Suddenly, scm-manager (1.38) was using all the CPU.
And we had strange failures in TeamCity checkouts like :
remote: internal server error fatal: protocol error: bad pack headerThe problem was that our central bare repository had never been garbage collected (with git gc). Even running git gc was failing with the error
warning: packfile ... cannot be accessed fatal: failed to read object ...: Too many open files error: failed to run repackHere is what I did :
- bare clone the repo on another machine
- run git gc
- bare clone the previous repo to the original machine
- run git gc
- switch last repo with the broken one (et voilĂ )
Ref : https://groups.google.com/forum/#!topic/git-users/6XChKIqdG_U