Upgrading the Version of a Heroku Postgres Database
Last updated September 27, 2022
Table of Contents
This article describes how to upgrade the major PostgreSQL version of a Heroku Postgres database. The most recent major PostgreSQL version supported by Heroku is 14.
If you want to upgrade your database’s minor version, or if you only want to change your Postgres plan or underlying infrastructure, see Changing the Plan or Infrastructure of a Heroku Postgres Database.
Upgrading your major PostgreSQL version can be done only via the Heroku CLI. It is a significant operation that should be done with care.
Heroku supports two methods for upgrading. Both methods require some application downtime to ensure that no data is lost during the upgrade.
| Upgrade Method | Description |
|---|---|
| pg:upgrade | Works for all Heroku Postgres plans except hobby-tier plans. This method is recommended, unless:
Requires downtime of about 30 minutes, although this can vary. |
| pg:copy |
Works for all Heroku Postgres plans (including hobby-tier plans), but only for databases under 10GB in size.
Requires downtime of about 3 minutes per GB, although this can vary substantially. |
If you use Heroku Connect to sync Salesforce data with your database, see Heroku Connect’s article on upgrading the Heroku Postgres database version.
Data checksums
Postgres 9.3 introduced data checksums for early data corruption detection. If you initially provisioned your Heroku Postgres database with a version of PostgreSQL prior to 9.3, checksums are probably not enabled for it.
Check whether your database has checksums enabled (on) by running this query in a heroku pg:psql session:
-- `on` means checksums are enabled, `off` means they are not enabled
=> SHOW data_checksums;
data_checksums
----------------
on
(1 row)
If your database currently does not have checksums enabled and the downtime required for a pg:copy upgrade is acceptable for your app, use the pg:copy upgrade method to ensure that checksums are enabled as part of the upgrade.
Database bloat
If your database has a lot of “bloat” (extra space taken up by dead rows) and the downtime required for a pg:copy upgrade is acceptable for your app, use the pg:copy upgrade method to recreate all of the tables and indexes from scratch. Doing so removes all of the bloat from the database and saves disk space.
If downtime is a major consideration, then use pg:upgrade instead.
Upgrading with pg:upgrade
The pg:upgrade command uses the PostgreSQL pg_upgrade utility to upgrade your PostgreSQL version in-place. This command can only be used to upgrade a follower database, which will remain on the same plan but stop following the current leader.
If you need to upgrade both the PostgreSQL version and your plan, you can provision a new follower on a different plan and perform a pg:upgrade as part of the changeover process.
The pg:upgrade command can fail due to out-of-memory errors for databases with a large number of schemas. If your database has more than 1,000 schemas or stores more than 10,000 objects, test the upgrade in advance and open a Support ticket for assistance if this upgrade fails.
You can upgrade your database to versions 10, 11, 12, 13 and 14. You can specify the version that pg:upgrade should use with the --version flag (e.g., --version 11). If no --version flag is set, the upgrade will default to 14.
Performing a pg:upgrade requires app downtime on the order of 30 minutes. This method is supported for all Heroku Postgres plans except hobby-tier plans.
1. Provision a follower database
To begin, create a new follower for your database and wait for the follower to catch up to the leader database. In the example below the standard-2 plan is used, but you should provision the plan best suited for your needs.
$ heroku addons:create heroku-postgresql:standard-2 --follow HEROKU_POSTGRESQL_LAVENDER_URL --app example-app
Adding heroku-postgresql:standard-2 to example-app... done, v71 ($200/mo)
Attached as HEROKU_POSTGRESQL_WHITE
Follower will become available for read-only queries when up-to-date
Use `heroku pg:wait` to track status
$ heroku pg:wait
Waiting for database HEROKU_POSTGRESQL_WHITE_URL... available
The follower is generally considered “caught up” when it is within 200 commits of the primary database. You can check how many commits the follower is behind with the pg:info command (see the Behind By row of the follower database):
$ heroku pg:info --app example-app
=== HEROKU_POSTGRESQL_LAVENDER
Plan: Standard 0
Status: available
...
=== HEROKU_POSTGRESQL_WHITE
Plan: Standard 2
Status: available
...
