Document how to use pgloader

1 files changed, 88 insertions, 55 deletions

diff --git a/doc/update/mysql_to_postgresql.md b/doc/update/mysql_to_postgresql.md
index 5dc8e6f65f8..a3eba38c91a 100644
--- a/doc/update/mysql_to_postgresql.md
+++ b/doc/update/mysql_to_postgresql.md
@@ -1,80 +1,113 @@
-*** NOTE: These instructions should be considered deprecated. In GitLab 10.0 we will be releasing new migration instructions using [pgloader](http://pgloader.io/).
+---
+last_updated: 2017-10-05
+---
 
-# Migrating GitLab from MySQL to Postgres
-*Make sure you view this [guide from the `master` branch](https://gitlab.com/gitlab-org/gitlab-ce/blob/master/doc/update/mysql_to_postgresql.md#migrating-gitlab-from-mysql-to-postgres) for the most up to date instructions.*
+# Migrating from MySQL to PostgreSQL
 
-If you are replacing MySQL with Postgres while keeping GitLab on the same server all you need to do is to export from MySQL, convert the resulting SQL file, and import it into Postgres. If you are also moving GitLab to another server, or if you are switching to omnibus-gitlab, you may want to use a GitLab backup file. The second part of this documents explains the procedure to do this.
+> **Note:** This guide assumes you have a working Omnibus GitLab instance with
+> MySQL and want to migrate to bundled PostgreSQL database.
 
-## Export from MySQL and import into Postgres
+## Prerequisites
 
-Use this if you are keeping GitLab on the same server.
+First, we'll need to enable the bundled PostgreSQL database with up-to-date
+schema. Next, we'll use [pgloader](http://pgloader.io) to migrate the data
+from the old MySQL database to the new PostgreSQL one.
 
-```
-sudo service gitlab stop
+Here's what you'll need to have installed:
 
-# Update /home/git/gitlab/config/database.yml
+- pgloader 3.4.1+
+- Omnibus GitLab
+- MySQL
 
-git clone https://github.com/gitlabhq/mysql-postgresql-converter.git -b gitlab
-cd mysql-postgresql-converter
-mysqldump --compatible=postgresql --default-character-set=utf8 -r gitlabhq_production.mysql -u root gitlabhq_production -p
-python db_converter.py gitlabhq_production.mysql gitlabhq_production.psql
-ed -s gitlabhq_production.psql < move_drop_indexes.ed
+## Enable bundled PostgreSQL database
 
-# Import the database dump as the application database user
-sudo -u git psql -f gitlabhq_production.psql -d gitlabhq_production
+1. Stop GitLab:
 
-# Install gems for PostgreSQL (note: the line below states '--without ... mysql')
-sudo -u git -H bundle install --without development test mysql --deployment
+    ``` bash
+    sudo gitlab-ctl stop
+    ```
 
-sudo service gitlab start
-```
+1. Edit `/etc/gitlab/gitlab.rb` to enable bundled PostgreSQL:
 
-## Converting a GitLab backup file from MySQL to Postgres
-**Note:** Please make sure to have Python 2.7.x (or higher) installed.
+    ```
+    postgresql['enable'] = true
+    ```
 
-GitLab backup files (`<timestamp>_gitlab_backup.tar`) contain a SQL dump. Using the lanyrd database converter we can replace a MySQL database dump inside the tar file with a Postgres database dump. This can be useful if you are moving to another server.
+1. Edit `/etc/gitlab/gitlab.rb` to use the bundled PostgreSQL. Please check
+   all the settings beginning with `db_`, such as `gitlab_rails['db_adapter']`
+   and alike. You could just comment all of them out so that we'll just use
+   the defaults.
 
