RDS PostgreSQL resizing automation

Purpose

RDS GP2 volumes can only increase in size. Disk space running out is not the only or possibly even the main reason for increasing disk size; you get more burst credits for your volume by having a bigger disk. So, if you have lots of disk activity, you may need to increase the volume to avoid volume I/O from choking. This increase expenditure and going back to a smaller disk later is not supported by AWS.

The tools in this repository automate much of the RDS resizing process.

Requirements

The main requirement is Podman. You can run the resizing script without Podman, but you need to make sure that your environment is set up just right.

You can use the rhel9-setup-rds-resize.sh to set up an environment with Podman and rds-resize.

Run Steps

Create a configuration file

Create a ./src/config.yaml file based on the example. The settings are as follows:

master_rds_identifier: the name of the old RDS instance to get settings and databases from.
new_rds_identifier: the name of the new RDS instance you want to create.
allocated_storage: size of the volume allocated to the new RDS instance. If you feel the urge to run this script then this should be smaller than the old RDS instance's volume: increase volume size does not require this convoluted process.
max_allocated_storage: the maximum allocated storage. This affects RDS automatic storage size increases only.
psql_admin: name of the RDS admin user. All SQL operations run as this user.
reuse_new_rds: use an existing new RDS instance, if present. Useful when debugging data dump/restore issues orwhen you get connection interruptions during the process and would otherwise have to start from scratch.
databases: a list of databases to dump from the old database and restore to the new database.
accounts: list of hashes where the key is the user and the value is the password. These will be set on new database.
aws_access_key_id AWS API key id.
aws_secret_access_key: AWS API key.
aws_region: default region for aws.

Ensure that environment variables are set

It should be enough to use

workon <virtualenv>

Build the container image

You can build the container image with build.sh. You need to have the BOLT_ssh_key_file and EYAML_CONFIG environment variables set for build.sh to work: this should be the case if you're in the correct Python virtualenv.

$ build.sh <container-image-name>

The container image name can be whatever you want.

Get the endpoint for the old RDS instance

This can be done with

$ aws rds describe-db-instances

or from the AWS Console.

Ensure that RDS is reachable

Your RDS instance should not be publicly accessible in most cases. If that is the case, you probably have a VPN set up. So, make sure your VPN connection is up.

You can use netcat for example to test that your old RDS instance is reachable:

$ nc -z -v -w3 <old-rds-endpoint>:5432
Ncat: Version 7.93 ( https://nmap.org/ncat )
Ncat: Connected to 10.10.10.200:5432.
Ncat: 0 bytes sent, 0 bytes received in 0.26 seconds.

Make sure that nothing is connected to the databases

The script checks whether any of the databases your dumping have active connections and if so, it will stop. However, it is best to make sure that all connections are closed before you even start the script. Here's a one-liner to do it for a single database:

PGPASSWORD=<rds-admin-password> psql -h <old-rds-instance-endpoint> postgres <rds-admin-user> -c "SELECT count(*) FROM pg_stat_activity WHERE datname = '<database>';"

You should get

 count
-------
     0

in return if you successfully stopped all processes that used the database. If count is not 0, you have something still using the database. You can get the IP of that something with the following spell:

PGPASSWORD=<rds-admin-password> psql -h <old-rds-instance-endpoint> postgres <rds-admin-user> -c "SELECT datname,client_addr FROM pg_stat_activity;"

If the offending IP belongs to a Linux machine you can locate the offending process fairly easily:

$ netstat -a --program|grep postgres

Typically the connected process would be a system service or a container. In that case it would be best to gracefully stop it. Otherwise just kill it:

$ kill <pid>
$ kill -9 <pid>

Alternatively you can try to kill the connection from server side. For details see this article.

Once all connections to the database are down you can proceed with running the script.

Launch the RDS resizing container

Just do

$ ./run.sh <container-image-name>

This gives you a terminal with a pre-built container environment for working with RDS and database dump/restore procedures. The src folder is mounted as a volume on the container.

Run the resize script

Inside the container do

$ cd /rds
$ ./resize.py

Resize logs will be written to resize.log (/rds in the container, src on the host). It is recommended to tail the logs while resizing is in progress:

$ tail -f ./src/resize.log

Known issues

Harmless errors

You may get harmless but nasty-looking errors in the RDS resize logs:

ERROR:  must be superuser to alter superusers
ERROR:  Cannot issue GRANT on the "rds_superuser" role.

Or, you may get something like this:

ERROR: 08/10/2023 08:17:53 b'pg_restore: [archiver (db)]
  Error while PROCESSING TOC:
  pg_restore: [archiver (db)]
  Error from TOC entry 4970; 1262 16396 DATABASE mydb myuser
  pg_restore: [archiver (db)] could not execute query:
  ERROR:  database "mydb" already exists
  Command was: CREATE DATABASE mydb WITH TEMPLATE = template0
  ENCODING = \'UTF8\' LC_COLLATE = \'en_US.UTF-8\'
  LC_CTYPE = \'en_US.UTF-8\';\n\n\
  WARNING: errors ignored on restore: 1\n

Both error types seem to be harmless and do not indicate failure of the resizing operation.

klapcsik/rds-resize