2020-10-30 18:08:56 +00:00
---
stage: none
group: unassigned
2020-11-26 06:09:20 +00:00
info: To determine the technical writer assigned to the Stage/Group associated with this page, see https://about.gitlab.com/handbook/engineering/ux/technical-writing/#assignments
2020-10-30 18:08:56 +00:00
---
2015-05-11 13:09:36 +00:00
# Migration Style Guide
When writing migrations for GitLab, you have to take into account that
2020-11-26 03:09:17 +00:00
these are run by hundreds of thousands of organizations of all sizes, some with
2015-05-11 13:09:36 +00:00
many years of data in their database.
2018-01-19 10:09:59 +00:00
In addition, having to take a server offline for an upgrade small or big is a
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
big burden for most organizations. For this reason, it is important that your
migrations are written carefully, can be applied online, and adhere to the style
2017-04-12 16:15:19 +00:00
guide below.
2015-05-11 13:09:36 +00:00
2017-04-12 16:15:19 +00:00
Migrations are **not** allowed to require GitLab installations to be taken
2021-04-07 00:09:26 +00:00
offline ever. Migrations always must be written in such a way to avoid
downtime. In the past we had a process for defining migrations that allowed for
downtime by setting a `DOWNTIME` constant. You may see this when looking at
2021-07-06 12:07:43 +00:00
older migrations. This process was in place for 4 years without ever being
used and as such we've learned we can always figure out how to write a migration
2021-04-07 00:09:26 +00:00
differently to avoid downtime.
2017-04-12 16:15:19 +00:00
2015-05-11 13:09:36 +00:00
When writing your migrations, also consider that databases might have stale data
2017-04-12 16:15:19 +00:00
or inconsistencies and guard for that. Try to make as few assumptions as
possible about the state of the database.
Please don't depend on GitLab-specific code since it can change in future
versions. If needed copy-paste GitLab code into the migration to make it forward
compatible.
2021-07-29 18:10:22 +00:00
## Choose an appropriate migration type
2020-02-21 15:09:05 +00:00
2021-07-29 18:10:22 +00:00
The first step before adding a new migration should be to decide which type is most appropriate.
There are currently three kinds of migrations you can create, depending on the kind of
work it needs to perform and how long it takes to complete:
1. [**Regular schema migrations.** ](#create-a-regular-schema-migration ) These are traditional Rails migrations in `db/migrate` that run _before_ new application code is deployed
(for GitLab.com before [Canary is deployed ](https://gitlab.com/gitlab-com/gl-infra/readiness/-/tree/master/library/canary/#configuration-and-deployment )).
This means that they should be relatively fast, no more than a few minutes, so as not to unnecessarily delay a deployment.
One exception is a migration that takes longer but is absolutely critical for the application to operate correctly.
For example, you might have indices that enforce unique tuples, or that are needed for query performance in critical parts of the application. In cases where the migration would be unacceptably slow, however, a better option might be to guard the feature with a [feature flag ](feature_flags/index.md )
and perform a post-deployment migration instead. The feature can then be turned on after the migration finishes.
1. [**Post-deployment migrations.** ](post_deployment_migrations.md ) These are Rails migrations in `db/post_migrate` and
run _after_ new application code has been deployed (for GitLab.com after the production deployment has finished).
They can be used for schema changes that aren't critical for the application to operate, or data migrations that take at most a few minutes.
Common examples for schema changes that should run post-deploy include:
- Clean-ups, like removing unused columns.
- Adding non-critical indices on high-traffic tables.
- Adding non-critical indices that take a long time to create.
1. [**Background migrations.** ](background_migrations.md ) These aren't regular Rails migrations, but application code that is
executed via Sidekiq jobs, although a post-deployment migration is used to schedule them. Use them only for data migrations that
exceed the timing guidelines for post-deploy migrations. Background migrations should _not_ change the schema.
Use the following diagram to guide your decision, but keep in mind that it is just a tool, and
the final outcome will always be dependent on the specific changes being made:
```mermaid
graph LR
A{Schema< br / > changed?}
A -->|Yes| C{Critical to< br / > speed or< br / > behavior?}
A -->|No| D{Is it fast?}
C -->|Yes| H{Is it fast?}
C -->|No| F[Post-deploy migration]
H -->|Yes| E[Regular migration]
H -->|No| I[Post-deploy migration< br / > + feature flag]
2021-09-01 18:08:49 +00:00
2021-07-29 18:10:22 +00:00
D -->|Yes| F[Post-deploy migration]
D -->|No| G[Background migration]
```
### How long a migration should take
In general, all migrations for a single deploy shouldn't take longer than
1 hour for GitLab.com. The following guidelines are not hard rules, they were
estimated to keep migration duration to a minimum.
NOTE:
Keep in mind that all durations should be measured against GitLab.com.
| Migration Type | Recommended Duration | Notes |
|----|----|---|
| Regular migrations | `<= 3 minutes` | A valid exception are changes without which application functionality or performance would be severely degraded and which cannot be delayed. |
| Post-deployment migrations | `<= 10 minutes` | A valid exception are schema changes, since they must not happen in background migrations. |
| Background migrations | `> 10 minutes` | Since these are suitable for larger tables, it's not possible to set a precise timing guideline, however, any single query must stay below [`1 second` execution time ](query_performance.md#timing-guidelines-for-queries ) with cold caches. |
## Create a regular schema migration
2021-07-06 12:07:43 +00:00
To create a migration you can use the following Rails generator:
```shell
bundle exec rails g migration migration_name_here
```
This generates the migration file in `db/migrate` .
2017-08-04 11:30:57 +00:00
## Schema Changes
2020-06-03 06:08:34 +00:00
Changes to the schema should be committed to `db/structure.sql` . This
2021-07-06 12:07:43 +00:00
file is automatically generated by Rails when you run
`bundle exec rails db:migrate` , so you normally should not
2020-04-29 15:09:58 +00:00
edit this file by hand. If your migration is adding a column to a
2020-11-26 03:09:17 +00:00
table, that column is added at the bottom. Please do not reorder
columns manually for existing tables as this causes confusion to
2020-04-29 15:09:58 +00:00
other people using `db/structure.sql` generated by Rails.
When your local database in your GDK is diverging from the schema from
2021-06-16 00:10:15 +00:00
`main` it might be hard to cleanly commit the schema changes to
2020-04-29 18:09:56 +00:00
Git. In that case you can use the `scripts/regenerate-schema` script to
2020-04-29 15:09:58 +00:00
regenerate a clean `db/structure.sql` for the migrations you're
2020-11-26 03:09:17 +00:00
adding. This script applies all migrations found in `db/migrate`
2020-04-29 15:09:58 +00:00
or `db/post_migrate` , so if there are any migrations you don't want to
commit to the schema, rename or remove them. If your branch is not
2021-06-16 00:10:15 +00:00
targeting `main` you can set the `TARGET` environment variable.
