# AWS says a DNS race condition in DynamoDB caused the October 2025 outage

> AWS says a latent race condition in its DynamoDB DNS system left the us-east-1 endpoint with an empty DNS record on Oct. 19, 2025.

*AWS says the failure spread to EC2 and Network Load Balancer in the same Region, and the event ran from 11:48 PM PDT on Oct. 19 to 2:20 PM PDT on Oct. 20.*

- Type: News
- Section: Tech
- By: Tech Desk
- Published: 2026-11-20T11:39:01-05:00
- Updated: 2026-11-20T11:39:01-05:00
- Checked: 2026-11-20T11:39:01-05:00
- URL: https://factsdesk.com/tech/what-caused-the-aws-outage-in-october-2025/

## Key facts

- AWS says the cause was a latent race condition in the DynamoDB DNS management system that left its us-east-1 regional endpoint with an incorrect empty DNS record. Source: [Amazon Web Services](https://aws.amazon.com/message/101925/). Confirmed 2026-11-20T11:39:01-05:00. Cite: https://factsdesk.com/tech/what-caused-the-aws-outage-in-october-2025/#aws-says-the-cause-was-a-latent-race-condition-in-the
- AWS says the event started at 11:48 PM PDT on Oct. 19, 2025 and ended at 2:20 PM PDT on Oct. 20, 2025, with three distinct periods of impact to customer applications. Source: [Amazon Web Services](https://aws.amazon.com/message/101925/). Confirmed 2026-11-20T11:39:01-05:00. Cite: https://factsdesk.com/tech/what-caused-the-aws-outage-in-october-2025/#aws-says-the-event-started-at-11-48-pm-pdt-on-oct-19-2025
- AWS says DynamoDB API error rates rose from 11:48 PM PDT on Oct. 19 to 2:40 AM PDT on Oct. 20, and new EC2 instance launches failed from 2:25 AM to 10:36 AM PDT on Oct. 20. Source: [Amazon Web Services](https://aws.amazon.com/message/101925/). Confirmed 2026-11-20T11:39:01-05:00. Cite: https://factsdesk.com/tech/what-caused-the-aws-outage-in-october-2025/#aws-says-dynamodb-api-error-rates-rose-from-11-48-pm-pdt-on
- AWS says it has disabled the DynamoDB DNS Planner and DNS Enactor automation worldwide and will fix the race condition before turning that automation back on. Source: [Amazon Web Services](https://aws.amazon.com/message/101925/). Confirmed 2026-11-20T11:39:01-05:00. Cite: https://factsdesk.com/tech/what-caused-the-aws-outage-in-october-2025/#aws-says-it-has-disabled-the-dynamodb-dns-planner-and-dns

On Oct. 19, 2025, a latent race condition in the DynamoDB DNS management system left the service's us-east-1 regional endpoint with an incorrect, empty DNS record, and [AWS says](https://aws.amazon.com/message/101925/) that failure triggered the disruption that spread through its N. Virginia (us-east-1) Region.

## What happened

[AWS says](https://aws.amazon.com/message/101925/) customers in the region could not connect to DynamoDB through its public endpoint from 11:48 PM PDT on Oct. 19, because the systems that needed the service hit DNS failures.

AWS describes its DNS system as two parts: a DNS Planner that writes plans of load balancers and weights, and a DNS Enactor that applies each plan to Amazon Route 53. [AWS says](https://aws.amazon.com/message/101925/) an Enactor that was unusually delayed applied an older plan just as another process cleaned up old plans. The older plan overwrote the newer one, and the clean-up then deleted it, which was now the active plan, so [all IP addresses for the regional endpoint were immediately removed](https://aws.amazon.com/message/101925/). The check that should have stopped the older plan was out of date by then, and the system needed manual operator intervention to correct.

[AWS also says](https://aws.amazon.com/message/101925/) the DynamoDB failure reached EC2. Its state checks depend on DynamoDB, so the leases that track physical servers timed out. After DynamoDB recovered at 2:25 AM PDT, engineers could not rebuild the leases fast enough, and at 4:14 AM they throttled incoming work and began selective restarts of the hosts that manage those leases. Full EC2 recovery came at 1:50 PM PDT on Oct. 20.

