Schema Evolution Manager (sem)

Intended Audience

• Engineers who regularly manage the creation of scripts to update the schema in a postgresql database.

• Engineers who want to simplify and/or standardize how other team members contribute schema changes to a postgresql database.

Purpose

Schema Evolution Manager (sem) makes it very simple for engineers to contribute schema changes to a postgresql database, managing the schema evolutions as proper source code. Schema changes are deployed as gzipped tarballs named with the corresponding git tag.

To apply schema changes to a particular database, download a tarball and use sem to figure out which scripts have not yet been applied, then apply those scripts in chronological order.

sem provides well tested, simple tools to manage the process of creating and applying schema upgrade scripts to databases in all environments.

• scripts are automatically named with a timestamp assigned at time of creation

• all scripts applied to the postgresql database are recorded in the table schema_evolution_manager.scripts - making it simple to see what has been applied, and when, if needed.

sem contains only tools for managing schema evolutions. The idea is that you create one git repository for each of your databases then use sem to manage the schema evolution of each database.

At Gilt Groupe, we started using sem in early 2012 and have observed an increase in the reliability of our production schema deploys across dozens of independent postgresql databases.

See INSTALLATION and GETTING STARTED for details.

Project Goals

• Absolutely minimal set of dependencies. We found that anything more complex led developers to prefer to manage their own schema evolutions. We prefer small sets of scripts that each do one thing well.

• Committed to true simplicity - features that would add complexity are not added. We hope that more advanced features might be built on top of schema evolution manager.

• Works for ALL applications - schema management is a first class task now so any application framework can leverage these migration tools.

• No rollback. We have found in practice that rolling back schema changes is not 100% reliable. Therefore we inentionally do NOT support rollback. This is an often debated element of sem, and although the design itself could be easily extended to support rollback, we currently have no plans to do so.

In place of rollback, we prefer to keep focus on the criticalness of schema changes, encouraging peer review and lots of smaller evolutions that themselves are relatively harmless.

This stems from the idea that we believe schema evolutions are fundamentally risky. We believe the best way to manage this risk is to:

1. Treat schema evolution changes as normal software releases as much as possible

2. Manage schema versions as simple tarballs - artifacts are critical to provide 100% reproducibility. This means the exact same artifacts can be applied in development then QA and finally production environments.

3. Isolate schema changes as their own deploy. This then guarantees that every other application itself can be rolled back if needed. In practice, we have seen greater risk when applications couple code changes with schema changes.

This last point bears some more detail. By fundamentally deciding to manage and release schema changes independent of application changes:

1. Schema changes are required to be incremental. For example, to rename a column takes 4 separate, independent production deploys:

a. add new column
b. deploy changes in application to use old and new column
c. remove old column
d. deploy changes in application to use only new column

Though at first this may seem more complex, each individual change itself is smaller and lower risk.

2. It is worth repeating that all application deploys can now be rolled back. This has been a huge win for our teams.

Talks

First presented at PGDay NYC 2013

Dependencies

• Ruby: Current testing against ruby 2.x. 1.8 and 1.9 are supported.

• Postgres: Only tested against 9.x. We minimize use of advanced features and should work against 8.x series. If you try 8.x and run into problems, please let us know so we can update.

• plpgsql must be available in the database. If needed you can:

  createlang plpgsql template1

• Git: Designed to use git for history (all versions since 1.7).

Installation

There are two ways to install schema evolution manager:

1. Direct install using ruby (no dependency on ruby gems)

    git clone git://github.com/mbryzek/schema-evolution-manager.git
    cd schema-evolution-manager
    git checkout 0.9.41
    ruby ./configure.rb
    sudo ./install.rb
   

2. If you have ruby gems:

    gem install schema-evolution-manager
   

Upgrading

Upgrading is as simple as following the Installation instructions for the new version. Each installation of sem will create a new directory for that specific version. When you install the newer version, a new directory will be created and symlinks updated to point to the latest version.

Getting Started

Initialization

git init /tmp/sample
sem-init --dir /tmp/sample --url postgresql://postgres@localhost/sample

Writing your first sql script

cd /tmp/sample
echo "create table tmp_table (id integer)" > new.sql
sem-add ./new.sql

Applying changes to your local database:

cd /tmp/sample
createdb sample
sem-apply --url postgresql://postgres@localhost/sample

Note that you can also pass in the username, db host, and db name explicitly:

sem-apply --host localhost --name sample --user postgres

Similarly, for non-standard setups, you can optionally pass in the port

sem-apply --host localhost --port 5433 --name sample --user postgres

When you are happy with your change, commit:

git commit -m "Adding a new tmp table to test sem process" scripts

Publishing a Release

cd /tmp/sample
sem-dist

By default, the sem-dist script will create the next micro git tag, and use that tag in the file name.