2020-04-29 15:09:58 +00:00
2020-05-19 18:08:11 +00:00
```shell
2021-06-16 00:10:15 +00:00
# Regenerate schema against `main`
2020-04-29 18:09:56 +00:00
scripts/regenerate-schema
2020-04-29 15:09:58 +00:00
# Regenerate schema against `12-9-stable-ee`
2020-04-29 18:09:56 +00:00
TARGET=12-9-stable-ee scripts/regenerate-schema
2020-04-29 15:09:58 +00:00
```
2017-08-04 11:30:57 +00:00
2021-04-07 00:09:26 +00:00
## Avoiding downtime
2015-05-11 13:09:36 +00:00
2021-04-07 00:09:26 +00:00
The document ["Avoiding downtime in migrations" ](avoiding_downtime_in_migrations.md ) specifies
various database operations, such as:
2015-05-11 13:09:36 +00:00
2021-04-07 00:09:26 +00:00
- [dropping and renaming columns ](avoiding_downtime_in_migrations.md#dropping-columns )
- [changing column constraints and types ](avoiding_downtime_in_migrations.md#changing-column-constraints )
- [adding and dropping indexes, tables, and foreign keys ](avoiding_downtime_in_migrations.md#adding-indexes )
2021-11-15 18:12:21 +00:00
- [migrating `integer` primary keys to `bigint` ](avoiding_downtime_in_migrations.md#migrating-integer-primary-keys-to-bigint )
2015-05-11 13:09:36 +00:00
2021-04-07 00:09:26 +00:00
and explains how to perform them without requiring downtime.
2015-11-02 15:14:34 +00:00
2016-06-24 16:29:23 +00:00
## Reversibility
2015-05-11 13:09:36 +00:00
2017-04-12 16:15:19 +00:00
Your migration **must be** reversible. This is very important, as it should
2015-05-11 13:09:36 +00:00
be possible to downgrade in case of a vulnerability or bugs.
In your migration, add a comment describing how the reversibility of the
2015-05-12 08:48:18 +00:00
migration was tested.
2020-02-27 21:09:17 +00:00
Some migrations cannot be reversed. For example, some data migrations can't be
reversed because we lose information about the state of the database before the migration.
You should still create a `down` method with a comment, explaining why
the changes performed by the `up` method can't be reversed, so that the
migration itself can be reversed, even if the changes performed during the migration
can't be reversed:
```ruby
def down
# no-op
# comment explaining why changes performed by `up` cannot be reversed.
end
```
2021-12-23 18:14:10 +00:00
Migrations like this are inherently risky and [additional actions ](database_review.md#preparation-when-adding-data-migrations )
are required when preparing the migration for review.
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
## Atomicity
By default, migrations are single transaction. That is, a transaction is opened
at the beginning of the migration, and committed after all steps are processed.
Running migrations in a single transaction makes sure that if one of the steps fails,
2020-11-26 03:09:17 +00:00
none of the steps are executed, leaving the database in valid state.
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
Therefore, either:
- Put all migrations in one single-transaction migration.
2019-08-27 11:31:43 +00:00
- If necessary, put most actions in one migration and create a separate migration
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
for the steps that cannot be done in a single transaction.
2019-08-27 11:31:43 +00:00
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
For example, if you create an empty table and need to build an index for it,
it is recommended to use a regular single-transaction migration and the default
rails schema statement: [`add_index` ](https://api.rubyonrails.org/v5.2/classes/ActiveRecord/ConnectionAdapters/SchemaStatements.html#method-i-add_index ).
2020-11-26 03:09:17 +00:00
This is a blocking operation, but it doesn't cause problems because the table is not yet used,
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
and therefore it does not have any records yet.
2021-05-03 09:10:02 +00:00
## Naming conventions
We keep column names consistent with [ActiveRecord's schema conventions ](https://guides.rubyonrails.org/active_record_basics.html#schema-conventions ).
Custom index names should follow the pattern `index_#{table_name}_on_#{column_1}_and_#{column_2}_#{condition}` .
Examples:
- `index_services_on_type_and_id_and_template_when_active`
- `index_projects_on_id_service_desk_enabled`
- `index_clusters_on_enabled_cluster_type_id_and_created_at`
2021-05-05 15:10:05 +00:00
### Truncate long index names
PostgreSQL [limits the length of identifiers ](https://www.postgresql.org/docs/current/limits.html ),
like column or index names. Column names are not usually a problem, but index names tend
to be longer. Some methods for shortening a name that's too long:
- Prefix it with `i_` instead of `index_` .
- Skip redundant prefixes. For example,
`index_vulnerability_findings_remediations_on_vulnerability_remediation_id` becomes
`index_vulnerability_findings_remediations_on_remediation_id` .
- Instead of columns, specify the purpose of the index, such as `index_users_for_unconfirmation_notification` .
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
## Heavy operations in a single transaction
2020-11-26 03:09:17 +00:00
When using a single-transaction migration, a transaction holds a database connection
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
for the duration of the migration, so you must make sure the actions in the migration
2021-03-17 15:09:03 +00:00
do not take too much time: GitLab.com's production database has a `15s` timeout, so
2020-11-18 12:09:16 +00:00
in general, the cumulative execution time in a migration should aim to fit comfortably
in that limit. Singular query timings should fit within the [standard limit ](query_performance.md#timing-guidelines-for-queries )
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
In case you need to insert, update, or delete a significant amount of data, you:
- Must disable the single transaction with `disable_ddl_transaction!` .
- Should consider doing it in a [Background Migration ](background_migrations.md ).
2021-09-01 18:08:49 +00:00
## Migration helpers and versioning
2021-09-09 12:09:09 +00:00
> [Introduced](https://gitlab.com/gitlab-org/gitlab/-/issues/339115) in GitLab 14.3.
2021-09-01 18:08:49 +00:00
Various helper methods are available for many common patterns in database migrations. Those
helpers can be found in `Gitlab::Database::MigrationHelpers` and related modules.
In order to allow changing a helper's behavior over time, we implement a versioning scheme
for migration helpers. This allows us to maintain the behavior of a helper for already
existing migrations but change the behavior for any new migrations.
For that purpose, all database migrations should inherit from `Gitlab::Database::Migration` ,
which is a "versioned" class. For new migrations, the latest version should be used (which
can be looked up in `Gitlab::Database::Migration::MIGRATION_CLASSES` ) to use the latest version
of migration helpers.
In this example, we use version 1.0 of the migration class:
```ruby
class TestMigration < Gitlab::Database::Migration [ 1 . 0 ]
def change
end
end
```
2021-09-09 12:09:09 +00:00
Do not include `Gitlab::Database::MigrationHelpers` directly into a
migration. Instead, use the latest version of `Gitlab::Database::Migration` , which exposes the latest
2021-09-01 18:08:49 +00:00
version of migration helpers automatically.
2021-09-09 12:09:09 +00:00
Migration helpers and versioning were [introduced ](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/68986 )
in GitLab 14.3.
For merge requests targeting previous stable branches, use the old format and still inherit from
`ActiveRecord::Migration[6.1]` instead of `Gitlab::Database::Migration[1.0]` .
2021-09-01 18:08:49 +00:00
2020-02-06 06:08:52 +00:00
## Retry mechanism when acquiring database locks
When changing the database schema, we use helper methods to invoke DDL (Data Definition
Language) statements. In some cases, these DDL statements require a specific database lock.
Example:
```ruby
def change
remove_column :users, :full_name, :string
end
```
Executing this migration requires an exclusive lock on the `users` table. When the table
is concurrently accessed and modified by other processes, acquiring the lock may take
a while. The lock request is waiting in a queue and it may also block other queries
on the `users` table once it has been enqueued.