Delays in network changes for new EC2 instances then made [Network Load Balancer health checks fail on and off](https://aws.amazon.com/message/101925/), which removed capacity from DNS. Engineers disabled automatic health check failovers for the load balancer at 9:36 AM PDT and re-enabled them at 2:09 PM PDT. AWS also lists errors in Lambda, Amazon Connect, the AWS Support Center and IAM sign-in, and says the [full list of affected services](https://aws.amazon.com/message/101925/) is in its event history.

## The three periods of impact

[AWS gives three distinct periods of impact to customer applications](https://aws.amazon.com/message/101925/). All times are Pacific, as AWS states them.

| Service | What customers saw | Times, as AWS gives them |
| --- | --- | --- |
| Amazon DynamoDB | Increased API error rates; customers and other services could not establish new connections | 11:48 PM Oct. 19 to 2:40 AM Oct. 20 |
| Network Load Balancer | Increased connection errors for some load balancers | 5:30 AM to 2:09 PM Oct. 20 |
| Amazon EC2 | New instance launches failed; some new instances had connectivity issues | 2:25 AM to 10:36 AM Oct. 20 for failed launches; connectivity issues resolved by 1:50 PM |

## What AWS says it is changing

- DynamoDB: [AWS says it has already disabled the DynamoDB DNS Planner and DNS Enactor automation worldwide](https://aws.amazon.com/message/101925/). Before that automation is turned back on, AWS says it will fix the race condition and add protections against the application of incorrect DNS plans.
- Network Load Balancer: [AWS says it is adding a velocity control](https://aws.amazon.com/message/101925/) that limits how much capacity a single load balancer can remove when health check failures cause an Availability Zone failover.
- EC2: [AWS says it is building an additional test suite](https://aws.amazon.com/message/101925/) for the recovery workflow of its droplet lease system, and it will rate limit incoming work in its EC2 data propagation systems based on the size of the waiting queue.

## What the record does not yet say

AWS's summary gives no date for turning the DynamoDB DNS automation back on, and says that will happen only after the race condition is fixed. AWS also says it will keep working through the details of the event across all of its services and look for more ways to avoid a similar impact ([AWS summary](https://aws.amazon.com/message/101925/)).

## Background

AWS lists this summary in its [post-event summaries index](https://aws.amazon.com/premiumsupport/technology/pes/) under its title and the date Oct. 19, 2025. For the status codes that web servers send when a service is unavailable or does not respond in time, see [what the HTTP 503 status code means](/tech/what-does-the-http-503-status-code-mean/) and [what the HTTP 504 status code means](/tech/what-does-the-http-504-status-code-mean/).

## Sources

1. [Summary of the Amazon DynamoDB Service Disruption in the Northern Virginia (US-EAST-1) Region](https://aws.amazon.com/message/101925/), Amazon Web Services, 2025-10-19 (primary source)
2. [AWS post-event summaries](https://aws.amazon.com/premiumsupport/technology/pes/), Amazon Web Services, 2025-10-19

## Asked next

- [CrowdStrike says a Windows content update caused its July 2024 outage](https://factsdesk.com/tech/what-caused-the-crowdstrike-outage/): CrowdStrike's root cause analysis says a content update for its Windows sensor, released July 19, 2024, caused the widespread outage.
- [What does the HTTP 503 status code mean?](https://factsdesk.com/tech/what-does-the-http-503-status-code-mean/): HTTP 503 Service Unavailable means the server cannot handle the request now due to a temporary overload or scheduled maintenance, under RFC 9110.
- [What does the HTTP 504 status code mean?](https://factsdesk.com/tech/what-does-the-http-504-status-code-mean/): HTTP 504 Gateway Timeout means a server acting as a gateway or proxy did not get a timely response from an upstream server, under RFC 9110.
- [What is a botnet?](https://factsdesk.com/tech/what-is-a-botnet/): NIST defines a botnet as infected machines a criminal can remotely manage. CISA and the FBI list the signs of infection and the steps to remove one.
- [Is AI being used in cyber attacks?](https://factsdesk.com/tech/is-ai-being-used-in-cyber-attacks/): Yes, according to threat reports from Anthropic, OpenAI and Google, which say they found attackers misusing their AI models, and the UK NCSC's assessment.
- [What is zero trust?](https://factsdesk.com/tech/what-is-zero-trust/): Zero trust is a set of cybersecurity concepts for per-request access decisions on a network viewed as compromised. NIST SP 800-207 sets out its tenets.
