zodbupdate 2.0

Custom software/eggs

It is important to install this egg in an interpreter/environment where your software is installed as well. If you’re using a regular Python installation or virtualenv, just installing the package using easy_install should be fine.

If you are using buildout, installing can be done using the egg recipe with this configuration:

[buildout]
parts += zodbupdate

[zodbupdate]
recipe = zc.recipe.egg
eggs = zodbupdate
    <list additional eggs here>

If you do not install zodbupdate together with the necessary software it will report missing classes and not touch your database.

Non-FileStorage configurations

You can configure any storage known to your ZODB installation by providing a ZConfig configuration file (similar to zope.conf). For example you can connect to a ZEO server by providing a config file zeo.conf:

<zeoclient>
    server 127.0.0.1:8100
    storage 1
</zeoclient>

And then running zodbupdate using:

$ zodbupdate -c zeo.conf

Pre-defined rename rules

Rename rules can be defined using an entry point called zodbupdate:

setup(...
      entry_points = """
      [zodbupdate]
      renames = mypackage.mymodule:rename_dict
      """)

These can also be defined in python:

setup(...
      entry_points={
        'zodbupdate': ['renames = mypackage.mymodule:rename_dict'],
      })

Those entry points must points to dictionaries that map old class names to new class names:

rename_dict = {
    'mypackage.mymodule ClassName':
    'otherpackage.othermodule OtherClass'}

As soon as you have rules defined, you can already remove the old import location mentioned in them.

Packing

The option --pack will pack the storage on success. (You tell your users to use that option. If they never pack their storage, it is a good occasion).

Converting to Python 3

zodbupdate can be used to migrate a database created with a Python 2 application to be usable with the same application in Python 3. To accomplish this, you need to:

1. Stop your application. Nothing should be written to the database while the migration is running.

2. Update your Python 2 application to use the latest ZODB version. It will not work with ZODB 3.

3. With Python 2, run zodbupdate --pack --convert-py3.

If you use a Data.fs we recommend you to use the -f option to specify your database. After the conversion the magic header of the database will be updated so that you will be able to open the database with Python 3.

If you use a different storage (like RelStorage), be sure you will be connecting to it using your Python 3 application after the migration. You will still be able to connect to your database and use your application with Python 2 without errors, but then you will need to convert it again to Python 3.

While the pack is not required, it is highly recommended.

The conversion will take care of the following tasks:

• Updating stored Python datetime, date and time objects to use Python 3 bytes,

• Updating ZODB references to use Python 3 bytes.

• Optionally convert stored strings to either unicode or bytes pending your configuration.

If your application expect to use bytes in Python 3, they must be stored as such in the database, and all other strings must be stored as unicode string, if they contain other characters than ascii characters.

When using --convert-py3, zodbupdate will load a set of decoders from the entry points:

setup(...
      entry_points = """
      [zodbupdate.decode]
      decodes = mypackage.mymodule:decode_dict
      """)

Decoders are dictionaries that specifies as keys attributes on Persistent classes that must either be encoded as bytes (if the value is binary) or decoded to unicode using value as encoding (for instance utf-8 here):

decode_dict = {
   'mypackage.mymodule ClassName attribute': 'binary',
   'otherpackage.othermodule OtherClass other_attribute': 'utf-8'}

Please note that for the moment only attributes on Persistent classes are supported.

Please also note that these conversion rules are _only_ selected for the class that is referenced in the pickle, rules for superclasses are _not_ applied. This means that you have to push down annotation rules to all the subclasses of a superclass that has a field that needs this annotation.

Converting to Python 3 from within Python 3

zodbupdate can also be run from within Python 3 to convert a database created with Python 2 to be usable in Python 3. However this works slightly differently than when running the conversion using Python 2. In Python 3 you must specify a default encoding to use while unpickling strings: zodbupdate --pack --convert-py3 --encoding utf-8.

