anybox.migration.openerp 0.2

Fast OpenERP migration tool

Warning

Reminder This tool is complex and highly experimental and is distributed in the hope that it might be useful, but WITHOUT ANY WARRANTY of FITNESS FOR A PARTICULAR PURPOSE. In particular, if you’re using OpenERP for your company, you should consider purchasing an OpenERP Enterprise contract that will provide you with the best and riskless migration path for your database and custom developments.

This tool has been developped with these initial goals in mind, in this priority order:

• Merging 2 different OpenERP databases into a single multicompany db

• Migrating from OpenERP 6.1 to 7.0

• Migrating data from a legacy business application (Access, Delphi, etc.)

• Migrating from Dolibarr to OpenERP

The principle of this tool is to export CSV data from an old application (only OpenERP for now), then to process CSV files in order to import them into a freshly installed OpenERP database. This is a completely different strategy than the in-place migration of OpenERP or OpenUpgrade. It also requires some treatment after the migration, such as recreating internal sequences. Import and export are done with the PostgreSQL-specific COPY command, and results in extremely fast exports and imports. Combined with a pure in-memory Python csv processing, this tool can often achieve overall migration rates over 1000 lines/sec.

Installation

This tool only works with Python 2.7!

You can install this tool in a virtualenv or buildout:

$ virtualenv sandbox
$ sandbox/bin/pip install anybox.migration.openerp

Usage

This tool offers a single migrate script:

$ sandbox/bin/migrate -h

You can list the available default mapping files:

$ sandbox/bin/migrate -l
openerp6.1-openerp7.0.yml

You should specify the source and target DBs, a selection of the source tables to migrate, and the mapping files to use. The tool then takes care of selecting the dependant tables:

$ sandbox/bin/migrate -s source_dbname -t target_dbname -r res_partner account_move -p openerp6.1-openerp7.0.yml custom.yml

If you want to inspect the temporary CSV files created, use the --keepcsv option. They will be stored in a temporary directory under the current directory.

This script won’t actually write anything in the target database unless you specify the -w option to commit the transaction at the end.

The most important part of the migration process is the YML mapping file, which describes how to handle data, table by table and column by column. A default mapping file is provided and is being used as a real mapping for a migration consisting in migrating two 6.1 databases into a single 7.0 multicompany database. You can mix the default 6.1 to 7.0 file provided, and augment it with other custom yml files, they will be merged.

Internals

This tool was very loosely inspired from:

• the external_referential OpenERP module

• the OpenUpgrade project

• Talend Open Studio

The different internal steps are:

• Exporting CSV from the old database

• Transforming CSV to match the target database

• Detect data existing in the target DB with discriminators

• Postprocessing CSV files to fix foreign keys

• Reinjecting into OpenERP

• Updating possible pre-existing data with incoming data

The processing of CSV files is done using a mapping file written in Yaml.

Mapping file

You should keep in mind that this migration tool is only dealing with database tables and columns: the OpenERP fields are unknown to it. Each table, each line, each cell of the source database is handled independently and the mapping file tells what to do with the current cell. This leads to limitations and this tool won’t be able to handle extremely complex migration. But it is powerful enough to allow to simultaneously merge and migrate two 6.1 databases into a 7.0 multicompany database.

For a real-life example, you can have a look at the OpenERP 6.1 to 7.0 mapping file provided in the mappings directory of this tool.

Copying data

The most simple and basic YML statement for a column mapping is the following:

module:
    table1.column1:
        table2.column2: __copy__

It tells that, if the OpenERP module is installed in the target database, the column1 of the table1 from the source DB should be copied to the column2 of the table2 in the target DB.

The __copy__ instruction can even be omitted and the previous statement is equivalent to this one:

module:
    table1.column1:
        table2.column2:

Internally, this statement is actually converted to a Python dict:

{'module':
    {'table1.column1':
        {'table2.column2': '__copy__'}}

And the whole yml file is converted to a large mapping dict whose leafs are statements or functions which are able to process data.

