2.22 Release notes

Requests for some file resources like a zip archive or raw file could enter a redirection loop.

A timeout could prevent some Issues and Pull Requests searches from providing complete search results.

Custom tabs with non-alphabetic characters in small screens did not render correctly.

An underlying behavior was causing failures when pushing content to a Git LFS-enabled repository.

In some rare cases issues could cause a 500 error when accessed via the web interface.

Authorization service was being detected as unhealthy due to a race condition in the bootstrap which led to restart of the service.

The Elasticsearch upgrade process was not getting captured by ghe-diagnostics.

Enabling GitHub Actions on an upgraded high availability configuration caused errors in replication.

An underlying behavior was causing a service to become unavailable during the hotpatch upgrade process.

Users connecting to an active replica would get an error connecting to the live updates websocket.

A subset of log forwarding SSL certificates was not being applied correctly.

Email notifications sent to suspended users when they were removed from a Team or an Organization.

The way SSH certificates were applied between Organizations and Businesses was inconsistent.

When an account was rate limited due to using incorrect passwords, it could be locked out for up to 24 hours.

Pull request synchronization on repositories with many references could cause worker queues to fall behind.

When signing in after attempting to visit a specific page, people were sent to the home page instead of their intended destination.

For GHES instances using built-in authentication with an internal SAML identity provider, users without an associated email address could not create a commit from the web interface.

The babeld logs were missing a separator between seconds and microseconds.

After upgrading GHES with a hotpatch, the ghe-actions-precheck and ghe-packages-precheck commands would fail with the error "docker load" accepts no arguments.

When the enterprise account "Repository visibility change" policy was set to "Enabled", organization owners could not change the visibility of repositories within the organization.

Audit logs could be attributed to 127.0.0.1 instead of the actual source IP address.

GitHub Actions could fail to start up successfully if it was previously enabled on an instance running 2.22.0 and was upgraded to 2.22.1 or 2.22.2.

Configuration files for GitHub Actions were not copied to the replica when setting up high availability replicas potentially leading to errors during ghe-repl-promote.

On a freshly set up 2.22.1 or 2.22.2 instance or after upgrading to 2.22.1 or 2.22.2, the activity feed on an organization's dashboard would not update.

Editing issues templates with filenames containing non-ASCII characters would fail with a "500 Internal Server Error".

A metric gathering method for background jobs increased CPU utilization. (updated 2020-11-03)

If the storage account settings failed to validate while configuring GitHub Actions, running ghe-actions-teardown was required before making a new attempt.

A custom proxy configuration could adversely affect the GitHub Actions environment.

On a change of an address on eth0, Nomad and Consul could get unresponsive.

When using self-signed certificates, GHES could have SSL validation exceptions upon configuring GitHub Actions.

Using a GitHub Action from a branch name with a + or / character resulted in an error: Unable to resolve action.

The enterprise account "Confirm two-factor requirement policy" messaging was incorrect.

On certain requests above 100MB, Kafka's buffer could be over-allocated.

The NameID Format dropdown in the Management Console would be reset to "unspecified" after setting it to "persistent".

Upgrading using a hotpatch could fail with an error: 'libdbi1' was not found

Saving settings via the management console would append a newline to the TLS/SSL certificate and key files which triggered unnecessary reloading of some services.

System logs for Dependency Graph were not rotating, allowing unbounded storage growth.

The MS SQL Server performance graph showed statistics from the primary instance even when a replica was selected.

ghe-actions-precheck would silently exit without running the storage checks if Actions was not enabled.

Upgrade could fail if the resqued workers override setting is in use.

Some services running in containers were not sending logs to the journal.

Links to GitHub Security Advisories would use a URL with the hostname of the GitHub Enterprise Server instance instead of GitHub.com, directing the user to a nonexistent URL.

When importing a repository with ghe-migrator, an unexpected exception could occur when inconsistent data is present.

The enterprise account security settings page showed a "View your organizations' current configurations" link for the "Two-factor authentication" setting when the authentication mode in use does not support built in two-factor authentication.

OAuth refresh tokens would be removed prematurely.

Search repair tasks would generate exceptions during the migration phase of configuration.

On the settings page for GitHub Apps, the "Beta Features" tab was not visible in some circumstances.

When using ghe-migrator to import PR review requests, records associated with deleted users would result in extraneous database records.

When importing users with ghe-migrator, an error of "Emails is invalid" would occur if the system-generated email address were longer than 100 characters.

Logging webhook activity could use large amounts of disk space and cause the root disk to become full.

Users experienced slower Git clone and fetch performance on an instance with high availability replicas due to reads being forwarded to a different node.

The repository Settings page of a repository for a user or organization GitHub Pages sites would fail with a "500 Internal Server Error".

Repository network maintenance operations could become stuck in a running state.

A repository being deleted immediately after uploading a code scanning result could cause a stall in the processing of code scanning results for all repositories.

When a large number of code scanning results were submitted at the same time, processing of batches could time out resulting in a stall in processing of code scanning results.

Creating a GitHub App from a manifest would fail.

GitHub usernames were changed unintentionally when using SAML authentication, when the GitHub username did not match the value of the attribute mapped to the username field in the Management Console.

The stafftools page for viewing pending collaborator showed a 500 Internal Server Error when there was a pending email invite.

The Repository Health Check in stafftools could give incorrect results on busy repositories.

A logged in user trying to accept an email invitation could get a 404 Not Found error.

If a user navigated to a repository whose name started with "repositories.", they were redirected to the owner's "Repositories" tab instead of landing on the repository overview page.

Labels in the dashboard timeline did not have enough contrast.