This gem provides a way is alter database schemas on large tables with little downtime. You run 2 commands to ultimately alter the database.
First, you generate the schema transform definitions and commands to be ran later on production. You will check these files into the rails project.
Second, you run 2 commands on production.
The first command will create a 'temporary' table with the altered schema and incrementally copy the data over until it is close to synced. You can run this command as many times as you want as it want - it work hurt. This first command is slow as it takes a while to copy the data over, especially if you have a really large tables that are several GBs in size.
The second command will do a switcheroo with with 'temporarily' new table and the current table. It will then remove the obsoleted table with the old schema structure. Because it is doing a rename (which can screw up replication on a heavily traffic site), this second command should be ran with maintenance page up. This second command is fast because it does a final incremental sync and quickly switches the new table into place.
gem install --no-ri --no-rdoc schema_transformer # sudo if you need to
Generate the schema transform definitions:
tung@walle $ schema_transformer generate What is the name of the table you want to alter? > tags What is the modification to the table? Examples 1: ADD COLUMN smart tinyint(1) DEFAULT '0' Examples 2: ADD INDEX idx_name (name) Examples 3: ADD COLUMN smart tinyint(1) DEFAULT '0', DROP COLUMN full_name > ADD COLUMN special tinyint(1) DEFAULT '0' *** Thanks *** Schema transform definitions have been generated and saved to: config/schema_transformations/tags.json Next you need to run 2 commands to alter the database. As explained in the README, the first can be ran with the site still up. The second command should be done with a maintenance page up. Here are the 2 commands you'll need to run later after checking in the tags.json file into your version control system: $ schema_transformer sync tags # can be ran over and over, it will just keep syncing the data $ schema_transformer switch tags # should be done with a maintenance page up, switches the tables *** Thank you *** tung@walle $ schema_transformer sync tags Creating temp table and syncing the data... (tail log/schema_transformer.log for status) *** Thanks *** There is now a tags_st_temp table with the new table schema and the data has been synced. Please run the next command after you put a maintenance page up: $ schema_transformer switch tags tung@walle $ schema_transformer switch tags *** Thanks *** The final sync ran and the table tags has been updated with the new schema. Get rid of that maintenance page and re-enable your site. Thank you. Have a very nice day. tung@walle $
It is strongly recommended that the tables that you are altering has updated_at timestamp columns with indexes on them. If the tables do not updated_at columns at all, then only the last 100,000 rows get updated in the final sync in the "schema_transformer switch ..." command. If the updated_at column is available then the final sync will use it to update all the data. However, because it uses the updated_at column, it is extremly important that the updated_at column is indexed or the final "schema_transformer switch ..." command possibly could be slow. Because of this, you should analyze your database schema for missing updated_at columns and indexes with the command "schema_transformer analyze".
tung@walle $ schema_transformer analyze Analyzing your database schema... There are no tables without the updated_at timestamp. GOOD These tables do have an updated_at timestamp, but no index: users tung@walle $
Q: What table alteration are supported?
A: I've only tested with adding columns and removing columns and indexes.
Q: Can I add and drop multiple columns and indexes at the same time? A: Yes.
For speed reasons the final sync is done by using the updated_at timestamp if available and syncing the data last updated since the last day. Data before that will not get synced in the final sync. So, having an updated_at timestamp and using it on the original table is very important.
For tables that do not have updated_at timestamps. I need to still limit the size of the final update so I'm limiting it to the last 100_000 records. Not much at all, so it is very important to have that updated_at timestamp.