Upgrade your PostgreSQL database using dump and restore

APPLIES TO: Azure Database for PostgreSQL - Single Server

You can upgrade your PostgreSQL server deployed in Azure Database for PostgreSQL by migrating your databases to a higher major version server using following methods.

• Offline method using PostgreSQL pg_dump and pg_restore which incurs downtime for migrating the data. This document addresses this method of upgrade/migration.
• Online method using Database Migration Service (DMS). This method provides a reduced downtime migration and keeps the target database in-sync with the source and you can choose when to cut-over. However, there are few prerequisites and restrictions to be addressed for using DMS. For details, see the DMS documentation.
• In-place Major Version Upgrade method using Azure Database for PostgreSQL - Flexible Server.In-place major version upgrade feature performs major version upgrade of the server with just a click. This simplifies the upgrade process minimizing the disruption to users and applications accessing the server. In-place upgrades are a simpler way to upgrade the major version of the instance, as they retain the server name and other settings of the current server after the upgrade, and don't require data migration or changes to the application connection strings. In-place upgrades are faster and involve shorter downtime than data migration.

The following table provides some recommendations based on database sizes and scenarios.

Database/Scenario	Dump/restore (Offline)	DMS (Online)
You have a small database and can afford downtime to upgrade	X
Small databases (< 10 GB)	X	X
Small-medium DBs (10 GB - 100 GB)	X	X
Large databases (> 100 GB)		X
Can afford downtime to upgrade (irrespective of the database size)	X
Can address DMS pre-requisites, including a reboot?		X
Can avoid DDLs and unlogged tables during the upgrade process?		X

This guide provides few offline migration methodologies and examples to show how you can migrate from your source server to the target server that runs a higher version of PostgreSQL.

Prerequisites for using dump and restore with Azure Database for PostgreSQL

To step through this how-to-guide, you need:

• A source PostgreSQL database server running a lower version of the engine that you want to upgrade.
• A target PostgreSQL database server with the desired major version Azure Database for PostgreSQL server - Single Server or Azure Database for PostgreSQL - Flexible Server.
• A PostgreSQL client system to run the dump and restore commands. It's recommended to use the higher database version. For example, if you're upgrading from PostgreSQL version 9.6 to 11, please use PostgreSQL version 11 client.
  • It can be a Linux or Windows client that has PostgreSQL installed and that has the pg_dump and pg_restore command-line utilities installed.
• Your PostgreSQL client preferably running in the same region as the source and target servers.

Additional details and considerations

• You can find the connection string to the source and target databases by selecting the "Connection Strings" from the portal.
• You may be running more than one database in your server. You can find the list of databases by connecting to your source server and running \l.
• Create corresponding databases in the target database server or add -C option to the pg_dump command which creates the databases.
• You must not upgrade azure_maintenance or template databases. If you have made any changes to template databases, you may choose to migrate the changes or make those changes in the target database.
• Refer to the tables above to determine the database is suitable for this mode of migration.
• If you want to use Azure Cli, please note that the session times out after 20 minutes. If your database size is < 10 GB, you may be able to complete the upgrade without the session timing out. Otherwise, you may have to keep the session open by other means, such as pressing any key once in 10-15 minutes.

Example database used in this guide

In this guide, the following source and target servers and database names are used to illustrate with examples.

Description	Value
Source server (v9.5)	pg-95.postgres.database.chinacloudapi.cn
Source database	bench5gb
Source database size	5 GB
Source user name	pg@pg-95
Target server (v11)	pg-11.postgres.database.chinacloudapi.cn
Target database	bench5gb
Target user name	pg@pg-11

Upgrade your databases using offline migration methods

You may choose to use one of the methods described in this section for your upgrades. You can use the following tips while performing the tasks.

• If you're using the same password for source and the target database, you can set the PGPASSWORD=yourPassword environment variable. Then you don't have to provide password every time you run commands like psql, pg_dump, and pg_restore. Similarly you can setup additional variables like PGUSER, PGSSLMODE etc. see to PostgreSQL environment variables.

• If your PostgreSQL server requires TLS/SSL connections (on by default in Azure Database for PostgreSQL servers), set an environment variable PGSSLMODE=require so that the pg_restore tool connects with TLS. Without TLS, the error may read FATAL: SSL connection is required. Please specify SSL options and retry.

• In the Windows command line, run the command SET PGSSLMODE=require before running the pg_restore command. In Linux or Bash run the command export PGSSLMODE=require before running the pg_restore command.

Migrate the Roles

Roles (Users) are global objects and needed to be migrated separately to the new cluster before restoring the database(s). You can use pg_dumpall binary with -r (--roles-only) option to dump them. To dump all the roles with passwords from the source Single Server:

pg_dumpall -r --host=mySourceServer --port=5432 --username=myUser@mySourceServer --database=mySourceDB > roles.sql

To dump all the roles names, without passwords from the source Flexible Server:

pg_dumpall -r --no-role-passwords --host=mySourceServer --port=5432 --username=myUser --database=mySourceDB > roles.sql

Edit the roles.sql and remove references of NOSUPERUSER and NOBYPASSRLS before restoring the content using psql in the target server:

psql -f roles.sql --host=myTargetServer --port=5432 --username=myUser --dbname=postgres

The dump script shouldn't be expected to run completely without errors. In particular, because the script will issue CREATE ROLE for every role existing in the source cluster, it's certain to get a "role already exists" error for the bootstrap superuser like azure_pg_admin or azure_superuser. This error is harmless and can be ignored. Use of the --clean option is likely to produce additional harmless error messages about non-existent objects, although you can minimize those by adding --if-exists.