More information about PostgresSQL locks: [Explicit Locking ](https://www.postgresql.org/docs/current/explicit-locking.html )
For stability reasons, GitLab.com has a specific [`statement_timeout` ](../user/gitlab_com/index.md#postgresql )
2020-11-26 03:09:17 +00:00
set. When the migration is invoked, any database query has
a fixed time to execute. In a worst-case scenario, the request sits in the
2020-02-06 06:08:52 +00:00
lock queue, blocking other queries for the duration of the configured statement timeout,
then failing with `canceling statement due to statement timeout` error.
This problem could cause failed application upgrade processes and even application
stability issues, since the table may be inaccessible for a short period of time.
To increase the reliability and stability of database migrations, the GitLab codebase
2021-09-15 12:11:13 +00:00
offers a method to retry the operations with different `lock_timeout` settings
and wait time between the attempts. Multiple shorter attempts to acquire the necessary
2020-02-06 06:08:52 +00:00
lock allow the database to process other statements.
2021-09-15 12:11:13 +00:00
There are two distinct ways to use lock retries:
1. Inside a transactional migration: use `enable_lock_retries!` .
1. Inside a non-transactional migration: use `with_lock_retries` .
If possible, enable lock-retries for any migration that touches a [high-traffic table ](#high-traffic-tables ).
### Usage with transactional migrations
Regular migrations execute the full migration in a transaction. We can enable the
lock-retry methodology by calling `enable_lock_retries!` at the migration level.
This leads to the lock timeout being controlled for this migration. Also, it can lead to retrying the full
migration if the lock could not be granted within the timeout.
Note that, while this is currently an opt-in setting, we prefer to use lock-retries for all migrations and
plan to make this the default going forward.
Occasionally a migration may need to acquire multiple locks on different objects.
To prevent catalog bloat, ask for all those locks explicitly before performing any DDL.
A better strategy is to split the migration, so that we only need to acquire one lock at the time.
2020-02-06 06:08:52 +00:00
2020-04-02 18:08:11 +00:00
**Removing a column:**
2020-02-06 06:08:52 +00:00
```ruby
2021-09-15 12:11:13 +00:00
enable_lock_retries!
2021-11-17 18:14:12 +00:00
def change
remove_column :users, :full_name, :string
2020-02-06 06:08:52 +00:00
end
```
2021-07-01 09:07:33 +00:00
**Multiple changes on the same table:**
2021-09-15 12:11:13 +00:00
With the lock-retry methodology enabled, all operations wrap into a single transaction. When you have the lock,
2021-07-01 09:07:33 +00:00
you should do as much as possible inside the transaction rather than trying to get another lock later.
Be careful about running long database statements within the block. The acquired locks are kept until the transaction (block) finishes and depending on the lock type, it might block other database operations.
```ruby
2021-09-15 12:11:13 +00:00
enable_lock_retries!
2021-07-01 09:07:33 +00:00
def up
2021-09-15 12:11:13 +00:00
add_column :users, :full_name, :string
add_column :users, :bio, :string
2021-07-01 09:07:33 +00:00
end
def down
2021-09-15 12:11:13 +00:00
remove_column :users, :full_name
remove_column :users, :bio
2021-07-01 09:07:33 +00:00
end
```
2020-04-02 18:08:11 +00:00
**Removing a foreign key:**
2020-02-06 06:08:52 +00:00
```ruby
2021-09-15 12:11:13 +00:00
enable_lock_retries!
2020-03-13 15:09:21 +00:00
def up
2021-09-15 12:11:13 +00:00
remove_foreign_key :issues, :projects
2020-02-06 06:08:52 +00:00
end
2020-03-13 15:09:21 +00:00
def down
2021-09-15 12:11:13 +00:00
add_foreign_key :issues, :projects
2020-03-13 15:09:21 +00:00
end
2020-02-06 06:08:52 +00:00
```
2020-04-02 18:08:11 +00:00
**Changing default value for a column:**
2020-02-06 06:08:52 +00:00
```ruby
2021-09-15 12:11:13 +00:00
enable_lock_retries!
2020-03-13 15:09:21 +00:00
def up
2021-09-15 12:11:13 +00:00
change_column_default :merge_requests, :lock_version, from: nil, to: 0
2020-02-06 06:08:52 +00:00
end
2020-03-13 15:09:21 +00:00
def down
2021-09-15 12:11:13 +00:00
change_column_default :merge_requests, :lock_version, from: 0, to: nil
2020-03-13 15:09:21 +00:00
end
2020-02-06 06:08:52 +00:00
```
2020-04-02 18:08:11 +00:00
**Creating a new table with a foreign key:**
2021-04-01 06:09:29 +00:00
We can wrap the `create_table` method with `with_lock_retries` :
2020-04-02 18:08:11 +00:00
```ruby
2021-09-15 12:11:13 +00:00
enable_lock_retries!
2020-04-02 18:08:11 +00:00
def up
2021-09-15 12:11:13 +00:00
create_table :issues do |t|
t.references :project, index: true, null: false, foreign_key: { on_delete: :cascade }
t.string :title, limit: 255
2020-04-02 18:08:11 +00:00
end
end
def down
2021-09-15 12:11:13 +00:00
drop_table :issues
2020-04-02 18:08:11 +00:00
end
```
**Creating a new table when we have two foreign keys:**
2021-09-17 00:09:10 +00:00
Only one foreign key should be created per transaction. This is because [the addition of a foreign key constraint requires a `SHARE ROW EXCLUSIVE` lock on the referenced table ](https://www.postgresql.org/docs/12/sql-createtable.html#:~:text=The%20addition%20of%20a%20foreign%20key%20constraint%20requires%20a%20SHARE%20ROW%20EXCLUSIVE%20lock%20on%20the%20referenced%20table ), and locking multiple tables in the same transaction should be avoided.
2021-07-19 18:08:23 +00:00
2020-11-26 03:09:17 +00:00
For this, we need three migrations:
2020-04-02 18:08:11 +00:00
1. Creating the table without foreign keys (with the indices).
1. Add foreign key to the first table.
1. Add foreign key to the second table.
Creating the table:
```ruby
def up
create_table :imports do |t|
t.bigint :project_id, null: false
t.bigint :user_id, null: false
t.string :jid, limit: 255
2021-04-19 15:09:08 +00:00
t.index :project_id
t.index :user_id
end
2020-04-02 18:08:11 +00:00
end
def down
drop_table :imports
end
```
Adding foreign key to `projects` :
2021-03-30 21:10:07 +00:00
We can use the `add_concurrent_foreign_key` method in this case, as this helper method
2020-10-16 15:08:46 +00:00
has the lock retries built into it.
2020-04-02 18:08:11 +00:00
```ruby
2020-10-16 15:08:46 +00:00
disable_ddl_transaction!
2020-04-02 18:08:11 +00:00
def up
2020-10-16 15:08:46 +00:00
add_concurrent_foreign_key :imports, :projects, column: :project_id, on_delete: :cascade
2020-04-02 18:08:11 +00:00
end
def down
with_lock_retries do
remove_foreign_key :imports, column: :project_id
end
end
```
Adding foreign key to `users` :
```ruby
2020-10-16 15:08:46 +00:00
disable_ddl_transaction!
2020-04-02 18:08:11 +00:00
def up
2020-10-16 15:08:46 +00:00
add_concurrent_foreign_key :imports, :users, column: :user_id, on_delete: :cascade
2020-04-02 18:08:11 +00:00
end
def down
with_lock_retries do
remove_foreign_key :imports, column: :user_id
end
end
```
2021-09-15 12:11:13 +00:00
### Usage with non-transactional migrations (`disable_ddl_transaction!`)
Only when we disable transactional migrations using `disable_ddl_transaction!` , we can use
the `with_lock_retries` helper to guard an individual sequence of steps. It opens a transaction
to execute the given block.
