcessor(BatchProcessor):
    ...   util_name = 'caveprocessor'
    ...   grok.name(util_name)
    ...   name = 'Cave Processor'
    ...   iface = ICave
    ...   location_fields = ['name']
    ...   factory_name = 'Lovely Cave'
    ...
    ...   def parentsExist(self, row, site):
    ...     return True
    ...
    ...   def getParent(self, row, site):
    ...     return stoneville
    ...
    ...   def entryExists(self, row, site):
    ...     return row['name'] in stoneville.keys()
    ...
    ...   def getEntry(self, row, site):
    ...     if not self.entryExists(row, site):
    ...       return None
    ...     return stoneville[row['name']]
    ...
    ...   def delEntry(self, row, site):
    ...     del stoneville[row['name']]
    ...
    ...   def addEntry(self, obj, row, site):
    ...     stoneville[row['name']] = obj
    ...
    ...   def updateEntry(self, obj, row, site):
    ...     # This is not strictly necessary, as the default
    ...     # updateEntry method does exactly the same
    ...     for key, value in row.items():
    ...       setattr(obj, key, value)

If we also want the results being logged, we must provide a logger
(this is optional):

    >>> import logging
    >>> logger = logging.getLogger('stoneville')
    >>> logger.setLevel(logging.DEBUG)
    >>> logger.propagate = False
    >>> handler = logging.FileHandler('stoneville.log', 'w')
    >>> logger.addHandler(handler)

Create the fellows:

    >>> processor = CaveProcessor()
    >>> result = processor.doImport('newcomers.csv',
    ...                   ['name', 'dinoports', 'owner', 'taxpayer'],
    ...                    mode='create', user='Bob', logger=logger)
    >>> result
    (4, 0, '/.../newcomers.finished.csv', None)

The result means: four entries were processed and no warnings
occured. Furthermore we get filepath to a CSV file with successfully
processed entries and a filepath to a CSV file with erraneous entries.
As everything went well, the latter is ``None``. Let's check:

    >>> sorted(stoneville.keys())
    [u'Barneys Home', ..., u'Wilmas Asylum']

The values of the Cave instances have correct type:

    >>> barney = stoneville['Barneys Home']
    >>> barney.dinoports
    2

which is a number, not a string.

Apparently, when calling the processor, we gave some more info than
only the CSV filepath. What does it all mean?

While the first argument is the path to the CSV file, we also have to
give an ordered list of headernames. These replace the header field
names that are actually in the file. This way we can override faulty
headers.

The ``mode`` paramter tells what kind of operation we want to perform:
``create``, ``update``, or ``remove`` data.

The ``user`` parameter finally is optional and only used for logging.

We can, by the way, see the results of our run in a logfile if we
provided a logger during the call:

    >>> print open('stoneville.log').read()
    --------------------
    Bob: Batch processing finished: OK
    Bob: Source: newcomers.csv
    Bob: Mode: create
    Bob: User: Bob
    Bob: Processing time: ... s (... s/item)
    Bob: Processed: 4 lines (4 successful/ 0 failed)
    --------------------

We cleanup the temporay dir created by doImport():

    >>> import shutil
    >>> import os
    >>> shutil.rmtree(os.path.dirname(result[2]))

As we can see, the processing was successful. Otherwise, all problems
could be read here as we can see, if we do the same operation again:

    >>> result = processor.doImport('newcomers.csv',
    ...                   ['name', 'dinoports', 'owner', 'taxpayer'],
    ...                    mode='create', user='Bob', logger=logger)
    >>> result
    (4, 4, '/.../newcomers.finished.csv', '/.../newcomers.pending.csv')

This time we also get a path to a .pending file.

The log file will tell us this in more detail:

    >>> print open('stoneville.log').read()
    --------------------
    ...
    --------------------
    Bob: Batch processing finished: FAILED
    Bob: Source: newcomers.csv
    Bob: Mode: create
    Bob: User: Bob
    Bob: Failed datasets: newcomers.pending.csv
    Bob: Processing time: ... s (... s/item)
    Bob: Processed: 4 lines (0 successful/ 4 failed)
    --------------------

This time a new file was created, which keeps all the rows we could not
process and an additional column with error messages:

    >>> print open(result[3]).read()
    owner,name,taxpayer,dinoports,--ERRORS--
    Barney,Barneys Home,1,2,This object already exists in the same container. Skipping.
    Wilma,Wilmas Asylum,1,1,This object already exists in the same container. Skipping.
    Fred,Freds Dinoburgers,0,10,This object already exists in the same container. Skipping.
    Joey,Joeys Drive-in,0,110,This object already exists in the same container. Skipping.

This way we can correct the faulty entries and afterwards retry without
having the already processed rows in the way.

We also notice, that the values of the taxpayer column are returned as
in the input file. There we wrote '1' for ``True`` and '0' for
``False`` (which is accepted by the converters).

Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))


We can also tell to ignore some cols from input by passing
``--IGNORE--`` as col name:

    >>> result = processor.doImport('newcomers.csv', ['name',
    ...                             '--IGNORE--', '--IGNORE--'],
    ...                    mode='update', user='Bob')
    >>> result
    (4, 0, '...', None)

Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))

If something goes wrong during processing, the respective --IGNORE--
cols won't be populated  in the resulting pending file:

    >>> result = processor.doImport('newcomers.csv', ['name', 'dinoports',
    ...                             '--IGNORE--', '--IGNORE--'],
    ...                    mode='create', user='Bob')
    >>> result
    (4, 4, '...', '...')