If you already have a tag:

sem-dist --tag 0.0.2

You will now have a single artifict - /tmp/sample/dist/sample-0.0.2.tar.gz - that you can manage in standard deploy process.

Deploying Schema Changes

Extract tarball on server

scp /tmp/sample/dist/sample-0.0.2.tar.gz <your server>:~/
ssh <your server>
tar xfz sample-0.0.2.tar.gz
cd sample-0.0.2

Do a dry run

sem-apply --url postgresql://postgres@localhost/sample --dry_run

You will likely see a number of create table statements (see data model section below). You should also see:

  [DRY RUN] Applying 20130318-214407.sql

which tells you that if you apply these changes, that sql script will be applied to the sample db

Specifying database password

There are two recommended ways in which to pass user passwords to psql:

1. Create a ~/.pgpass file with the appropriate credentials

2. Specify a [--password] flag when running sem-apply. You will then be prompted to enter your password once. sem will create a temporary file to store your password, using that file during the duration of the command and ensuring the file is deleted after sem completed.

   Example:

   sem-apply --url postgresql://postgres@localhost/sample --password
   

Apply the changes

sem-apply --url postgresql://postgres@localhost/sample

You will see:

  Upgrading schema for postgres@localhost/sample
  Applying 20130318-214407.sql

Attempt to apply again:

sem-apply --url postgresql://postgres@localhost/sample

You will see:

  Upgrading schema for postgres@localhost/sample
    All scripts have been previously applied

Baselines

If you have an existing database, and you want to start using schema evolution manager, we support the notion of creating a baseline. The sem-baseline script will record that all of the scripts have been applied to the database, without actually applying them. From this point forward, only new scripts will be applied to the database.

sem-baseline --url postgresql://postgres@localhost/sample

Data Model

sem will create a new postgresql schema in your database named 'schema_evolution_manager'

psql sample
set search_path to schema_evolution_manager;
\dt

    Schema    |       Name        | Type  |  Owner
 -------------+-------------------+-------+----------
 schema_evolution_manager | bootstrap_scripts | table | postgres
 schema_evolution_manager | scripts           | table | postgres

Each of these tables has a column named 'filename' which keeps track of the sql files applied to each database.

• The scripts table is used for your application.
• The bootstrap_scripts table is used to manage upgrades to the sem application itself.

For details on these tables, see scripts/*sql where the tables themselves are defined.

PLPGSQL Utilities

We've included a copy of the schema conventions we practice at Gilt Groupe. There are also a number of utility plpgsql functions to help developers apply these conventions in a systematic way.

The helpers are defined in

scripts/20130318-105456.sql

We have found these utilities incredibly useful - and are committed to providing only the most relevant, high quality, and extremely clear helpers as possible.

In CONVENTIONS.md you will find a simple example of these conventions and utilities in practice.

Command Line Utilities

• sem-init: Initialize a git repository for sem support
• sem-add: Adds a database upgrade script
• sem-dist: Create a distribution tar.gz file containing schema upgrade scripts
• sem-apply: Apply any deltas from a distribution tarball to a particular database
• sem-baseline: Add any migration scripts to the schema tables without actually applying them. See Migrating

Attributes supported in sql migration scripts

Sometimes you may want to adjust the specific options used by SEM when applying SQL scripts. Attributes can be specified within each SQL file in comments.

To specify an attribute, add a comment of the following format anywhere in your SQL file (but at the top by convention):

-- sem.attribute.[name] = [value]

Currently supported attributes:

• transaction

  • single (default): the entire file is applied within a transaction (by using the psql command line argument --single-transaction)
  • none: Each command in the file will be applied in order. If a later command in the file fails, there will be no rollback.

  Examples:

  • -- sem.attribute.transaction = none
  • -- sem.attribute.transaction = single

Migrating

In some cases you may be migrating from no schema evolutions, or another schema evolution model.

For these cases, sem provides a 'baseline' command.

Current workflow:

1. sem-add your current schema 1. Either via a database dump 1. Or by sem-adding existing DB scripts
2. Use sem-baseline to bootstrap the sem tables and add existing schema files to sem's migration table without actually applying them

License

Copyright 2013-2016 Gilt Groupe, Inc.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.