Look Before You Leap – The Coming Leap Second and AWS (Updated)
My colleague Mingxue Zhao sent me a guest post designed to make sure that you are aware of an important time / clock issue.
Note: This post was first published on May 18, 2015. We made some important additions and corrections on May 25, 2015.
The International Earth Rotation and Reference Systems (IERS) recently announced that an extra second will be injected into civil time at the end of June 30th, 2015. This means that the last minute of June 30th, 2015 will have 61 seconds. If a clock is synchronized to the standard civil time, it should show an extra second 23:59:60 on that day between 23:59:59 and 00:00:00. This extra second is called a leap second. There have been 25 such leap seconds since 1972. The last one took place on June 30th, 2012.
Clocks in IT systems do not always follow the standard above and can behave in many different ways. For example:
- Some Linux kernels implement a one-seconds backwards jump instead of the extra “:60” second, repeating the 59th second (see the article, Resolve Leap Second Issues in Red Hat Enterprise Linux for more information).
- Windows time servers ignore the leap second signal and will sync to the correct time after the leap second (see How the Windows Time Service Treats a Leap Second for more information).
- Some organizations, including Amazon Web Services, plan to spread the extra second over many hours surrounding the leap second by making every second slightly longer.
- If a clock doesn’t connect to a time synchronization system, it drifts on its own and will not implement any leap second or an adjustment for it.
If you want to know whether your applications and systems can properly handle the leap second, contact your providers. If you run time-sensitive workloads and need to know how AWS clocks will behave, read this document carefully. In general, there are three affected parts:
- The AWS Management Console and backend systems
- Amazon EC2 instances
- Other AWS managed resources
For more information about comparing AWS clocks to UTC, see the AWS Adjusted Time section of this post.
AWS Management Console and Backend Systems
The AWS Management Console and backend systems will NOT implement the leap second. Instead, we will spread the one extra second over a 24-hour period surrounding the leap second by making each second slightly longer. During these 24 hours, AWS clocks may be up to 0.5 second behind or ahead of the standard civil time (see the AWS Adjusted Time section for more information).
You can see adjusted times in consoles (including resource creation timestamps), metering records, billing records, Amazon CloudFront logs, and AWS CloudTrail logs. You will not see a “:60” second in these places and your usage will be billed according to the adjusted time.
Amazon EC2 Instances
Each EC2 instance has its own clock and is fully under your control; AWS does not manage instance clocks. An instance clock can have any of the behaviors listed at the beginning of this post. Contact your OS provider to understand the expected behavior of your operating system.
If you use the Amazon Linux AMI, your instance will implement the one-second backwards jump and will see “23:59:59” twice. You may find the following information useful:
- Forum post: Amazon Linux AMI and the 2015-06-30 leap second.
- Configuring NTP for instances using Amazon Linux AMIs: Setting the Time for Your Linux Instance.
- Configuring NTP for instances using Amazon-provided Microsoft Windows AMIs: Setting the Time for a Windows Instance.
If you use SUSE Linux Enterprise Server, take a look at Fixes and Workaround to Avoid Issues Caused by Leap Second 2015.
Other AWS Managed Resources
Other AWS resources may also have their own clocks. Unlike EC2 instances, these resources are fully or partially managed by AWS.
The following resources will implement the one-second backwards jump and will see :23:59:59″ twice:
- Amazon CloudSearch clusters
- Amazon EC2 Container Service instances
- Amazon EMR Clusters
- Amazon RDS instances
- Amazon Redshift instances
To enable time synchronization on EMR clusters, your VPC has to allow access to NTP. Make sure that your EMR clusters have access to the Internet, and that your security groups and network ACLs allow outbound UDP traffic on port 123.
AWS Adjusted Time
This section provides specific details on how clocks will behave in the AWS Management Console and backend systems.
Starting at 12:00:00 PM on June 30th, 2015, we will slow down AWS clocks by 1/86400. Every second on AWS clocks will take 1+1/86400 seconds of “real” time, until 12:00:00 PM on July 1st, 2015, when AWS clocks will be behind by a full second. Meanwhile, the standard civil time (UTC) will implement the leap second at the end of June 30th, 2015 and fall behind by a full second, too. Therefore, at 12:00:00 PM July 1st, 2015, AWS clocks will be synchronized to UTC again. The table below illustrates these changes.
|UTC||AWS Adjusted Clock||AWS vs. UTC||Notes|
|11:59:59 AM June 30th, 2015||11:59:59 AM June 30th, 2015||+0||AWS clocks are synchronized to UTC.|
|12:00:00 PM||12:00:00 PM||+0|
|12:00:01||Each second is 1/86400 longer and AWS clocks fall behind UTC. The gap gradually increases to up to 1/2 second.|
|23:59:60||Leap second injected to UTC.|
|00:00:00 AM July 1st, 2015||-1/2||AWS clocks gain 1/2 second ahead of UTC.|
|00:00:00 AM July 1st, 2015||AWS clocks keep falling behind and the gap with UTC shrinks gradually.|
|12:00:00 PM July 1st ,2015||12:00:00 PM July 1st ,2015||+0||The gap shrinks to zero. AWS clocks synchronize to UTC again.|
— Mingxue Zhao, Senior Product Manager