Fast storage volume initialization is now supported for Amazon RDS

Fast storage volume initialization is now supported for Amazon RDS

We compared Amazon RDS fast storage volume initialization on RDS for MySQL with gp3 and gp2. For Multi-AZ conversion and read replica creation, gp3 and io2 Express instances can be up to 4 TiB/hour faster. Enabled by default, no extra cost.
2026.11.20

This page has been translated by machine translation. View original

Introduction

On November 18, 2026, Amazon RDS fast storage volume initialization was announced.

https://aws.amazon.com/about-aws/whats-new/2026/11/amazon-rds-fast-storage-volume-initialization/

For PITR, read replica creation, and conversion from Single-AZ to Multi-AZ, volumes are created from snapshots and data blocks are read from S3. According to the announcement, the gp3 and io2 Express volumes created by these operations are initialized at up to 4 TiB/hour. This feature is enabled by default and requires no configuration on the user's side.

The User Guide states that the target is gp3 and io2 Express volumes.

https://docs.aws.amazon.com/AmazonRDS/latest/UserGuide/CHAP_Storage.html#CHAP_Storage.initialization

In this article, I newly created one RDS for MySQL instance each on gp3 and gp2, and compared how initialization progressed when converting to Multi-AZ and creating a read replica.

Test Environment

I used RDS for MySQL 9.7.4 in the ap-northeast-1 region. The instance class was db.t4g.small and the capacity was 20 GiB. I newly created one instance with gp3 storage (3000 IOPS, 125 MiB/s) and one with gp2, both as Single-AZ.

Since the announcement does not describe a way to disable fast initialization, I used gp2, which is not a target, as the comparison.

Test Method

Creating the Instances and Writing Data

I created the instances with create-db-instance. The main items in the JSON passed to --cli-input-json are as follows. The gp2 JSON differs only in StorageType and the identifier.

{
  "DBInstanceIdentifier": "devio-init-gp3",
  "DBInstanceClass": "db.t4g.small",
  "Engine": "mysql",
  "EngineVersion": "9.7.4",
  "AllocatedStorage": 20,
  "StorageType": "gp3",
  "MultiAZ": false,
  "BackupRetentionPeriod": 1,
  "PubliclyAccessible": false
}
aws rds create-db-instance --cli-input-json file://create-gp3.json

After creation, I wrote the following data to both instances using the mysql client from a working EC2 instance in the same VPC.

CREATE TABLE t (id INT AUTO_INCREMENT PRIMARY KEY, pad VARBINARY(1000) NOT NULL) ENGINE=InnoDB;
INSERT INTO t(pad) VALUES (RANDOM_BYTES(1000));
-- Repeat the next statement 16 times (the number of rows doubles)
INSERT INTO t(pad) SELECT pad FROM t;
INSERT INTO t(pad) SELECT pad FROM t LIMIT 34464;

As a result, both had 100000 rows, and the table size was 112.6 MB.

Converting to Multi-AZ and Creating Read Replicas

I ran the Multi-AZ conversion on both instances at the same time.

aws rds modify-db-instance \
  --db-instance-identifier devio-init-gp3 \
  --multi-az \
  --apply-immediately

After Initializing disappeared following the conversion, I created read replicas for both instances at the same time. Network and tag options are omitted.

aws rds create-db-instance-read-replica \
  --db-instance-identifier devio-init-gp3-rr \
  --source-db-instance-identifier devio-init-gp3 \
  --db-instance-class db.t4g.small \
  --storage-type gp3 \
  --no-multi-az

For gp2, I changed the identifiers to devio-init-gp2 and devio-init-gp2-rr, and --storage-type to gp2.

Retrieving Progress

Progress can be tracked in the response of describe-db-instances. The values I checked were DBInstanceStatus, StorageOperationStatus, and StorageOperationPercentProgress. I ran the following command at intervals of about 30 seconds and retrieved all four instances together.

aws rds describe-db-instances \
  --query "DBInstances[?starts_with(DBInstanceIdentifier,'devio-init-')].[DBInstanceIdentifier,DBInstanceStatus,StorageType,MultiAZ,StorageOperationStatus,StorageOperationPercentProgress]" \
  --output text

I checked events for each instance with describe-events.

aws rds describe-events \
  --source-type db-instance \
  --source-identifier devio-init-gp3 \
  --duration 1440

Results

New Creation

When creating new Single-AZ instances, StorageOperationStatus never became Initializing for either gp3 or gp2. Initializing was first returned partway through the Multi-AZ conversion.

Conversion to Multi-AZ

The following is an excerpt of the output observed for gp3. The columns are, in order, observation time, identifier, DBInstanceStatus, storage type, MultiAZ, StorageOperationStatus, and StorageOperationPercentProgress. None means that the item was not in the response.

23:53:40 devio-init-gp3	modifying	gp3	False	Initializing	0
23:58:46 devio-init-gp3	available	gp3	True	Initializing	0
00:10:29 devio-init-gp3	available	gp3	True	None	None

I ran the conversion at 23:50:28. For gp3 and gp2, the time Initializing was first observed, the time it was last observed, and the time of the first observation after it disappeared were all the same.

Observation gp3 gp2
Initializing first observed 23:53:40 (modifying) 23:53:40 (modifying)
Last Initializing 00:09:59 (available) 00:09:59 (available)
First observation after Initializing disappeared 00:10:29 00:10:29

Initializing continued even after DBInstanceStatus became available. From the first observation until it disappeared, it took about 17 minutes for both.

In the RDS events, Finished applying modification to convert to a Multi-AZ DB Instance was recorded at 23:57:12 for both gp3 and gp2. Initializing then continued for about 13 minutes until the observation at 00:10:29. There were no events indicating the start or completion of initialization.

Creating Read Replicas

I created the read replicas at 00:12:33.

Observation gp3 gp2
Initializing first observed 00:16:37 (creating) 00:16:37 (creating)
available first observed 00:22:13 00:21:12
Last Initializing 00:35:29 (available) 00:35:29 (available)
First observation after Initializing disappeared 00:36:00 00:36:00

From the first observation until it disappeared, it took about 19 minutes for both gp3 and gp2.

In the events, Replication for the Read Replica resumed was recorded at 00:20:30 (gp3) and 00:20:27 (gp2). Initializing then continued for about 15 minutes.

Progress Values

In all observations where Initializing was returned, StorageOperationPercentProgress remained at 0 for both gp3 and gp2.

Summary

I measured the effect of Amazon RDS fast storage volume initialization through Multi-AZ conversion and read replica creation. In observations at about 30-second intervals in this environment (db.t4g.small, 20 GiB, about 112 MB of data), no difference was seen in the time taken for initialization between gp3 and gp2.

With gp3 and io2 Express volumes, fast storage volume initialization is available at no additional cost for Multi-AZ conversion, read replica creation, and PITR.

If you need to reduce the performance degradation immediately after a restore or replica creation, we recommend choosing a storage type that supports fast storage volume initialization, within a range that fits your budget.

Share this article

AWSのお困り事はクラスメソッドへ