Very slow CVMFS client startup when rlimit_nofile is too high

[gman0]

CVMFS client startup time is proportional to rlimit_nofile. When it forks to run exec(), it will first try to close all file descriptors, from 3 up until the maximum possible fd number. If rlimit_nofile is too high, this may take a very long time.

xref: cvmfs/cvmfs#3158
See details in https://cernvm-forum.cern.ch/t/cvmfs-config-setup-hangs/275

This issue will be kept open until there is a more permanent fix in the client itself. Until then, these are some of the possible workarounds:

• set node-wide rlimit_nofile to a reasonable number
• set rlimit_nofile in container runtime
• run ulimit -n <value> inside cvmfs-csi nodeplugin container (deployment: temporary set rlimit_nofile #58)

[gman0]

This was fixed in cvmfs/cvmfs#3182. Let's close this issue once we release an image with CVMFS client containing this fix.

[gman0]

#117 removes the workaround. The client version we're using already contains the fix. Closing.