neutron/doc/source/devref/alembic_migrations.rst
jeckxie d17befe586 Fix dir doc typo error
Change-Id: I74de73e8f21ce44abbfa293369c774834d105c80
2017-01-06 08:54:46 +08:00

20 KiB

Alembic Migrations

Introduction

The migrations in the alembic/versions contain the changes needed to migrate from older Neutron releases to newer versions. A migration occurs by executing a script that details the changes needed to upgrade the database. The migration scripts are ordered so that multiple scripts can run sequentially to update the database.

The Migration Wrapper

The scripts are executed by Neutron's migration wrapper neutron-db-manage which uses the Alembic library to manage the migration. Pass the --help option to the wrapper for usage information.

The wrapper takes some options followed by some commands:

neutron-db-manage <options> <commands>

The wrapper needs to be provided with the database connection string, which is usually provided in the neutron.conf configuration file in an installation. The wrapper automatically reads from /etc/neutron/neutron.conf if it is present. If the configuration is in a different location:

neutron-db-manage --config-file /path/to/neutron.conf <commands>

Multiple --config-file options can be passed if needed.

Instead of reading the DB connection from the configuration file(s) the --database-connection option can be used:

neutron-db-manage --database-connection mysql+pymysql://root:secret@127.0.0.1/neutron?charset=utf8 <commands>

The branches, current, and history commands all accept a --verbose option, which, when passed, will instruct neutron-db-manage to display more verbose output for the specified command:

neutron-db-manage current --verbose

For some commands the wrapper needs to know the entrypoint of the core plugin for the installation. This can be read from the configuration file(s) or specified using the --core_plugin option:

neutron-db-manage --core_plugin neutron.plugins.ml2.plugin.Ml2Plugin <commands>

When giving examples below of using the wrapper the options will not be shown. It is assumed you will use the options that you need for your environment.

For new deployments you will start with an empty database. You then upgrade to the latest database version via:

neutron-db-manage upgrade heads

For existing deployments the database will already be at some version. To check the current database version:

neutron-db-manage current

After installing a new version of Neutron server, upgrading the database is the same command:

neutron-db-manage upgrade heads

To create a script to run the migration offline:

neutron-db-manage upgrade heads --sql

To run the offline migration between specific migration versions:

neutron-db-manage upgrade <start version>:<end version> --sql

Upgrade the database incrementally:

neutron-db-manage upgrade --delta <# of revs>

NOTE: Database downgrade is not supported.

Migration Branches

Neutron makes use of alembic branches for two purposes.

1. Independent Sub-Project Tables

Various sub-projects can be installed with Neutron. Each sub-project registers its own alembic branch which is responsible for migrating the schemas of the tables owned by the sub-project.

The neutron-db-manage script detects which sub-projects have been installed by enumerating the neutron.db.alembic_migrations entrypoints. For more details see the Entry Points section of Contributing extensions to Neutron.

The neutron-db-manage script runs the given alembic command against all installed sub-projects. (An exception is the revision command, which is discussed in the Developers section below.)

2. Offline/Online Migrations

Since Liberty, Neutron maintains two parallel alembic migration branches.

The first one, called 'expand', is used to store expansion-only migration rules. Those rules are strictly additive and can be applied while neutron-server is running. Examples of additive database schema changes are: creating a new table, adding a new table column, adding a new index, etc.

The second branch, called 'contract', is used to store those migration rules that are not safe to apply while neutron-server is running. Those include: column or table removal, moving data from one part of the database into another (renaming a column, transforming single table into multiple, etc.), introducing or modifying constraints, etc.

The intent of the split is to allow invoking those safe migrations from 'expand' branch while neutron-server is running, reducing downtime needed to upgrade the service.

For more details, see the Expand and Contract Scripts section below.

Developers

A database migration script is required when you submit a change to Neutron or a sub-project that alters the database model definition. The migration script is a special python file that includes code to upgrade the database to match the changes in the model definition. Alembic will execute these scripts in order to provide a linear migration path between revisions. The neutron-db-manage command can be used to generate migration scripts for you to complete. The operations in the template are those supported by the Alembic migration library.

Running neutron-db-manage without devstack

When, as a developer, you want to work with the Neutron DB schema and alembic migrations only, it can be rather tedious to rely on devstack just to get an up-to-date neutron-db-manage installed. This section describes how to work on the schema and migration scripts with just the unit test virtualenv and mysql. You can also operate on a separate test database so you don't mess up the installed Neutron database.

Setting up the environment

Install mysql service

This only needs to be done once since it is a system install. If you have run devstack on your system before, then the mysql service is already installed and you can skip this step.

Mysql must be configured as installed by devstack, and the following script accomplishes this without actually running devstack:

INSTALL_MYSQL_ONLY=True ./tools/configure_for_func_testing.sh ../devstack

Run this from the root of the neutron repo. It assumes an up-to-date clone of the devstack repo is in ../devstack.

