An AWS Lambda function that runs pg_dump and streams the output to s3.
It can be configured to run periodically using CloudWatch events.
-
Create an AWS lambda function:
- Author from scratch
- Runtime: Node.js 20.x
- Architecture: x86_64
-
tab "Code" -> "Upload from" -> ".zip file":
- Upload (pgdump-aws-lambda.zip)
- tab "Configuration" -> "General Configuration" -> "Edit"
- Timeout: 15 minutes
- Edit the role and attach the policy "AmazonS3FullAccess"
- Save
-
Give your lambda permissions permissions to write to S3:
- tab "Configuration" -> "Permissions"
- click the existing Execution role
- "Add permissions" -> "Attach policies"
- select "AmazonS3FullAccess" and click "Add Permissions"
-
Test
- Create new test event, e.g.:
{ "PGDATABASE": "dbname", "PGUSER": "postgres", "PGPASSWORD": "password", "PGHOST": "host", "S3_BUCKET": "db-backups", "ROOT": "hourly-backups" }- Test and check the output
-
Create a CloudWatch rule:
- Event Source: Schedule -> Fixed rate of 1 hour
- Targets: Lambda Function (the one created in step #1)
- Configure input -> Constant (JSON text) and paste your config (as per previous step)
This function will store your backup with the following s3 key:
s3://${S3_BUCKET}${ROOT}/YYYY-MM-DD/YYYY-MM-DD_HH-mm-ss.backup
- If you run the Lambda function outside a VPC, you must enable public access to your database instance, a non VPC Lambda function executes on the public internet.
- If you run the Lambda function inside a VPC, you must allow access from the Lambda Security Group to your database instance. Also you must either add a NAT gateway (chargeable) to your VPC so the Lambda can connect to S3 over the Internet, or add an S3 VPC endpoint (free) and allow traffic to the appropriate S3 prefixlist.
You can add an encryption key to your event, e.g.
{
"PGDATABASE": "dbname",
"PGUSER": "postgres",
"PGPASSWORD": "password",
"PGHOST": "host",
"S3_BUCKET": "db-backups",
"ROOT": "hourly-backups",
"ENCRYPT_KEY": "c0d71d7ae094bdde1ef60db8503079ce615e71644133dc22e9686dc7216de8d0"
}The key should be exactly 64 hex characters (32 hex bytes).
When this key is present the function will do streaming encryption directly from pg_dump -> S3.
It uses the aes-256-cbc encryption algorithm with a random IV for each backup file. The IV is stored alongside the backup in a separate file with the .iv extension.
You can decrypt such a backup with the following bash command:
openssl enc -aes-256-cbc -d \
-in [email protected] \
-out [email protected] \
-K c0d71d7ae094bdde1ef60db8503079ce615e71644133dc22e9686dc7216de8d0 \
-iv $(< [email protected])If you experience lamba timeouts while uploading file parts to S3 you can try increasing the part size of each file chunk (might need to increase lambda resources). For instance on a 2GB file using the default part size of 5MB would result on ~400 parts, pushing all this parts was exceeding the 15min timeout for lambdas, by increasing the part size to 1GB the transmit time was reduced to ~3 minutes.
{
"S3_PART_SIZE": 1073741824,
}Your context may require that you use IAM-based authentication to log into the Postgres service. Support for this can be enabled my making your Cloudwatch Event look like this.
{
"PGDATABASE": "dbname",
"PGUSER": "postgres",
"PGHOST": "host",
"S3_BUCKET": "db-backups",
"ROOT": "hourly-backups",
"USE_IAM_AUTH": true
}If you supply USE_IAM_AUTH with a value of true, the PGPASSWORD var may be omitted in the CloudWatch event.
If you still provide it, it will be ignored.
If you prefer to not send DB details/credentials in the event parameters, you can store such details in SecretsManager and just provide the SecretId, then the function will fetch your DB details/credentials from the secret value.
NOTE: the execution role for the Lambda function must have access to GetSecretValue for the given secret.
Support for this can be enabled by setting the SECRETS_MANAGER_SECRET_ID, so your Cloudwatch Event looks like this:
{
"SECRETS_MANAGER_SECRET_ID": "my/secret/id",
"S3_BUCKET": "db-backups",
"ROOT": "hourly-backups"
}If you supply SECRETS_MANAGER_SECRET_ID, you can ommit the 'PG*' keys, and they will be fetched from your SecretsManager secret value instead with the following mapping:
| Secret Value | PG-Key |
|---|---|
| username | PGUSER |
| password | PGPASSWORD |
| dbname | PGDATABASE |
| host | PGHOST |
| port | PGPORT |
You can provide overrides in your event to any PG* keys as event parameters will take precedence over secret values.
If you'd like to export multiple databases in a single event, you can add a comma-separated list of database names to the PGDATABASE setting. The results will return in a list.
{
"PGDATABASE": "dbname1,dbname2,dbname3",
"PGUSER": "postgres",
"PGPASSWORD": "password",
"PGHOST": "host",
"S3_BUCKET": "db-backups",
"ROOT": "hourly-backups"
}NOTE: The 15 minute timeout for lambda still applies.
- Launch an EC2 instance with the Amazon Linux 2023 AMI (ami-0649bea3443ede307)
- Connect via SSH and:
# install packages required for building
sudo dnf install make automake gcc gcc-c++ readline-devel zlib-devel openssl-devel libicu-devel
# build and install postgres from source
wget https://ftp.postgresql.org/pub/source/v16.3/postgresql-16.3.tar.gz
tar zxf postgresql-16.3.tar.gz
cd postgresql-16.3
./configure --with-ssl=openssl
make
sudo make install
exitmkdir bin/postgres-16.3
scp ec2-user@your-ec2-server:/usr/local/pgsql/bin/pg_dump ./bin/postgres-16.3/pg_dump
scp ec2-user@your-ec2-server:/usr/local/pgsql/lib/libpq.so.5 ./bin/postgres-16.3/libpq.so.5- To use the new postgres binary pass PGDUMP_PATH in the event:
{
"PGDUMP_PATH": "bin/postgres-16.3"
}npm run makezip
Please submit issues and PRs.