Method 1: Using pg_dump and psql

This method involves two steps. First is to dump a SQL file from the source server using pg_dump. The second step is to import the file to the target server using psql. Please see the Migrate using export and import documentation for details.

Method 2: Using pg_dump and pg_restore

In this method of upgrade, you first create a dump from the source server using pg_dump. Then you restore that dump file to the target server using pg_restore. Please see the Migrate using dump and restore documentation for details.

Method 3: Using streaming the dump data to the target database

The database dump is streamed directly to the target database server and doesn't store the dump in the client. Hence, this can be used with a client with limited storage.

1. Make sure the database exists in the target server using \l command. If the database doesn't exist, then create the database.

    psql "host=myTargetServer port=5432 dbname=postgres user=myUser password=###### sslmode=mySSLmode"
   

   postgres> \l   
   postgres> create database myTargetDB;
   

2. Run the dump and restore as a single command line using a pipe.

   pg_dump -Fc --host=mySourceServer --port=5432 --username=myUser --dbname=mySourceDB | pg_restore  --no-owner --host=myTargetServer --port=5432 --username=myUser --dbname=myTargetDB
   

   For example,

   pg_dump -Fc --host=pg-95.postgres.database.chinacloudapi.cn --port=5432 --username=pg@pg-95 --dbname=bench5gb | pg_restore --no-owner --host=pg-11.postgres.database.chinacloudapi.cn --port=5432 --username=pg@pg-11 --dbname=bench5gb
   

3. Once the upgrade (migration) process completes, you can test your application with the target server.

4. Repeat this process for all the databases within the server.

As an example, the following table illustrates the time it took to migrate using streaming dump method. The sample data is populated using pgbench. As your database can have different number of objects with varied sizes than pgbench generated tables and indexes, it's highly recommended to test dump and restore of your database to understand the actual time it takes to upgrade your database.

Database Size	Approx. time taken
1 GB	1-2 minutes
5 GB	8-10 minutes
10 GB	15-20 minutes
50 GB	1-1.5 hours
100 GB	2.5-3 hours

Method 4: Using parallel dump and restore

You can consider this method if you have few larger tables in your database and you want to parallelize the dump and restore process for that database. You also need enough storage in your client system to accommodate backup dumps. This parallel dump and restore process reduces the time consumption to complete the whole migration. For example, the 50 GB pgbench database which took 1-1.5 hrs to migrate was completed using Method 1 and 2 took less than 30 minutes using this method.

1. For each database in your source server, create a corresponding database at the target server.

   psql "host=myTargetServer port=5432 dbname=postgres user=myuser password=###### sslmode=mySSLmode"
   

   postgres> create database myDB;
   

   For example,

   psql "host=pg-11.postgres.database.chinacloudapi.cn port=5432 dbname=postgres user=pg@pg-11 password=###### sslmode=require"
   psql (12.3 (Ubuntu 12.3-1.pgdg18.04+1), server 13.3)
   SSL connection (protocol: TLSv1.3, cipher: TLS_AES_256_GCM_SHA384, bits: 256, compression: off)
   Type "help" for help.
   
   postgres> create database bench5gb;
   postgres> \q
   

2. Run the pg_dump command in a directory format with number of jobs = 4 (number of tables in the database). With larger compute tier and with more tables, you can increase it to a higher number. That pg_dump will create a directory to store compressed files for each job.

   pg_dump -Fd -v --host=sourceServer --port=5432 --username=myUser --dbname=mySourceDB -j 4 -f myDumpDirectory
   

   For example,

   pg_dump -Fd -v --host=pg-95.postgres.database.chinacloudapi.cn --port=5432 --username=pg@pg-95 --dbname=bench5gb -j 4 -f dump.dir
   

3. Then restore the backup at the target server.

   $ pg_restore -v --no-owner --host=myTargetServer --port=5432 --username=myUser --dbname=myTargetDB -j 4 myDumpDir
   

   For example,

   $ pg_restore -v --no-owner --host=pg-11.postgres.database.chinacloudapi.cn --port=5432 --username=pg@pg-11 --dbname=bench5gb -j 4 dump.dir
   

Post upgrade/migrate

After the major version upgrade is complete, we recommend to run the ANALYZE command in each database to refresh the pg_statistic table. Otherwise, you may run into performance issues.

postgres=> analyze;
ANALYZE

Next steps

• After you're satisfied with the target database function, you can drop your old database server.
• For Azure Database for PostgreSQL - Single server only. If you want to use the same database endpoint as the source server, then after you had deleted your old source database server, you can create a read replica with the old database server name. Once the steady replication state is established, you can stop the replica, which will promote the replica server to be an independent server. See Replication for more details.