Note that you must know the mysql root password. It is derived from (in order of precedence):

  • $MYSQL_PASSWORD in your environment
  • $MYSQL_PASSWORD in ../devstack/local.conf
  • $MYSQL_PASSWORD in ../devstack/localrc
  • default of 'secretmysql' from tools/configure_for_func_testing.sh
Work on a test database

Rather than using the neutron database when working on schema and alembic migration script changes, we can work on a test database. In the examples below, we use a database named testdb.

To create the database:

mysql -e "create database testdb;"

You will often need to clear it to re-run operations from a blank database:

mysql -e "drop database testdb; create database testdb;"

To work on the test database instead of the neutron database, point to it with the --database-connection option:

neutron-db-manage --database-connection mysql+pymysql://root:secretmysql@127.0.0.1/testdb?charset=utf8 <commands>

You may find it convenient to set up an alias (in your .bashrc) for this:

alias test-db-manage='neutron-db-manage --database-connection mysql+pymysql://root:secretmysql@127.0.0.1/testdb?charset=utf8'
Create and activate the virtualenv

From the root of the neutron (or sub-project) repo directory, run:

tox --notest -r -e py27
source .tox/py27/bin/activate

Now you can use the test-db-manage alias in place of neutron-db-manage in the script auto-generation instructions below.

When you are done, exit the virtualenv:

deactivate

Script Auto-generation

This section describes how to auto-generate an alembic migration script for a model change. You may either use the system installed devstack environment, or a virtualenv + testdb environment as described in neutron-db-manage-without-devstack.

Stop the neutron service. Work from the base directory of the neutron (or sub-project) repo. Check out the master branch and do git pull to ensure it is fully up to date. Check out your development branch and rebase to master.

NOTE: Make sure you have not updated the CONTRACT_HEAD or EXPAND_HEAD yet at this point.

Start with an empty database and upgrade to heads:

mysql -e "drop database neutron; create database neutron;"
neutron-db-manage upgrade heads

The database schema is now created without your model changes. The alembic revision --autogenerate command will look for differences between the schema generated by the upgrade command and the schema defined by the models, including your model updates:

neutron-db-manage revision -m "description of revision" --autogenerate

This generates a prepopulated template with the changes needed to match the database state with the models. You should inspect the autogenerated template to ensure that the proper models have been altered. When running the above command you will probably get the following error message:

Multiple heads are present; please specify the head revision on which the
new revision should be based, or perform a merge.

This is alembic telling you that it does not know which branch (contract or expand) to generate the revision for. You must decide, based on whether you are doing contracting or expanding changes to the schema, and provide either the --contract or --expand option. If you have both types of changes, you must run the command twice, once with each option, and then manually edit the generated revision scripts to separate the migration operations.

In rare circumstances, you may want to start with an empty migration template and manually author the changes necessary for an upgrade. You can create a blank file for a branch via:

neutron-db-manage revision -m "description of revision" --expand
neutron-db-manage revision -m "description of revision" --contract

NOTE: If you use above command you should check that migration is created in a directory that is named as current release. If not, please raise the issue with the development team (IRC, mailing list, launchpad bug).

NOTE: The "description of revision" text should be a simple English sentence. The first 30 characters of the description will be used in the file name for the script, with underscores substituted for spaces. If the truncation occurs at an awkward point in the description, you can modify the script file name manually before committing.

The timeline on each alembic branch should remain linear and not interleave with other branches, so that there is a clear path when upgrading. To verify that alembic branches maintain linear timelines, you can run this command:

neutron-db-manage check_migration

If this command reports an error, you can troubleshoot by showing the migration timelines using the history command:

neutron-db-manage history

Expand and Contract Scripts

The obsolete "branchless" design of a migration script included that it indicates a specific "version" of the schema, and includes directives that apply all necessary changes to the database at once. If we look for example at the script 2d2a8a565438_hierarchical_binding.py, we will see:

# .../alembic_migrations/versions/2d2a8a565438_hierarchical_binding.py

def upgrade():

    # .. inspection code ...

    op.create_table(
        'ml2_port_binding_levels',
        sa.Column('port_id', sa.String(length=36), nullable=False),
        sa.Column('host', sa.String(length=255), nullable=False),
        # ... more columns ...
    )

    for table in port_binding_tables:
        op.execute((
            "INSERT INTO ml2_port_binding_levels "
            "SELECT port_id, host, 0 AS level, driver, segment AS segment_id "
            "FROM %s "
            "WHERE host <> '' "
            "AND driver <> '';"
        ) % table)

    op.drop_constraint(fk_name_dvr[0], 'ml2_dvr_port_bindings', 'foreignkey')
    op.drop_column('ml2_dvr_port_bindings', 'cap_port_filter')
    op.drop_column('ml2_dvr_port_bindings', 'segment')
    op.drop_column('ml2_dvr_port_bindings', 'driver')

    # ... more DROP instructions ...

The above script contains directives that are both under the "expand" and "contract" categories, as well as some data migrations. the op.create_table directive is an "expand"; it may be run safely while the old version of the application still runs, as the old code simply doesn't look for this table. The op.drop_constraint and op.drop_column directives are "contract" directives (the drop column more so than the drop constraint); running at least the op.drop_column directives means that the old version of the application will fail, as it will attempt to access these columns which no longer exist.