2020-05-04 09:09:36 +00:00
2021-09-15 12:11:13 +00:00
A custom RuboCop rule ensures that only allowed methods can be placed within the lock retries block.
2020-05-04 09:09:36 +00:00
```ruby
disable_ddl_transaction!
def up
with_lock_retries do
2021-09-15 12:11:13 +00:00
add_column :users, :name, :text unless column_exists?(:users, :name)
2020-05-04 09:09:36 +00:00
end
add_text_limit :users, :name, 255 # Includes constraint validation (full table scan)
end
```
2020-11-26 03:09:17 +00:00
The RuboCop rule generally allows standard Rails migration methods, listed below. This example causes a Rubocop offense:
2020-05-04 09:09:36 +00:00
```ruby
2020-10-06 09:08:32 +00:00
disable_ddl_transaction!
2020-05-04 09:09:36 +00:00
def up
with_lock_retries do
add_concurrent_index :users, :name
end
end
```
2020-02-06 06:08:52 +00:00
### When to use the helper method
2021-09-15 12:11:13 +00:00
You can **only** use the `with_lock_retries` helper method when the execution is not already inside
an open transaction (using Postgres subtransactions is discouraged). It can be used with
2020-02-06 06:08:52 +00:00
standard Rails migration helper methods. Calling more than one migration
helper is not a problem if they're executed on the same table.
Using the `with_lock_retries` helper method is advised when a database
2021-03-05 18:09:17 +00:00
migration involves one of the [high-traffic tables ](#high-traffic-tables ).
2020-02-06 06:08:52 +00:00
Example changes:
- `add_foreign_key` / `remove_foreign_key`
- `add_column` / `remove_column`
- `change_column_default`
2020-04-02 12:08:18 +00:00
- `create_table` / `drop_table`
2020-02-06 06:08:52 +00:00
2020-11-04 00:09:12 +00:00
The `with_lock_retries` method **cannot** be used within the `change` method, you must manually define the `up` and `down` methods to make the migration reversible.
2020-03-13 15:09:21 +00:00
2020-02-06 06:08:52 +00:00
### How the helper method works
1. Iterate 50 times.
1. For each iteration, set a pre-configured `lock_timeout` .
1. Try to execute the given block. (`remove_column`).
1. If `LockWaitTimeout` error is raised, sleep for the pre-configured `sleep_time`
and retry the block.
1. If no error is raised, the current iteration has successfully executed the block.
For more information check the [`Gitlab::Database::WithLockRetries` ](https://gitlab.com/gitlab-org/gitlab/-/blob/master/lib/gitlab/database/with_lock_retries.rb ) class. The `with_lock_retries` helper method is implemented in the [`Gitlab::Database::MigrationHelpers` ](https://gitlab.com/gitlab-org/gitlab/-/blob/master/lib/gitlab/database/migration_helpers.rb ) module.
In a worst-case scenario, the method:
- Executes the block for a maximum of 50 times over 40 minutes.
- Most of the time is spent in a pre-configured sleep period after each iteration.
2020-11-26 03:09:17 +00:00
- After the 50th retry, the block is executed without `lock_timeout` , just
2020-02-06 06:08:52 +00:00
like a standard migration invocation.
2020-11-26 03:09:17 +00:00
- If a lock cannot be acquired, the migration fails with `statement timeout` error.
2020-02-06 06:08:52 +00:00
The migration might fail if there is a very long running transaction (40+ minutes)
accessing the `users` table.
2017-06-13 11:44:13 +00:00
## Removing indexes
2015-05-12 08:48:18 +00:00
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
If the table is not empty when removing an index, make sure to use the method
`remove_concurrent_index` instead of the regular `remove_index` method.
The `remove_concurrent_index` method drops indexes concurrently, so no locking is required,
and there is no need for downtime. To use this method, you must disable single-transaction mode
2018-08-14 00:51:27 +00:00
by calling the method `disable_ddl_transaction!` in the body of your migration
class like so:
2015-05-12 08:48:18 +00:00
2016-11-09 10:59:15 +00:00
```ruby
2021-09-01 18:08:49 +00:00
class MyMigration < Gitlab::Database::Migration [ 1 . 0 ]
2017-04-05 22:53:57 +00:00
disable_ddl_transaction!
2020-12-03 21:09:35 +00:00
INDEX_NAME = 'index_name'
2021-09-01 18:08:49 +00:00
2017-04-05 22:53:57 +00:00
def up
2020-12-03 21:09:35 +00:00
remove_concurrent_index :table_name, :column_name, name: INDEX_NAME
2017-04-05 22:53:57 +00:00
end
end
2015-05-12 08:48:18 +00:00
```
2018-03-20 13:38:43 +00:00
Note that it is not necessary to check if the index exists prior to
2020-09-16 21:09:52 +00:00
removing it, however it is required to specify the name of the
index that is being removed. This can be done either by passing the name
as an option to the appropriate form of `remove_index` or `remove_concurrent_index` ,
or more simply by using the `remove_concurrent_index_by_name` method. Explicitly
specifying the name is important to ensure the correct index is removed.
2018-03-20 13:38:43 +00:00
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
For a small table (such as an empty one or one with less than `1,000` records),
it is recommended to use `remove_index` in a single-transaction migration,
combining it with other operations that don't require `disable_ddl_transaction!` .
2020-11-16 18:09:15 +00:00
### Disabling an index
There are certain situations in which you might want to disable an index before removing it.
See the [maintenance operations guide ](database/maintenance_operations.md#disabling-an-index )
for more details.
2017-06-13 11:44:13 +00:00
## Adding indexes
2015-05-12 08:48:18 +00:00
2020-02-20 18:08:51 +00:00
Before adding an index, consider if this one is necessary. There are situations in which an index
might not be required, like:
- The table is small (less than `1,000` records) and it's not expected to exponentially grow in size.
- Any existing indexes filter out enough rows.
- The reduction in query timings after the index is added is not significant.
Additionally, wide indexes are not required to match all filter criteria of queries, we just need
to cover enough columns so that the index lookup has a small enough selectivity. Please review our
[Adding Database indexes ](adding_database_indexes.md ) guide for more details.
2015-05-12 08:48:18 +00:00
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
When adding an index to a non-empty table make sure to use the method
2019-08-27 11:31:43 +00:00
`add_concurrent_index` instead of the regular `add_index` method.
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
The `add_concurrent_index` method automatically creates concurrent indexes
2019-08-27 11:31:43 +00:00
when using PostgreSQL, removing the need for downtime.
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
To use this method, you must disable single-transactions mode
by calling the method `disable_ddl_transaction!` in the body of your migration
class like so:
2016-05-09 13:05:19 +00:00
2016-11-09 10:59:15 +00:00
```ruby
2021-09-01 18:08:49 +00:00
class MyMigration < Gitlab::Database::Migration [ 1 . 0 ]
2016-05-09 13:05:19 +00:00
disable_ddl_transaction!