Copying all columns of a table

If your target table has the same structure as the source table, you can avoid specifying one mapping statement for each column and use a wildcard:

module:
    table1.*:

It means: copy all the columns of table1 from the source db to table1 in the target db. This kind of mapping is often used as a starting point when source and table structures are similar. You can then add mapping statements for specific columns to override this wildcard.

Copying all columns to a different table

If the source table has only been renamed, you can copy all the columns of the source table1 to the target table2:

module:
    table1.*:
        table2.*:

Copying everything

If the source and target db have exactly the same structure and you just want to transfer data, you may use a global wildcard (but we have not had the opportunity to try this one for real yet):

module:
    .*:

It means: copy all tables to the target database without processing. It may seem unuseful compared to a bare dump and restore, but remind that this way you can append data to the target DB, not only replace it. In that case you should take care of existing data, if the table has constraints (see discriminators below)

Splitting one source line to several tables

For a single source line coming from a source table, you can feed data in several target tables. This can be done just by putting several target lines like this:

module:
    table1.column1:
        table2.column2:
        table3.column3:

It means: for each column1 in the table1 of the source DB, create two target lines: one for table2 and one for table3.

During the processing of the current line, other mapping statements can feed the same target lines. Take this example:

module:
    table1.column1:
        table2.column2:
        table3.column3:
    table1.column2:
        table2.column2:
        table3.column4:

In this case, data in the table1 will be directed to table2 and table3. You can then add more lines to handle all the columns of table1

However in the example above, there is a conflict since two source cells are directed to the same target cell (table2.column2). In this scenario, there is no way to predict which one will be used (because the mapping is a Python dict and a dict is not ordered). You should avoid this kind of conflicts.

In case of an OpenERP 6.1 to 7.0 migration, this kind of mapping is actually used to migrate one source res_users line to three different lines: one in res_users + one in res_partner + one in mail_alias. See the default mapping for a real example.

Not migrating a column

If you want to get rid of a specific column in a table, use the __forget__ statement:

module:
    table1.column1: __forget__

This statement is useful if you defined a wildcard, to prevent from migrating a specific column.

Transforming data with Python code

Instead of just copying data with the __copy__ statement, you can use any Python code. The Python code should be written in a literal Yaml block and is executed as is, as a function body, so that you have to insert a return statement somewhere.

Example from the mail module:

mail:
    mail_message.type:
        mail_message.type: return 'email'

It means the type column of the mail_message table will be filled with 'email' strings, whatever data the source column had.

The eventual signature of the function constructed using the Python code block is

def mapping_function(self, source_row, target_rows):

It means that in the function body you can access the full source_row, which is a dict containing all the keys (column names) and values of the current line being processed. But keep in mind that at this time, you are dealing with one specific cell of this line, and you should return the value that will be inserted in the corresponding cell of the target table. This can be used to aggregate data from two source cells into a target cell:

base:
    table1.firstname: __forget__
    table1.name:
        table1.name: return source_row['firstname'] + ' ' + source_row['name']

You can also access the target_rows beeing filled during the processing of the line, so that data coming from a source cell can influence several cells in the target lines, or even different target tables. Here is an example:

base:
    table1.id:
        table1.id:
        table2.id:
    table1.name:
        table1.name: |
            name = source_row['firstname'] + ' ' + source_row['name']
            target_rows['table1']['display_name'] = name
            target_rows['table2']['display_name'] = name
            return name
        table2.name

Note that in the example above, the Python code spans on several lines, and you should define a Yaml literal block using |. The example above eventually means: append firstname to name coming from the table1, and put it in the display_name cell of the target table1 and table2. The target name cell will contain a copy of the source name cell.