The data migrations in this script are adding new rows to the newly added ml2_port_binding_levels table.

Under the new migration script directory structure, the above script would be stated as two scripts; an "expand" and a "contract" script:

# expansion operations
# .../alembic_migrations/versions/liberty/expand/2bde560fc638_hierarchical_binding.py

def upgrade():

    op.create_table(
        'ml2_port_binding_levels',
        sa.Column('port_id', sa.String(length=36), nullable=False),
        sa.Column('host', sa.String(length=255), nullable=False),
        # ... more columns ...
    )


# contraction operations
# .../alembic_migrations/versions/liberty/contract/4405aedc050e_hierarchical_binding.py

def upgrade():

    for table in port_binding_tables:
        op.execute((
            "INSERT INTO ml2_port_binding_levels "
            "SELECT port_id, host, 0 AS level, driver, segment AS segment_id "
            "FROM %s "
            "WHERE host <> '' "
            "AND driver <> '';"
        ) % table)

    op.drop_constraint(fk_name_dvr[0], 'ml2_dvr_port_bindings', 'foreignkey')
    op.drop_column('ml2_dvr_port_bindings', 'cap_port_filter')
    op.drop_column('ml2_dvr_port_bindings', 'segment')
    op.drop_column('ml2_dvr_port_bindings', 'driver')

    # ... more DROP instructions ...

The two scripts would be present in different subdirectories and also part of entirely separate versioning streams. The "expand" operations are in the "expand" script, and the "contract" operations are in the "contract" script.

For the time being, data migration rules also belong to contract branch. There is expectation that eventually live data migrations move into middleware that will be aware about different database schema elements to converge on, but Neutron is still not there.

Scripts that contain only expansion or contraction rules do not require a split into two parts.

If a contraction script depends on a script from expansion stream, the following directive should be added in the contraction script:

depends_on = ('<expansion-revision>',)

Expand and Contract Branch Exceptions

In some cases, we have to have "expand" operations in contract migrations. For example, table 'networksegments' was renamed in contract migration, so all operations with this table are required to be in contract branch as well. For such cases, we use the contract_creation_exceptions that should be implemented as part of such migrations. This is needed to get functional tests pass.

Usage:

def contract_creation_exceptions():
    """Docstring should explain why we allow such exception for contract
    branch.
    """
    return {
        sqlalchemy_obj_type: ['name']
        # For example: sa.Column: ['subnets.segment_id']
    }

HEAD files for conflict management

In directory neutron/db/migration/alembic_migrations/versions there are two files, CONTRACT_HEAD and EXPAND_HEAD. These files contain the ID of the head revision in each branch. The purpose of these files is to validate the revision timelines and prevent non-linear changes from entering the merge queue.

When you create a new migration script by neutron-db-manage these files will be updated automatically. But if another migration script is merged while your change is under review, you will need to resolve the conflict manually by changing the down_revision in your migration script.

Applying database migration rules

To apply just expansion rules, execute:

neutron-db-manage upgrade --expand

After the first step is done, you can stop neutron-server, apply remaining non-expansive migration rules, if any:

neutron-db-manage upgrade --contract

and finally, start your neutron-server again.

If you have multiple neutron-server instances in your cloud, and there are pending contract scripts not applied to the database, full shutdown of all those services is required before 'upgrade --contract' is executed. You can determine whether there are any pending contract scripts by checking the message returned from the following command:

neutron-db-manage has_offline_migrations

If you are not interested in applying safe migration rules while the service is running, you can still upgrade database the old way, by stopping the service, and then applying all available rules:

neutron-db-manage upgrade head[s]

It will apply all the rules from both the expand and the contract branches, in proper order.

Tagging milestone revisions

When named release (liberty, mitaka, etc.) is done for neutron or a sub-project, the alembic revision scripts at the head of each branch for that release must be tagged. This is referred to as a milestone revision tag.

For example, here is a patch that tags the liberty milestone revisions for the neutron-fwaas sub-project. Note that each branch (expand and contract) is tagged.

Tagging milestones allows neutron-db-manage to upgrade the schema to a milestone release, e.g.:

neutron-db-manage upgrade liberty

Generation of comparable metadata with current database schema

Directory neutron/db/migration/models contains module head.py, which provides all database models at current HEAD. Its purpose is to create comparable metadata with the current database schema. The database schema is generated by alembic migration scripts. The models must match, and this is verified by a model-migration sync test in Neutron's functional test suite. That test requires all modules containing DB models to be imported by head.py in order to make a complete comparison.

When adding new database models, developers must update this module, otherwise the change will fail to merge.