2021-01-06 09:10:31 +00:00
INDEX_NAME = 'index_name'
2017-04-12 16:15:19 +00:00
def up
2021-01-06 09:10:31 +00:00
add_concurrent_index :table, :column, name: INDEX_NAME
2017-04-12 16:15:19 +00:00
end
2016-05-09 13:05:19 +00:00
2017-04-12 16:15:19 +00:00
def down
2021-01-06 09:10:31 +00:00
remove_concurrent_index :table, :column, name: INDEX_NAME
2016-05-09 13:05:19 +00:00
end
end
```
2020-09-16 21:09:52 +00:00
You must explicitly name indexes that are created with more complex
definitions beyond table name, column name(s) and uniqueness constraint.
Consult the [Adding Database Indexes ](adding_database_indexes.md#requirements-for-naming-indexes )
guide for more details.
2020-02-20 18:08:51 +00:00
If you need to add a unique index, please keep in mind there is the possibility
of existing duplicates being present in the database. This means that should
always _first_ add a migration that removes any duplicates, before adding the
unique index.
2019-08-27 11:31:43 +00:00
For a small table (such as an empty one or one with less than `1,000` records),
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
it is recommended to use `add_index` in a single-transaction migration, combining it with other
operations that don't require `disable_ddl_transaction!` .
2020-09-16 21:09:52 +00:00
## Testing for existence of indexes
If a migration requires conditional logic based on the absence or
presence of an index, you must test for existence of that index using
its name. This helps avoids problems with how Rails compares index definitions,
which can lead to unexpected results. For more details, review the
[Adding Database Indexes ](adding_database_indexes.md#why-explicit-names-are-required )
guide.
The easiest way to test for existence of an index by name is to use the
`index_name_exists?` method, but the `index_exists?` method can also
be used with a name option. For example:
```ruby
2021-09-01 18:08:49 +00:00
class MyMigration < Gitlab::Database::Migration [ 1 . 0 ]
2020-09-16 21:09:52 +00:00
INDEX_NAME = 'index_name'
def up
# an index must be conditionally created due to schema inconsistency
unless index_exists?(:table_name, :column_name, name: INDEX_NAME)
add_index :table_name, :column_name, name: INDEX_NAME
end
end
def down
# no op
end
end
```
Keep in mind that concurrent index helpers like `add_concurrent_index` ,
`remove_concurrent_index` , and `remove_concurrent_index_by_name` already
perform existence checks internally.
2018-08-01 10:05:37 +00:00
## Adding foreign-key constraints
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
When adding a foreign-key constraint to either an existing or a new column also
remember to add an index on the column.
2018-08-01 10:05:37 +00:00
2021-07-19 18:08:23 +00:00
This is **required** for all foreign-keys, for example, to support efficient cascading
2019-05-23 12:45:52 +00:00
deleting: when a lot of rows in a table get deleted, the referenced records need
to be deleted too. The database has to look for corresponding records in the
2020-11-26 03:09:17 +00:00
referenced table. Without an index, this results in a sequential scan on the
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
table, which can take a long time.
2018-08-01 10:05:37 +00:00
Here's an example where we add a new column with a foreign key
constraint. Note it includes `index: true` to create an index for it.
```ruby
2021-09-01 18:08:49 +00:00
class Migration < Gitlab::Database::Migration [ 1 . 0 ]
2018-08-01 10:05:37 +00:00
def change
add_reference :model, :other_model, index: true, foreign_key: { on_delete: :cascade }
end
end
```
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
When adding a foreign-key constraint to an existing column in a non-empty table,
we have to employ `add_concurrent_foreign_key` and `add_concurrent_index`
2018-08-01 10:05:37 +00:00
instead of `add_reference` .
2021-02-24 21:11:16 +00:00
If you have a new or empty table that doesn't reference a
2021-03-05 18:09:17 +00:00
[high-traffic table ](#high-traffic-tables ),
2021-02-24 21:11:16 +00:00
we recommend that you use `add_reference` in a single-transaction migration. You can
combine it with other operations that don't require `disable_ddl_transaction!` .
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
2020-02-28 03:09:04 +00:00
You can read more about adding [foreign key constraints to an existing column ](database/add_foreign_key_to_existing_column.md ).
2020-06-05 12:08:24 +00:00
## `NOT NULL` constraints
> [Introduced](https://gitlab.com/gitlab-org/gitlab/-/issues/38358) in GitLab 13.0.
See the style guide on [`NOT NULL` constraints ](database/not_null_constraints.md ) for more information.
2016-05-09 13:05:19 +00:00
## Adding Columns With Default Values
2020-10-15 00:08:42 +00:00
With PostgreSQL 11 being the minimum version in GitLab 13.0 and later, adding columns with default values has become much easier and
2020-05-15 15:08:04 +00:00
the standard `add_column` helper should be used in all cases.
2016-05-09 13:05:19 +00:00
2020-05-15 15:08:04 +00:00
Before PostgreSQL 11, adding a column with a default was problematic as it would
have caused a full table rewrite. The corresponding helper `add_column_with_default`
2020-11-26 03:09:17 +00:00
has been deprecated and is scheduled to be removed in a later release.
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
2020-05-25 18:08:35 +00:00
If a backport adding a column with a default value is needed for %12.9 or earlier versions,
2020-06-15 21:08:35 +00:00
it should use `add_column_with_default` helper. If a [large table ](https://gitlab.com/gitlab-org/gitlab/-/blob/master/rubocop/rubocop-migrations.yml#L3 )
2020-05-25 18:08:35 +00:00
is involved, backporting to %12.9 is contraindicated.
2019-10-07 18:06:24 +00:00
## Changing the column default
One might think that changing a default column with `change_column_default` is an
expensive and disruptive operation for larger tables, but in reality it's not.
Take the following migration as an example:
```ruby
2021-09-01 18:08:49 +00:00
class DefaultRequestAccessGroups < Gitlab::Database::Migration [ 1 . 0 ]
2020-02-21 12:09:07 +00:00
def change
change_column_default(:namespaces, :request_access_enabled, from: false, to: true)
2019-10-07 18:06:24 +00:00
end
end
```
Migration above changes the default column value of one of our largest
tables: `namespaces` . This can be translated to:
```sql
ALTER TABLE namespaces
ALTER COLUMN request_access_enabled
2020-09-29 09:09:49 +00:00
SET DEFAULT false
2019-10-07 18:06:24 +00:00
```
In this particular case, the default value exists and we're just changing the metadata for
`request_access_enabled` column, which does not imply a rewrite of all the existing records
in the `namespaces` table. Only when creating a new column with a default, all the records are going be rewritten.
2020-12-04 21:09:29 +00:00
NOTE:
2020-07-16 06:09:33 +00:00
A faster [ALTER TABLE ADD COLUMN with a non-null default ](https://www.depesz.com/2018/04/04/waiting-for-postgresql-11-fast-alter-table-add-column-with-a-non-null-default/ )
2020-01-09 03:07:56 +00:00
was introduced on PostgresSQL 11.0, removing the need of rewriting the table when a new column with a default value is added.
2019-10-07 18:06:24 +00:00
For the reasons mentioned above, it's safe to use `change_column_default` in a single-transaction migration
without requiring `disable_ddl_transaction!` .
2017-11-17 16:02:10 +00:00
## Updating an existing column
To update an existing column to a particular value, you can use
2020-11-26 03:09:17 +00:00
`update_column_in_batches` . This splits the updates into batches, so we
2017-11-17 16:02:10 +00:00
don't update too many rows at in a single statement.
