- 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
- 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"
- 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"
Count processes again and monitor the initially referred dashboard