What’s New in Release 2.17
There have been reports of our backpressure settings being too aggressive due to lack of nomad garbage collecting, as well as issues with the Vm Service overwhelming AWS API Limits. If you are on 2.17, you should upgrade to 2.17.1.
IMPORTANT We have upgraded Replicated to version 2.34.1 which requires Docker 17.12.1. Please follow the instructions provided in the Updating Server Replicated Version guide before upgrading.
- When using Github Enterprise, operators will now need to provide CircleCI with a Github User Token in the Management Console Settings. This is required due to a change in how the Github Enterprise API allows us to verify Organizations. It is recommended that you use a shared ops account to generate this token. After upgrading to CircleCI Server v2.17, please follow these steps to enact this change:
- Sign into your GitHub Enterprise instance.
- Navigate to Personal Settings (top right) > Developer Settings > Personal Access Tokens.
- Click “generate new token”. Name the token appropriately to prevent accidental deletion. Do not tick any of the checkboxes.
- Copy the new token and paste it into the Management Console settings. (Github Integration > GitHub Enterprise Default API Token)
- Save the settings in Replicated.
Notes and Best Practices
- Fill out this form to receive updates about CircleCI through email.
- It is currently a best practice to use a Services Machine with a minimum of 32GB of RAM. Starting in 2.18 it will become required. See docs for our recommendation(s).
- We have updated our software packages to the following versions. This is not a required update for those with externalized environments at this time, but will be when v2.18 is released.
- We are removing the 1.0 Single-Box options from CircleCI 2.0. We found a few critical vulnerabilities in our 1.0 build image, and we have long stopped recommending it for trials. If this is absolutely critical to your workflow please reach out to us. This does not impact people who are running 1.0 in clustered mode.
- Workflows now has a Slack Integration! Users can choose to receive Slack notifications when their workflows complete.
- Administrators can now restrict what organizations are allowed into their CircleCI installation. For more details on how to enable this feature, please see the User Management Section of the 2.17 Ops Manual.
- We changed the renamed org flow so orgs that have been renamed will no longer result in errors going forward. Existing users who had applied workarounds for this use case will now no longer require said workarounds.
- Workflows now take up less DB space and will be easier to maintain going forward.
- Improved the cache in front of GraphQL API resulting in overall improved performance.
- Added backpressure to avoid overwhelming nomad with requests, this will result in increased performance from existing nomad clusters.
New machine executor AMIs based on Ubuntu 16.04 for AWS.
- Ubuntu 16.04 with Docker 18.06.3 has apt-daily and apt-daily-upgrade services disabled.
- It is highly recommended that customers try to experiment with the AMIs below before officially switching:
- The new images are as follows
- They are replacing:
- Fixed some bugs related to GitHub API response handling and webhook handling.
- Fixed issue with Scheduled Workflows when the services machine is restarted.
- Fixed changing the RabbitMQ hostname for Scheduled Workflows when externalizing.
- You can no longer create contexts with no names. If you are using a context with no names, it will no longer be accessible from your execution environment.
- We have optimized our handling of large amounts of build output and test results XML, to avoid out-of-memory errors.
- The CIRCLE_PULL_REQUEST environment variable was not being populated correctly in all cases when building across forks. This has been fixed.
- Fixed a bug where some commits with [ci skip] in the message were still being built.
- Fixed a bug causing Workflows to get stuck when infrastructure_failure happens after a job fails.
- Fixed a bug causing duplicate docker networks on same nomad client (if running build using machine:true AND vm-provider=on_host).
- Improved performance when using local storage. Previously, caching issues had been experienced when local storage was used rather than the default option of using S3 (selecting None under Storage Driver options from the Management Console).
- We have added more error checking and validation around Github’s API so the existing list commit endpoint no longer causes issues.
- Datadog API token field was stored in plaintext, now set as a password field.
- Fixed issue where workflows were constrained from fanning out to large number of jobs.