    >>> print open(result[3], 'rb').read()
    name,dinoports,--ERRORS--
    Barneys Home,2,This object already exists in the same container. Skipping.
    Wilmas Asylum,1,This object already exists in the same container. Skipping.
    Freds Dinoburgers,10,This object already exists in the same container. Skipping.
    Joeys Drive-in,110,This object already exists in the same container. Skipping.


Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))




Updating entries
----------------

To update entries, we just call the batchprocessor in a different
mode:

    >>> result = processor.doImport('newcomers.csv', ['name',
    ...                             'dinoports', 'owner'],
    ...                    mode='update', user='Bob')
    >>> result
    (4, 0, '...', None)

Now we want to tell, that Wilma got an extra port for her second dino:

    >>> open('newcomers.csv', 'wb').write(
    ... """name,dinoports,owner
    ... Wilmas Asylum,2,Wilma
    ... """)

    >>> wilma = stoneville['Wilmas Asylum']
    >>> wilma.dinoports
    1

Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))


We start the processor:

    >>> result = processor.doImport('newcomers.csv', ['name',
    ...                    'dinoports', 'owner'], mode='update', user='Bob')
    >>> result
    (1, 0, '...', None)

    >>> wilma = stoneville['Wilmas Asylum']
    >>> wilma.dinoports
    2

Wilma's number of dinoports raised.

Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))


If we try to update an unexisting entry, an error occurs:

    >>> open('newcomers.csv', 'wb').write(
    ... """name,dinoports,owner
    ... NOT-WILMAS-ASYLUM,2,Wilma
    ... """)

    >>> result = processor.doImport('newcomers.csv', ['name',
    ...                             'dinoports', 'owner'],
    ...                    mode='update', user='Bob')
    >>> result
    (1, 1, '/.../newcomers.finished.csv', '/.../newcomers.pending.csv')

Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))

    
Also invalid values will be spotted:

    >>> open('newcomers.csv', 'wb').write(
    ... """name,dinoports,owner
    ... Wilmas Asylum,NOT-A-NUMBER,Wilma
    ... """)

    >>> result = processor.doImport('newcomers.csv', ['name',
    ...                             'dinoports', 'owner'],
    ...                    mode='update', user='Bob')
    >>> result
    (1, 1, '...', '...')

Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))


We can also update only some cols, leaving some out. We skip the
'dinoports' column in the next run:

    >>> open('newcomers.csv', 'wb').write(
    ... """name,owner
    ... Wilmas Asylum,Barney
    ... """)

    >>> result = processor.doImport('newcomers.csv', ['name', 'owner'],
    ...                             mode='update', user='Bob')
    >>> result
    (1, 0, '...', None)

    >>> wilma.owner
    u'Barney'

Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))


We can however, not leave out the 'location field' ('name' in our
case), as this one tells us which entry to update:

    >>> open('newcomers.csv', 'wb').write(
    ... """name,dinoports,owner
    ... 2,Wilma
    ... """)

    >>> processor.doImport('newcomers.csv', ['dinoports', 'owner'],
    ...                    mode='update', user='Bob')
    Traceback (most recent call last):
    ...
    FatalCSVError: Need at least columns 'name' for import!

This time we get even an exception!

We can tell to set dinoports to ``None`` although this is not a
number, as we declared the field not required in the interface:

    >>> open('newcomers.csv', 'wb').write(
    ... """name,dinoports,owner
    ... "Wilmas Asylum",,"Wilma"
    ... """)

    >>> result = processor.doImport('newcomers.csv', ['name',
    ...                             'dinoports', 'owner'],
    ...                    mode='update', user='Bob')
    >>> result
    (1, 0, '...', None)

    >>> wilma.dinoports is None
    True

Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))

Generally, empty strings are considered as ``None``:

    >>> open('newcomers.csv', 'wb').write(
    ... """name,dinoports,owner
    ... "Wilmas Asylum","","Wilma"
    ... """)

    >>> result = processor.doImport('newcomers.csv', ['name',
    ...                             'dinoports', 'owner'],
    ...                    mode='update', user='Bob')
    >>> result
    (1, 0, '...', None)

    >>> wilma.dinoports is None
    True

Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))


Removing entries
----------------

In 'remove' mode we can delete entries. Here validity of values in
non-location fields doesn't matter because those fields are ignored.

    >>> open('newcomers.csv', 'wb').write(
    ... """name,dinoports,owner
    ... "Wilmas Asylum","ILLEGAL-NUMBER",""
    ... """)

    >>> result = processor.doImport('newcomers.csv', ['name',
    ...                             'dinoports', 'owner'],
    ...                    mode='remove', user='Bob')
    >>> result
    (1, 0, '...', None)

    >>> sorted(stoneville.keys())
    [u'Barneys Home', "Fred's home", u'Freds Dinoburgers', u'Joeys Drive-in']

Oops! Wilma is gone.

Clean up:

    >>> shutil.rmtree(os.path.dirname(result[2]))


Clean up:

    >>> import os
    >>> os.unlink('newcomers.csv')
    >>> os.unlink('stoneville.log')