This updates the column `foo` in the `projects` table to 10, where `some_column`
is `'hello'` :
```ruby
update_column_in_batches(:projects, :foo, 10) do |table, query|
query.where(table[:some_column].eq('hello'))
end
```
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
If a computed update is needed, the value can be wrapped in `Arel.sql` , so Arel
2020-05-21 06:08:25 +00:00
treats it as an SQL literal. It's also a required deprecation for [Rails 6 ](https://gitlab.com/gitlab-org/gitlab/-/issues/28497 ).
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
The below example is the same as the one above, but
2017-11-17 16:02:10 +00:00
the value is set to the product of the `bar` and `baz` columns:
```ruby
update_value = Arel.sql('bar * baz')
update_column_in_batches(:projects, :foo, update_value) do |table, query|
query.where(table[:some_column].eq('hello'))
end
```
Like `add_column_with_default` , there is a RuboCop cop to detect usage of this
on large tables. In the case of `update_column_in_batches` , it may be acceptable
to run on a large table, as long as it is only updating a small subset of the
rows in the table, but do not ignore that without validating on the GitLab.com
staging environment - or asking someone else to do so for you - beforehand.
2016-09-13 22:15:14 +00:00
2022-01-04 09:13:39 +00:00
## Removing a foreign key constraint
When removing a foreign key constraint, we need to acquire a lock on both tables
that are related to the foreign key. For tables with heavy write patterns, it's a good
idea to use `with_lock_retries` , otherwise you might fail to acquire a lock in time.
You might also run into deadlocks when acquiring a lock, because ordinarily
the application writes in `parent,child` order. However, removing a foreign
key acquires the lock in `child,parent` order. To resolve this, you can
explicitly acquire the lock in `parent,child` , for example:
```ruby
disable_ddl_transaction!
def up
with_lock_retries do
execute('lock table ci_pipelines, ci_builds in access exclusive mode')
remove_foreign_key :ci_builds, to_table: :ci_pipelines, column: :pipeline_id, on_delete: :cascade, name: 'the_fk_name'
end
end
def down
add_concurrent_foreign_key :ci_builds, :ci_pipelines, column: :pipeline_id, on_delete: :cascade, name: 'the_fk_name'
end
```
2020-02-06 06:08:52 +00:00
## Dropping a database table
Dropping a database table is uncommon, and the `drop_table` method
provided by Rails is generally considered safe. Before dropping the table,
please consider the following:
2021-03-05 18:09:17 +00:00
If your table has foreign keys on a [high-traffic table ](#high-traffic-tables ) (like `projects` ), then
the `DROP TABLE` statement is likely to stall concurrent traffic until it fails with **statement timeout** error.
2020-02-06 06:08:52 +00:00
Table **has no records** (feature was never in use) and **no foreign
keys**:
2021-04-01 06:09:29 +00:00
- Use the `drop_table` method in your migration.
2020-02-06 06:08:52 +00:00
```ruby
def change
drop_table :my_table
end
```
Table **has records** but **no foreign keys** :
- First release: Remove the application code related to the table, such as models,
controllers and services.
- Second release: Use the `drop_table` method in your migration.
```ruby
def up
drop_table :my_table
end
def down
# create_table ...
end
```
Table **has foreign keys** :
- First release: Remove the application code related to the table, such as models,
controllers, and services.
- Second release: Remove the foreign keys using the `with_lock_retries`
helper method. Use `drop_table` in another migration file.
**Migrations for the second release:**
Removing the foreign key on the `projects` table:
```ruby
# first migration file
def up
with_lock_retries do
remove_foreign_key :my_table, :projects
end
end
def down
with_lock_retries do
add_foreign_key :my_table, :projects
end
end
```
Dropping the table:
```ruby
# second migration file
def up
drop_table :my_table
end
def down
# create_table ...
end
```
2016-09-13 22:15:14 +00:00
## Integer column type
By default, an integer column can hold up to a 4-byte (32-bit) number. That is
2020-11-26 03:09:17 +00:00
a max value of 2,147,483,647. Be aware of this when creating a column that
holds file sizes in byte units. If you are tracking file size in bytes, this
2016-09-13 22:15:14 +00:00
restricts the maximum file size to just over 2GB.
To allow an integer column to hold up to an 8-byte (64-bit) number, explicitly
2020-11-26 03:09:17 +00:00
set the limit to 8-bytes. This allows the column to hold a value up to
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
`9,223,372,036,854,775,807` .
2016-09-13 22:15:14 +00:00
Rails migration example:
2016-11-09 10:59:15 +00:00
```ruby
2020-05-15 15:08:04 +00:00
add_column(:projects, :foo, :integer, default: 10, limit: 8)
2016-09-13 22:15:14 +00:00
```
2020-05-22 09:08:09 +00:00
## Strings and the Text data type
> [Introduced](https://gitlab.com/gitlab-org/gitlab/-/issues/30453) in GitLab 13.0.
See the [text data type ](database/strings_and_the_text_data_type.md ) style guide for more information.
2017-06-13 11:44:13 +00:00
## Timestamp column type
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
By default, Rails uses the `timestamp` data type that stores timestamp data
2021-09-28 15:11:30 +00:00
without time zone information. The `timestamp` data type is used by calling
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
either the `add_timestamps` or the `timestamps` method.
Also, Rails converts the `:datetime` data type to the `timestamp` one.
2017-06-13 11:44:13 +00:00
Example:
```ruby
# timestamps
create_table :users do |t|
t.timestamps
end
# add_timestamps
def up
add_timestamps :users
end
# :datetime
def up
add_column :users, :last_sign_in, :datetime
end
```
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
Instead of using these methods, one should use the following methods to store
2021-09-28 15:11:30 +00:00
timestamps with time zones:
2017-06-13 11:44:13 +00:00
2018-11-13 06:07:16 +00:00
- `add_timestamps_with_timezone`
- `timestamps_with_timezone`
2019-09-23 18:06:14 +00:00
- `datetime_with_timezone`
2017-06-13 11:44:13 +00:00
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
This ensures all timestamps have a time zone specified. This, in turn, means
2021-09-28 15:11:30 +00:00
existing timestamps don't suddenly use a different time zone when the system's
time zone changes. It also makes it very clear which time zone was used in the
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
first place.
2017-06-13 11:44:13 +00:00
2018-09-02 14:35:15 +00:00
## Storing JSON in database
The Rails 5 natively supports `JSONB` (binary JSON) column type.
Example migration adding this column:
```ruby
2021-09-01 18:08:49 +00:00
class AddOptionsToBuildMetadata < Gitlab::Database::Migration [ 1 . 0 ]
2018-09-02 14:35:15 +00:00
def change
add_column :ci_builds_metadata, :config_options, :jsonb
end
end
```
2019-08-06 14:08:28 +00:00
You have to use a serializer to provide a translation layer:
2018-09-02 14:35:15 +00:00
```ruby
class BuildMetadata
2021-06-11 18:10:13 +00:00
serialize :config_options, Serializers::Json # rubocop:disable Cop/ActiveRecordSerialize
2018-09-02 14:35:15 +00:00
end
```
2017-06-13 11:44:13 +00:00
2020-06-18 12:09:25 +00:00
When using a `JSONB` column, use the [JsonSchemaValidator ](https://gitlab.com/gitlab-org/gitlab/-/blob/master/app/validators/json_schema_validator.rb ) to keep control of the data being inserted over time.