Following: HEROKU_POSTGRESQL_LAVENDER (DATABASE_URL)
Behind By: 125 commits
2. Enter maintenance mode to prevent database writes
It is important that no new data is written to your current primary database during the upgrade process, because it will not be transferred to the new database. To accomplish this, place your app into maintenance mode. If you have scheduler jobs running as well, you should disable them. Databases continue to accrue billing hours while in maintenance mode.
Maintenance mode does not automatically scale down dynos. Web and any non-web dynos should be scaled down (e.g. heroku ps:scale worker=0) to ensure that no connections are writing data to the database.
Your application will be unavailable starting at this point in the upgrade process.
$ heroku maintenance:on --app example-app
Enabling maintenance mode for example-app... done
3. Upgrade the follower database
Now that you are in maintenance mode and no additional data is being written to the primary database, you can upgrade the follower database.
Wait for the follower database to fully catch up to the primary (as indicated by being behind by 0 commits).
$ heroku pg:info --app example-app
=== HEROKU_POSTGRESQL_LAVENDER_URL
Plan: Standard 0
Status: available
...
=== HEROKU_POSTGRESQL_WHITE_URL
Plan: Standard 2
Status: available
...
Following: HEROKU_POSTGRESQL_LAVENDER_URL (DATABASE_URL)
Behind By: 0 commits
If you do not wait for the follower to catch up then you may get an error message:
▸ database must not be too far behind leader, please wait until your follower catches up with its leader.
Once the follower is caught up, use the pg:upgrade command to update the PostgreSQL version of the follower in place. Upgrading also causes the follower to unfollow the primary database. This step usually requires about 20 minutes to complete.
$ heroku pg:upgrade HEROKU_POSTGRESQL_WHITE --app example-app
You can monitor the progress of the upgrade with pg:wait.
$ heroku pg:wait --app example-app
As part of the pg:upgrade process, Heroku Postgres runs ANALYZE on your database: this recalculates statistics for your database to make sure the Postgres query planner has up to date information even after a version upgrade.
Versions of Rails pre 5.0 have a known compatibility issue with Postgres 10 and above. To work around this, upgrade your version of Rails or use the monkey patch suggested in the Rails issue.
4. Promote the new database
Promote the newly upgraded database to set it as the primary database (DATABASE_URL) used by your application. pg:promote also creates an alternate attachment for the old primary database, assigned with a new HEROKU_POSTGRESQL_<color>_URL config var. The promotion process triggers a release and restarts the app.
$ heroku pg:promote HEROKU_POSTGRESQL_WHITE --app example-app
Promoting HEROKU_POSTGRESQL_WHITE_URL to DATABASE_URL... done
The follower database is now the primary database (although the application is not yet receiving new requests).
If your original primary database was attached to multiple apps, you need to attach your new database to those apps with heroku addons:attach.
After the promotion, followers of your original primary database do not automatically start to follow your new primary.
Create new followers for the new primary database as needed:
$ heroku addons:create heroku-postgresql:standard-0 --follow DATABASE_URL --app example-app
Be sure to deprovision your old followers after you no longer need them.
If your old primary was using connection pooling, and it was attached with the default name DATABASE_CONNECTION_POOL, the promote will reattach the connection pooler to the new primary under the same name DATABASE_CONNECTION_POOL.
Attachments under non-default names will not be reattached and you should activate connection pooling on your new primary if you wish to keep using connection pooling on your new primary with the same non-default name as the old primary:
$ heroku pg:connection-pooling:attach DATABASE_URL --as MY_DATABASE_CONNECTION_POOL -a example-app
5. Exit maintenance mode
To resume normal application operation, scale any non-web dynos back to their original levels (e.g., heroku ps:scale worker=1).
Finally, turn off maintenance mode:
$ heroku maintenance:off --app example-app
Your application is now receiving requests to your updated database instance. You can confirm this by running heroku pg:info. The database denoted by DATABASE_URL is considered the primary database.
If your Heroku Postgres database is not connected to a Heroku application, you need to retrieve the HEROKU_POSTGRESQL_WHITE_URL and update your application to use it as your primary database.
Upgrading with pg:copy
The pg:copy upgrade method uses native PostgreSQL backup and restore utilities. Instead of writing a database backup to disk, it streams the backup data directly to the restore process of a newly provisioned database.
