I Tried Recovering a Deleted User Using Point-in-Time Recovery (PITR) in Amazon Cognito

I Tried Recovering a Deleted User Using Point-in-Time Recovery (PITR) in Amazon Cognito

Using Amazon Cognito point-in-time recovery (PITR), I restored deleted users. Their sub, password, group membership and TOTP MFA came back unchanged, while users created after the recovery point were deleted. Writes were rejected during verification.
2026.11.19

This page has been translated by machine translation. View original

Introduction

On November 18, 2026, Amazon Cognito added support for point-in-time recovery (PITR).

https://aws.amazon.com/jp/about-aws/whats-new/2026/11/cognito-point-in-time-recovery/

PITR is a feature that restores the user data of a user pool to its state at a past point in time. In this article, I deleted and modified users and groups in an Essentials plan user pool, and then recovered them with the AWS CLI.

Prerequisites

PITR is enabled automatically for Essentials and Plus user pools. The range you can specify for the recovery point is the past 1 day for Essentials and the past 14 days for Plus.

I used AWS CLI 2.41.1.

The IAM policy must allow the following five actions on the user pool resource. All of them are included in the AmazonCognitoPowerUser managed policy.

cognito-idp:InitiatePointInTimeRecovery
cognito-idp:DescribePointInTimeRecovery
cognito-idp:StartPointInTimeRecovery
cognito-idp:CancelPointInTimeRecovery
cognito-idp:ListPointInTimeRecoveries

https://docs.aws.amazon.com/cognito/latest/developerguide/user-pool-point-in-time-recovery.html

Test Data

I created users alice, bob, carol, and dave together with the groups admins and editors, and then performed the following operations.

User State at creation Operation after creation
alice Group admins, TOTP MFA Deleted
bob Group editors Deleted
carol No group Updated name to carol-updated, changed password
dave No group No changes
eve Does not exist Created (group temp)
frank Does not exist Created

After that, I deleted the group editors and created the group temp.

Commands to create the pool and app client

I created the user pool specifying Essentials, and enabled TOTP MFA as OPTIONAL.

aws cognito-idp create-user-pool --pool-name pitr-verify-1119 --user-pool-tier ESSENTIALS --deletion-protection INACTIVE
aws cognito-idp set-user-pool-mfa-config --user-pool-id <USER_POOL_ID> --mfa-configuration OPTIONAL --software-token-mfa-configuration Enabled=true
aws cognito-idp create-user-pool-client --user-pool-id <USER_POOL_ID> --client-name pitr-client --no-generate-secret --explicit-auth-flows ALLOW_USER_PASSWORD_AUTH ALLOW_REFRESH_TOKEN_AUTH

Recovery Steps

Specifying the Recovery Point and Previewing

For initiate, you pass the recovery point as a UTC timestamp (ending in Z). You must specify a time at least 15 minutes before the present. This time, I specified the time right after the creation of the 4 users and the groups.

aws cognito-idp initiate-point-in-time-recovery --user-pool-id <USER_POOL_ID> --restore-timestamp <RESTORE_TIMESTAMP_UTC>
{
    "Restoration": {
        "RestoreId": "<RESTORE_ID>",
        "RestoreStatus": "PREVIEW_GENERATING",
        "RestoreTimestamp": "2026-11-19T11:25:10+09:00",
        "TriggerTimestamp": "2026-11-19T11:51:12.629000+09:00"
    }
}

For the subsequent describe and start, pass the RestoreId from this response.

It took about 16 minutes from initiate until RestoreStatus became PREVIEW_GENERATED.

Use describe to check the preview counts.

aws cognito-idp describe-point-in-time-recovery --user-pool-id <USER_POOL_ID> --restore-id <RESTORE_ID>
{
    "Restoration": {
        "RestoreId": "<RESTORE_ID>",
        "RestoreStatus": "PREVIEW_GENERATED",
        "RestoreTimestamp": "2026-11-19T11:25:10+09:00",
        "TriggerTimestamp": "2026-11-19T11:51:12.629000+09:00",
        "RestorePreview": [
            {
                "Entity": "Passkeys",
                "CreateCount": 0,
                "UpdateCount": 0,
                "DeleteCount": 0
            },
            {
                "Entity": "User MFAs",
                "CreateCount": 1,
                "UpdateCount": 0,
                "DeleteCount": 0
            },
            {
                "Entity": "Users",
                "CreateCount": 2,
                "UpdateCount": 1,
                "DeleteCount": 2
            },
            {
                "Entity": "Groups",
                "CreateCount": 1,
                "UpdateCount": 0,
                "DeleteCount": 1
            }
        ]
    }
}

The CreateCount for Users is the number of deleted users being recreated, and the DeleteCount is the number of users created after the recovery point that will be deleted. The actual differences are as follows.

Item Count Actual difference
Users create 2 Recreate alice and bob
Users update 1 carol
Users delete 2 eve, frank
Groups create 1 editors
Groups delete 1 temp
User MFAs create 1 alice's TOTP
Passkeys All 0 None

Although admins lost a member because alice was deleted, the UpdateCount for Groups is 0.

Behavior While Reviewing the Preview

These are the results of performing writes, reads, and sign-ins against the user pool while in PREVIEW_GENERATED.

Operation Request type Result
AdminCreateUser, AdminUpdateUserAttributes, CreateGroup Signed request ValidationException: Cannot modify user pool resources while point-in-time recovery is in progress.
ListUsers, AdminGetUser Signed request Success
InitiateAuth (dave, who has no MFA configured) Request from the user Success
UpdateUserAttributes, ChangePassword, SignUp Request from the user ForbiddenException: A required service is currently unavailable. If this issue persists, contact your administrator.

The error message for requests from users does not reveal that a recovery is in progress.

Starting the Recovery

After checking the preview, run start.

aws cognito-idp start-point-in-time-recovery --user-pool-id <USER_POOL_ID> --restore-id <RESTORE_ID>

The RestoreStatus in the start response was RESTORING. I first confirmed RESTORE_COMPLETED about 70 seconds after start.

Recovery Results

What Was Restored

Item At the recovery point Just before recovery (after changes) After recovery
alice's sub 5774ead8… No user 5774ead8…
bob's sub 47342a18… No user 47342a18…
alice's sign-in Success (with TOTP challenge) UserNotFoundException Succeeded with the original password. Completed the challenge with a TOTP code registered before the recovery point
bob's sign-in Success UserNotFoundException Succeeded with the original password
Group membership alice in admins, bob in editors Neither alice nor bob exists alice in admins, bob in editors
Group list admins, editors admins, temp admins, editors
carol's name carol carol-updated carol
eve Does not exist Exists UserNotFoundException
frank Does not exist Exists UserNotFoundException
dave Exists No changes No changes

Changed Passwords Are Preserved

As stated in the Developer Guide, PITR does not revert password changes. The user who changed the password (carol) was able to sign in with the changed password.

Summary

Amazon Cognito user pools now support PITR.

Previously, to recover user data after an operational mistake, you had no choice but to use a backup and recovery mechanism you prepared yourself. Recovered users had to reset their passwords, and their sub values were also new ones. With PITR, you can now restore to a specified point in time while preserving passwords and sub values.

It is effective as a means of recovering user data in production environments, but pool settings, app clients, IdPs, and so on are out of scope. In addition, writes to the user pool are rejected while the preview is being reviewed, and generating the preview took about 16 minutes this time.

I recommend trying recovery in an evaluation environment beforehand, understanding the impact and the time required, and then using it.

Share this article

AWSのお困り事はクラスメソッドへ