```ruby
class BuildMetadata
2020-07-02 06:09:26 +00:00
validates :config_options, json_schema: { filename: 'build_metadata_config_option' }
2020-06-18 12:09:25 +00:00
end
```
2021-05-24 18:10:28 +00:00
## Encrypted attributes
> [Introduced](https://gitlab.com/gitlab-org/gitlab/-/issues/227779) in GitLab 14.0.
Do not store `attr_encrypted` attributes as `:text` in the database; use
`:binary` instead. This uses the `bytea` type in PostgreSQL and makes storage more
efficient:
```ruby
2021-09-01 18:08:49 +00:00
class AddSecretToSomething < Gitlab::Database::Migration [ 1 . 0 ]
2021-05-24 18:10:28 +00:00
def change
add_column :something, :encrypted_secret, :binary
add_column :something, :encrypted_secret_iv, :binary
end
end
```
When storing encrypted attributes in a binary column, we need to provide the
`encode: false` and `encode_iv: false` options to `attr_encrypted` :
```ruby
class Something < ApplicationRecord
attr_encrypted :secret,
mode: :per_attribute_iv,
key: Settings.attr_encrypted_db_key_base_32,
algorithm: 'aes-256-gcm',
encode: false,
encode_iv: false
end
```
2015-05-12 08:48:18 +00:00
## Testing
2019-09-24 06:06:02 +00:00
See the [Testing Rails migrations ](testing_guide/testing_migrations_guide.md ) style guide.
2015-05-12 08:48:18 +00:00
## Data migration
2017-04-12 16:15:19 +00:00
Please prefer Arel and plain SQL over usual ActiveRecord syntax. In case of
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
using plain SQL, you need to quote all input manually with `quote_string` helper.
2015-05-12 08:48:18 +00:00
Example with Arel:
2016-11-09 10:59:15 +00:00
```ruby
2015-05-12 08:48:18 +00:00
users = Arel::Table.new(:users)
users.group(users[:user_id]).having(users[:id].count.gt(5))
2016-05-30 05:31:39 +00:00
#update other tables with these results
2015-05-12 08:48:18 +00:00
```
Example with plain SQL and `quote_string` helper:
2016-11-09 10:59:15 +00:00
```ruby
2015-05-12 08:48:18 +00:00
select_all("SELECT name, COUNT(id) as cnt FROM tags GROUP BY name HAVING COUNT(id) > 1").each do |tag|
tag_name = quote_string(tag["name"])
duplicate_ids = select_all("SELECT id FROM tags WHERE name = '#{tag_name}'").map{|tag| tag["id"]}
origin_tag_id = duplicate_ids.first
duplicate_ids.delete origin_tag_id
execute("UPDATE taggings SET tag_id = #{origin_tag_id} WHERE tag_id IN(#{duplicate_ids.join(",")})")
execute("DELETE FROM tags WHERE id IN(#{duplicate_ids.join(",")})")
end
2016-05-09 13:05:19 +00:00
```
2017-04-12 16:15:19 +00:00
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
If you need more complex logic, you can define and use models local to a
2017-04-12 16:15:19 +00:00
migration. For example:
```ruby
2021-09-01 18:08:49 +00:00
class MyMigration < Gitlab::Database::Migration [ 1 . 0 ]
2017-04-12 16:15:19 +00:00
class Project < ActiveRecord::Base
self.table_name = 'projects'
end
2020-07-16 12:09:22 +00:00
def up
# Reset the column information of all the models that update the database
# to ensure the Active Record's knowledge of the table structure is current
Project.reset_column_information
# ... ...
end
2017-04-12 16:15:19 +00:00
end
```
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
When doing so be sure to explicitly set the model's table name, so it's not
2017-04-12 16:15:19 +00:00
derived from the class name or namespace.
2017-05-02 12:10:55 +00:00
2020-08-06 15:09:42 +00:00
Be aware of the limitations [when using models in migrations ](#using-models-in-migrations-discouraged ).
2020-07-16 12:09:22 +00:00
2021-06-25 15:08:34 +00:00
### Modifying existing data
In most circumstances, prefer migrating data in **batches** when modifying data in the database.
We introduced a new helper [each_batch_range ](https://gitlab.com/gitlab-org/gitlab/-/blob/cd3e0a5cddcb464cb9b8c6e3275839cf57dfa6e2/lib/gitlab/database/dynamic_model_helpers.rb#L28-32 ) which facilitates the process of iterating over a collection in a performant way. The default size of the batch is defined in the `BATCH_SIZE` constant.
See the following example to get an idea.
**Purging data in batch:**
```ruby
include ::Gitlab::Database::DynamicModelHelpers
disable_ddl_transaction!
def up
each_batch_range('ci_pending_builds', scope: ->(table) { table.ref_protected }, of: BATCH_SIZE) do |min, max|
execute < < ~SQL
DELETE FROM ci_pending_builds
USING ci_builds
WHERE ci_builds.id = ci_pending_builds.build_id
AND ci_builds.status != 'pending'
AND ci_builds.type = 'Ci::Build'
AND ci_pending_builds.id BETWEEN #{min} AND #{max}
SQL
end
end
```
- The first argument is the table being modified: `'ci_pending_builds'` .
- The second argument calls a lambda which fetches the relevant dataset selected (the default is set to `.all` ): `scope: ->(table) { table.ref_protected }` .
- The third argument is the batch size (the default is set in the `BATCH_SIZE` constant): `of: BATCH_SIZE` .
Here is an [example MR ](https://gitlab.com/gitlab-org/gitlab/-/merge_requests/62195 ) illustrating how to use our new helper.
2017-05-02 12:10:55 +00:00
### Renaming reserved paths
Migrations guide: use atomic steps, when possible
Currently, the DB migrations guide says that "you must" use non-blocking
operations (such as CREATE INDEX CONCURRENTLY), always.
But this does not make sense in cases of empty tables and leads to
splitting the work to multiple non-atomic
(with disable_ddl_transaction!) DB migrations.
To follow KISS principle, to have fewer DB migrations steps,
to have them atomic when it's possible and simplify deployment
and troubleshooting, the following exceptions were added
to the doc:
- index creation,
- index dropping,
- defining an FK,
- adding a column with DEFAULT,
2019-08-27 10:22:12 +00:00
When a new route for projects is introduced, it could conflict with any
existing records. The path for these records should be renamed, and the
2017-05-02 12:10:55 +00:00
related data should be moved on disk.
Since we had to do this a few times already, there are now some helpers to help
with this.
To use this you can include `Gitlab::Database::RenameReservedPathsMigration::V1`
2020-11-26 03:09:17 +00:00
in your migration. This provides 3 methods which you can pass one or more
2017-05-02 12:10:55 +00:00
paths that need to be rejected.
2020-11-26 03:09:17 +00:00
- **`rename_root_paths`**: Renames the path of all _namespaces_ with the
2017-05-02 12:10:55 +00:00
given name that don't have a `parent_id` .
2020-11-26 03:09:17 +00:00
- **`rename_child_paths`**: Renames the path of all _namespaces_ with the
2017-05-02 12:10:55 +00:00
given name that have a `parent_id` .
2020-11-26 03:09:17 +00:00
- **`rename_wildcard_paths`**: Renames the path of all _projects_ , and all
2017-05-02 12:10:55 +00:00
_namespaces_ that have a `project_id` .
