Admin users always get 500 timeout in GitUtils.has_commit?, while normal users work fine

Community Building
1

Hi everyone,

I’m facing a strange issue with a fresh Discourse installation and I’d appreciate any guidance.

Environment

  • Discourse Docker (latest stable)
  • Ubuntu 24.04
  • Fresh installation using discourse_docker
  • No plugins except OpenID Connect
  • Git repository is clean
  • Reverse proxy with nginx
  • HTTP only (no HTTPS yet)
  • Local Git repository exists and is on latest branch

The strange behavior

Anonymous users:

  • :white_check_mark: Home page loads normally

Normal authenticated users:

  • :white_check_mark: Work normally

Users authenticated through OpenID Connect:

  • :white_check_mark: Work normally

But every local Discourse admin user gets a timeout and finally:



Oops

The software powering this discussion forum encountered an unexpected problem.

or sometimes



Server is currently experiencing high load.

What we tested

We created a brand new local admin using



rake admin:create

Result:

  • user as Admin → :cross_mark: 500 timeout
  • same user after changing


u.admin = false
u.save!

→ :white_check_mark: works immediately

So the issue is reproducible only when admin == true.

Log

The worker always times out after about 30 seconds.

Backtrace:



GitUtils.try_git
GitUtils.has_commit?
DiscourseUpdates.new_features
DiscourseUpdates.has_unseen_features?
CurrentUserSerializer
ApplicationLayoutPreloader

The timeout always happens here:



/var/www/discourse/lib/git_utils.rb:33
Kernel#`

Important observation

Running Git manually inside the container is instantaneous.

For example:



git rev-parse HEAD

returns in about 7 ms.

Also:



git merge-base --is-ancestor <current_commit> HEAD

returns immediately.

So Git itself is not slow.

Redis

new_features does not seem to exist:



redis-cli exists new_features latest_new_feature_created_at

(integer) 0

Question

Why would GitUtils.has_commit? block only for admin users while every Git command executes instantly from the shell?

Is there any known issue in recent versions of Discourse where DiscourseUpdates or new_features causes admin-only timeouts?

Any ideas on where to investigate next would be greatly appreciated.

Thanks!