For each string in the database, zodbupdate will convert it as follows:

1. If it’s an attribute configured explicitly via a decoder as described above, it will be decoded or encoded as specified there.

2. Otherwise the value will be decoded using the encoding specified on the command line.

3. If there is an error while decoding using the encoding specified on the command line, the value will be stored as bytes.

Problems and solutions

2.0 (2023-02-09)

• Add support for Python 3.9, 3.10, 3.11.

• Drop support for Python 2.7, 3.5, 3.6.

• Test with history-free and history-preserving RelStorage. Note that history-preserving RelStorage requires RelStorage 3.3 or above, and Python 2.7 or Python 3.6 and above. (#30)

1.5 (2020-07-28)

• Fixed incompatibility with ZODB 5.6 (#35)

• Added support for history-free RelStorage (#28)

• Support zope.interface >= 5 in tests. (issue 32)

1.4 (2019-08-23)

• Fail with explanation when opening a Python 2 ZODB with –dry-run on Python 3 (#22)

1.3 (2019-07-30)

• Support converting sets.Set() objects from ancient Python 2 versions. (issue 23)

• Convert set objects to builtins.set without relying on ZODB.broken.rebuild. (issue 25)

1.2 (2019-05-09)

• Enable fallback encodings for Python 3 conversion for old/grown ZODBs using the new command line option --encoding-fallback. (#15)

• Switch to use argparse as optparse is deprecated.

• Add ability to run the Python 3 migration with a default encoding for str objects. (#14)

• Fix updating records that reference a broken interface when the interface’s top-level module is missing.

• Fixed skipping of blob records so that oids in references to blobs are still converted.

• Add support for Python 3.8a3.

• Drop support for Python 3.4.

1.1 (2018-10-05)

• Skip records for ZODB.blob when migrating database to Python 3 to not break references to blobfiles.

• When migrating databases to Python 3, do not fail when converting attributes containing None.

• Fix tests on Python 2 with ZODB >= 5.4.0, which now uses pickle protocol 3.

• Fix is_broken check for old-style class instances.

• Add support for Python 3.7.

• Drop PyPy support.

1.0 (2018-02-13)

• Support Python 2.7 and 3.4, 3.5 and 3.6 and pypy 3. Drop any older version of Python.

• The option to the select the pickler (--pickler) has been removed. This was only useful if you had extension classes with Python 2.5 or less.

• Added an option to convert a database to Python 3.

0.5 (2010-10-07)

• More debug logging shows now the currently processed OID (that is helpful to determine which object misses the factory).

• Support for missing factories have been improved: an error used to occur if a pickle needed an update and contained a reference to a missing class (not instance of this class). This case is now fixed.

• Python 2.4 is no longer supported. Please stick to version 0.3 if you need Python 2.4 support.

0.4 (2010-07-14)

• Add an option to debug broken records.

• Add an option to skip records.

• Add an option to use Python unPickler instead of C one. This let you debug records. As well Python unPickler let you update old ExtensionClass records who had a special hack in the past.

• Broken interfaces are well supported now (if you did alsoProvides with them).

0.3 (2010-02-02)

• Unplickle and re-pickle the code to rename references to moved classes. This make the script works on database created with older versions of ZODB.

• If you are working directly with a FileStorage, POSKeyError are reported but non-fatal.

• Remove superfluous code that tried to prevent commits when no changes happened: ZODB does this all by itself already.

0.2 (2009-06-23)

• Add option to store the rename rules into a file.

• Don’t commit transactions that have no changes.

• Load rename rules from entry points zodbupdate.

• Compatibility with Python 2.4

• Rename from zodbupgrade to zodbupdate.

• Add ‘verbose’ option.

• Improve logging.

• Suppress duplicate log messages (e.g. if the same class is missing in multiple objects).

• Improve the updating process: rewrite pickle opcodes instead of blindly touching a class. This also allows updating pickles that can’t be unpickled due to missing classes.

0.1 (2009-06-08)

• First release.