2020-11-26 03:09:17 +00:00
The `path` column for these rows are renamed to their previous value followed
2017-05-02 12:10:55 +00:00
by an integer. For example: `users` would turn into `users0`
2020-08-06 15:09:42 +00:00
## Using models in migrations (discouraged)
The use of models in migrations is generally discouraged. As such models are
[contraindicated for background migrations ](background_migrations.md#isolation ),
the model needs to be declared in the migration.
If using a model in the migrations, you should first
[clear the column cache ](https://api.rubyonrails.org/classes/ActiveRecord/ModelSchema/ClassMethods.html#method-i-reset_column_information )
using `reset_column_information` .
2021-06-08 15:10:00 +00:00
If using a model that leverages single table inheritance (STI), there are [special
considerations](single_table_inheritance.md#in-migrations).
2020-08-06 15:09:42 +00:00
This avoids problems where a column that you are using was altered and cached
in a previous migration.
### Example: Add a column `my_column` to the users table
It is important not to leave out the `User.reset_column_information` command, in order to ensure that the old schema is dropped from the cache and ActiveRecord loads the updated schema information.
```ruby
2021-09-01 18:08:49 +00:00
class AddAndSeedMyColumn < Gitlab::Database::Migration [ 1 . 0 ]
2020-08-06 15:09:42 +00:00
class User < ActiveRecord::Base
self.table_name = 'users'
end
def up
User.count # Any ActiveRecord calls on the model that caches the column information.
add_column :users, :my_column, :integer, default: 1
User.reset_column_information # The old schema is dropped from the cache.
User.find_each do |user|
user.my_column = 42 if some_condition # ActiveRecord sees the correct schema here.
user.save!
end
end
end
```
The underlying table is modified and then accessed via ActiveRecord.
Note that this also needs to be used if the table is modified in a previous, different migration,
if both migrations are run in the same `db:migrate` process.
This results in the following. Note the inclusion of `my_column` :
```shell
== 20200705232821 AddAndSeedMyColumn: migrating ==============================
D, [2020-07-06T00:37:12.483876 #130101 ] DEBUG -- : (0.2ms) BEGIN
D, [2020-07-06T00:37:12.521660 #130101 ] DEBUG -- : (0.4ms) SELECT COUNT(*) FROM "user"
-- add_column(:users, :my_column, :integer, {:default=>1})
D, [2020-07-06T00:37:12.523309 #130101 ] DEBUG -- : (0.8ms) ALTER TABLE "users" ADD "my_column" integer DEFAULT 1
-> 0.0016s
D, [2020-07-06T00:37:12.650641 #130101 ] DEBUG -- : AddAndSeedMyColumn::User Load (0.7ms) SELECT "users".* FROM "users" ORDER BY "users"."id" ASC LIMIT $1 [["LIMIT", 1000]]
D, [2020-07-18T00:41:26.851769 #459802 ] DEBUG -- : AddAndSeedMyColumn::User Update (1.1ms) UPDATE "users" SET "my_column" = $1, "updated_at" = $2 WHERE "users"."id" = $3 [["my_column", 42], ["updated_at", "2020-07-17 23:41:26.849044"], ["id", 1]]
D, [2020-07-06T00:37:12.653648 #130101 ] DEBUG -- : ↳ config/initializers/config_initializers_active_record_locking.rb:13:in `_update_row'
== 20200705232821 AddAndSeedMyColumn: migrated (0.1706s) =====================
```
If you skip clearing the schema cache (`User.reset_column_information`), the column is not
used by ActiveRecord and the intended changes are not made, leading to the result below,
where `my_column` is missing from the query.
```shell
== 20200705232821 AddAndSeedMyColumn: migrating ==============================
D, [2020-07-06T00:37:12.483876 #130101 ] DEBUG -- : (0.2ms) BEGIN
D, [2020-07-06T00:37:12.521660 #130101 ] DEBUG -- : (0.4ms) SELECT COUNT(*) FROM "user"
-- add_column(:users, :my_column, :integer, {:default=>1})
D, [2020-07-06T00:37:12.523309 #130101 ] DEBUG -- : (0.8ms) ALTER TABLE "users" ADD "my_column" integer DEFAULT 1
-> 0.0016s
D, [2020-07-06T00:37:12.650641 #130101 ] DEBUG -- : AddAndSeedMyColumn::User Load (0.7ms) SELECT "users".* FROM "users" ORDER BY "users"."id" ASC LIMIT $1 [["LIMIT", 1000]]
D, [2020-07-06T00:37:12.653459 #130101 ] DEBUG -- : AddAndSeedMyColumn::User Update (0.5ms) UPDATE "users" SET "updated_at" = $1 WHERE "users"."id" = $2 [["updated_at", "2020-07-05 23:37:12.652297"], ["id", 1]]
D, [2020-07-06T00:37:12.653648 #130101 ] DEBUG -- : ↳ config/initializers/config_initializers_active_record_locking.rb:13:in `_update_row'
== 20200705232821 AddAndSeedMyColumn: migrated (0.1706s) =====================
```
2021-03-05 18:09:17 +00:00
## High traffic tables
2021-03-08 15:08:54 +00:00
Here's a list of current [high-traffic tables ](https://gitlab.com/gitlab-org/gitlab/-/blob/master/rubocop/rubocop-migrations.yml ).
2021-03-05 18:09:17 +00:00
Determining what tables are high-traffic can be difficult. Self-managed instances might use
2021-03-08 15:08:54 +00:00
different features of GitLab with different usage patterns, thus making assumptions based
2021-03-05 18:09:17 +00:00
on GitLab.com not enough.
To identify a high-traffic table for GitLab.com the following measures are considered.
Note that the metrics linked here are GitLab-internal only:
- [Read operations ](https://thanos.gitlab.net/graph?g0.range_input=2h&g0.max_source_resolution=0s&g0.expr=topk(500%2C%20sum%20by%20(relname )%20(rate(pg_stat_user_tables_seq_tup_read%7Benvironment%3D%22gprd%22%7D%5B12h%5D)%20%2B%20rate(pg_stat_user_tables_idx_scan%7Benvironment%3D%22gprd%22%7D%5B12h%5D)%20%2B%20rate(pg_stat_user_tables_idx_tup_fetch%7Benvironment%3D%22gprd%22%7D%5B12h%5D)))& g0.tab=1)
2021-03-29 18:09:37 +00:00
- [Number of records ](https://thanos.gitlab.net/graph?g0.range_input=2h&g0.max_source_resolution=0s&g0.expr=topk(500%2C%20max%20by%20(relname )%20(pg_stat_user_tables_n_live_tup%7Benvironment%3D%22gprd%22%7D))& g0.tab=1)
- [Size ](https://thanos.gitlab.net/graph?g0.range_input=2h&g0.max_source_resolution=0s&g0.expr=topk(500%2C%20max%20by%20(relname )%20(pg_total_relation_size_bytes%7Benvironment%3D%22gprd%22%7D))& g0.tab=1) is greater than 10 GB
2021-03-05 18:09:17 +00:00
Any table which has some high read operation compared to current [high-traffic tables ](https://gitlab.com/gitlab-org/gitlab/-/blob/master/rubocop/rubocop-migrations.yml#L4 ) might be a good candidate.
2021-08-09 09:22:41 +00:00
As a general rule, we discourage adding columns to high-traffic tables that are purely for
analytics or reporting of GitLab.com. This can have negative performance impacts for all
self-managed instances without providing direct feature value to them.