The pg:copy method requires approximately 3 minutes of app downtime per GB of your current database, although this can vary substantially depending on your schema and database plan. You can estimate your required downtime by performing a dry run (perform the upgrade process on a new database without then promoting it).
The pg:copy method supports upgrades between all supported Heroku Postgres plans and versions. It is the only upgrade method that supports hobby-tier plans. pg:copy will only copy your default credential and the data it has access to. Any additional credentials and data that only they can access will not be copied.
For Hobby databases that use the scheduled backups feature, your schedules are lost after upgrading to production-tier plans. Set up your schedules again after the upgrade to avoid missing a backup.
1. Provision a new database
Provision a new Heroku Postgres database with your desired plan (in the example below the standard-0 plan is used, but you should provision the plan best suited for your needs):
$ heroku addons:create heroku-postgresql:standard-0 --app example-app
Adding heroku-postgresql:standard-0 on example-app... done, v122 ($50/mo)
The database should be available in 3-5 minutes
If you want to upgrade your database to a version of PostgreSQL other than than the most recent supported version, specify the version to use with the --version flag (e.g., --version 11).
Standard, Premium, Private, and Shield tiers take a few minutes to provision. You can use the pg:wait command to notify you when provisioning is complete:
$ heroku pg:wait -a example-app
Waiting for database HEROKU_POSTGRESQL_PINK_URL... available
2. Enter maintenance mode to prevent database writes
It is important that no new data is written to your current primary database during the upgrade process, because it will not be transferred to the new database. To accomplish this, place your app into maintenance mode. If you have scheduler jobs running as well, you should disable them.
Maintenance mode does not automatically scale down dynos. Web and any non-web dynos should be scaled down (e.g. heroku ps:scale worker=0) to ensure that no connections are writing data to the database.
Your application will be unavailable starting at this point in the upgrade process.
$ heroku maintenance:on -a example-app
Enabling maintenance mode for example-app... done
3. Transfer data to the new database
To copy data from your current database to the newly provisioned database, use the pg:copy command with the HEROKU_POSTGRESQL_COLOR name of your new database.
In this example, the DATABASE_URL is the source of the data in the transfer and HEROKU_POSTGRESQL_PINK is the target database.
$ heroku pg:copy DATABASE_URL HEROKU_POSTGRESQL_PINK --app example-app
! WARNING: Destructive Action
! Transfering data from DATABASE_URL to HEROKU_POSTGRESQL_PINK
! This command will affect the app: example-app
! To proceed, type "example-app" or re-run this command with --confirm example-app
> example-app
4. Promote the new database
At this point, the new database is populated with the data from the original database, but is not yet the active database for your application. To make the new upgraded database the primary database for your application, you promote it like so:
$ heroku pg:promote HEROKU_POSTGRESQL_PINK --app example-app
Promoting HEROKU_POSTGRESQL_PINK_URL to DATABASE_URL... done
The upgraded database is now the primary database (although the application is not yet receiving new requests).
If your original primary database was attached to multiple apps, you need to attach your new database to those apps with heroku addons:attach.
After the promotion, followers of your original primary database do not automatically start to follow your new primary.
Create new followers for the new primary database as needed:
$ heroku addons:create heroku-postgresql:standard-0 --follow DATABASE_URL -a example-app
Be sure to deprovision your old followers after you no longer need them.
5. Exit maintenance mode
To resume normal application operation, scale any non-web dynos back to their original levels (e.g., heroku ps:scale worker=1).
Finally, turn off maintenance mode:
$ heroku maintenance:off --app example-app
Your application is now receiving requests to your new database instance. You can confirm this by running heroku pg:info. The database denoted by DATABASE_URL is considered the primary database.
If your Heroku Postgres database is not connected to a Heroku application, you need to retrieve the HEROKU_POSTGRESQL_WHITE_URL and update your application to use it as your primary database.
Deprovisioning the old primary database
After you upgrade your database, be sure to deprovision your old primary database:
$ heroku addons:destroy HEROKU_POSTGRESQL_LAVENDER --app example-app
Dataclips that were associated with the old primary database must be reassigned to the new database. Follow the instructions on Dataclip recovery to resolve any recoverable Dataclips.
