Workers under heavy load because of being used as a CDN

Symptoms

• skyrocket increase of load in workers
  • 
• decrease of connections
  • 
• decrease of database load
  • 
• increase of git http processes
  • 
• Relevant Dashboard where to find all these graphs
  • http://dashboards.gitlab.net/dashboard/db/fleet-overview

Possible checks

• Perform a count of all the http ssh processes that are a cat-file blob

knife ssh 'role:<cluster-role>' "ps -U git -o cmd | grep 'cat-file blob' | grep -v grep | sort | uniq -c -u"

Resolution

• Set the project as logically deleted to prevent requests from reaching the git layer.

sudo gitlab-rails console
proj = Project.find_with_namespace('group/project_name') # the path without the .git
proj.pending_delete = true
proj.save

• Kill them all

knife ssh 'role:<cluster-role>' "for p in \$(ps -U git -o pid,cmd | grep <object-id> | grep -v grep | awk '{ print \$1 }'); do kill -9 \$p; done"

Post Checks

Count processes again and monitor the initially referred dashboard