If the target line is not supposed to have the same id as the source line, you can create a new id with the newid() function. This function returns a different value at each call and is responsible of incrementing the id. Here is an example:

base:
    res_users.id:
        res_users.id:
        res_users.partner_id:
        res_partner.notification_email_send: return 'comment'
        res_partner.id: |
            i = newid()
            target_rows['res_users']['partner_id'] = i
            target_rows['res_partner']['id'] = i
            target_rows['res_partner']['name'] = source_row['name']
            target_rows['res_partner']['email'] = source_row['user_email']
            return i

Each res_users line will generate a new res_partner line with a new id, while the res_users id will be the same as the source. (Actually it will not be the same, because an offset is applied to all ids).

Feeding a new column

If a target column should contain data but has no equivalent in the source table, you can use ‘_’ as a substitute to the not existing source column name:

base:
    res_partner._:
        res_partner.is_company: return False

Merging with existing data

When data is inserted in the target table, you may want to merge it with existing data.

Imagine the target res_users table already contains an admin account, and you don’t want to duplicate this account by migrating data from the source res_users table. In this case you should tell the mapping how to recognize existing data. This is done by replacing the source column name with the __discriminator__ statement, and by providing a list of column names that will be used to recognize existing data:

base:
    res_users.__discriminator__:
        - login

Using this statement, you can install a new OpenERP database with its admin account, and merge all existing accounts with data coming from the source table. The login column will be used to match data. The preexisting admin account won’t be duplicated but will be updated with the admin account from the source table.

Another use case in a multicompany scenario is to merge partners existing in the target database, but keep them separate for the two companies:

base:
    res_partner.__discriminator__:
        - name
        - company_id

Foreign keys without constraints

The first step of the migration is to automatically detect all the foreign keys of the source and target tables. Sometimes, OpenERP defines foreign keys without constraints. This mainly happens with related fields with store=True, which create a column of integers without constraints. If you don’t want to __forget__ such columns, you have to tell the mapping what the target of the foreign key is, like in the real example below:

account:
    account_move.company_id:
        account_move.company_id: __fk__ res_company

Handle cyclic dependant tables

During the last step, the migrated CSV files are imported one by one. Some tables depend on other tables through foreign key constraints, and such dependencies sometimes happen to be cyclic. In that case, there is no way to import tables because they all depend on another one. One solution is to __forget__ the column, which is rarely desirable because you lose data. To be able to keep such data, you should use the __defer__ statement, so that the column will be updated after all the data is imported:

base:
    res_users.create_uid:
        res_users.create_uid: __defer__
    res_users.write_uid:
        res_users.write_uid: __defer__

Understanding errors

The most difficult part of using this tool is to understand the errors during the processing, as it requires a deep knowledge of how it internally works. Most errors generally come from an erroneous mapping file. Errors can happen during the processing of the CSV files, but the most difficult ones come from the last import step, because some tables may fail to be imported. In this case, you should carefully look at the logging messages at the end, and try to understand the constraint errors or why tables cannot be imported. You also should use the --keepcsv option, and inspect the intermediate CSV files to understand the problem. By using this option, you will end up with a directory containing five CSV files for each table.

For instance, for the res_partner table you will find these files:

• res_partner.csv is the original data exported from the source database

• res_partner.target.csv contains data after the first processing with the mapping file, but wrong foreign keys

• res_partner.target2.csv contains final data with fixed foreign keys, that will eventually be imported at the end

• res_partner.update.csv contains data which have been detected as existing in the target database, with wrong foreign keys.

• res_partner.update2.csv contains the final existing data with fixed foreign keys, that will be used to update the target table after import.

Contribute

Authors and contributors:

• Christophe Combelles

• Florent Jouatte

• Guy-Clovis Nzouendjou

Code

• Code repository and bug tracker: https://bitbucket.org/anybox/anybox.migration.openerp

Please don’t hesitate to give us feedback, report bugs or contribute the mapping files on Bitbucket.

Changes

0.2 (2013-07-01)

• initial release