-```
-# Stop GitLab
-sudo service gitlab stop
+1. [Reconfigure GitLab] for the changes to take effect:
+
+    ``` bash
+    sudo gitlab-ctl reconfigure
+    ```
+
+1. Start Unicorn and PostgreSQL so that we could prepare the schema:
+
+    ``` bash
+    sudo gitlab-ctl start unicorn
+    sudo gitlab-ctl start posgresql
+    ```
+
+1. Run the following commands to prepare the schema:
+
+    ``` bash
+    sudo gitlab-rake db:create db:migrate
+    ```
+
+1. Stop Unicorn in case it's interfering the next step:
 
-# Create the backup
-cd /home/git/gitlab
-sudo -u git -H bundle exec rake gitlab:backup:create RAILS_ENV=production
+    ``` bash
+    sudo gitlab-ctl stop unicorn
+    ```
 
-# Note the filename of the backup that was created. We will call it
-# TIMESTAMP_gitlab_backup.tar below.
+After these steps, you'll have a fresh PostgreSQL database with up-to-date schema.
 
-# Move the backup file we will convert to its own directory
-sudo -u git -H mkdir -p tmp/backups/postgresql
-sudo -u git -H mv tmp/backups/TIMESTAMP_gitlab_backup.tar tmp/backups/postgresql/
+## Migrate data from MySQL to PostgreSQL
 
-# Create a separate database dump with PostgreSQL compatibility
-cd tmp/backups/postgresql
-sudo -u git -H mysqldump --compatible=postgresql --default-character-set=utf8 -r gitlabhq_production.mysql -u root gitlabhq_production -p
+Now, you can use pgloader to migrate the data from MySQL to PostgreSQL:
 
-# Clone the database converter
-sudo -u git -H git clone https://github.com/gitlabhq/mysql-postgresql-converter.git -b gitlab
+1. Save the following snippet in a `commands.load` file, and edit with your
+   database `username`, `password` and `host`:
 
-# Convert gitlabhq_production.mysql
-sudo -u git -H mkdir db
-sudo -u git -H python mysql-postgresql-converter/db_converter.py gitlabhq_production.mysql db/database.sql
-sudo -u git -H ed -s db/database.sql < mysql-postgresql-converter/move_drop_indexes.ed
+    ```
+    LOAD DATABASE
+         FROM mysql://username:password@host/gitlabhq_production
+         INTO postgresql://gitlab-psql@unix://var/opt/gitlab/postgresql:/gitlabhq_production
 
-# Compress database backup
-# Warning: If you have Gitlab 7.12.0 or older skip this step and import the database.sql directly into the backup with:
-# sudo -u git -H tar rf TIMESTAMP_gitlab_backup.tar db/database.sql
-# The compressed databasedump is not supported at 7.12.0 and older.
-sudo -u git -H gzip db/database.sql
+    WITH include no drop, truncate, disable triggers, create no tables,
+         create no indexes, preserve index names, no foreign keys,
+         data only
 
-# Replace the MySQL dump in TIMESTAMP_gitlab_backup.tar.
+    ALTER SCHEMA 'gitlabhq_production' RENAME TO 'public'
 
-# Warning: if you forget to replace TIMESTAMP below, tar will create a new file
-# 'TIMESTAMP_gitlab_backup.tar' without giving an error.
+    ;
+    ```
 
-sudo -u git -H tar rf TIMESTAMP_gitlab_backup.tar db/database.sql.gz
+1. Start the migration:
 
-# Done! TIMESTAMP_gitlab_backup.tar can now be restored into a Postgres GitLab
-# installation.
-# See https://gitlab.com/gitlab-org/gitlab-ce/blob/master/doc/raketasks/backup_restore.md for more information about backups.
+    ``` bash
+    sudo -u gitlab-psql pgloader commands.load
+    ```
+
+1. Once the migration finishes, start GitLab:
+
+    ``` bash
+    sudo gitlab-ctl start
+    ```
+
+Now, you can verify that everything worked by visiting GitLab.
+
+## Troubleshooting
+
+### Experiencing 500 errors after the migration
+
+If you experience 500 errors after the migration, try to clear the cache:
+
+``` bash
+sudo gitlab-rake cache:clear
 ```
+
+[reconfigure GitLab]: ../administration/restart_gitlab.md#omnibus-gitlab-reconfigure