Restarting autofs on SL7 breaks CVMFS repositories

On SL6 machines with the CVMFS client installed (e.g. worker nodes), autofs can be restarted and CVMFS does not seem to be affected. However, on SL7, if the repository is in use while autofs restarted, any operations result in "Transport endpoint is not connected" errors, e.g.

[root@vm4 ~]# ls /cvmfs/cms.cern.ch/
ls: cannot open directory /cvmfs/cms.cern.ch/: Transport endpoint is not connected

I can reproduce this with 2.3.2 and 2.3.3, just by doing this on one terminal:

cd /cvmfs/cms.cern.ch/SITECONF

and then running "systemctl restart autofs" in another terminal. The repository only becomes useable again after running:

umount -l /cvmfs/cms.cern.ch

Obviously we generally don't restart autofs on worker nodes running jobs, but occasionally when our configuration management system does updates it results in systemd restarting autofs.

Here's an example from /var/log/messages of what happens on a worker node running jobs at the time autofs was restarted. In this case CVMFS seems to crash:

2017-02-28T16:55:23.026438+00:00 lcg1662 systemd: Stopping Automounts filesystems on demand...
2017-02-28T16:55:23.046421+00:00 lcg1662 cvmfs2: (lhcb.cern.ch) cache manager disappeared, aborting
2017-02-28T16:55:23.046701+00:00 lcg1662 cvmfs2: (cms.cern.ch) cache manager disappeared, aborting
2017-02-28T16:55:23.046953+00:00 lcg1662 cvmfs2: (alice.cern.ch) cache manager disappeared, aborting
2017-02-28T16:55:23.171872+00:00 lcg1662 cvmfs2: (grid.cern.ch) cache manager disappeared, aborting
2017-02-28T16:55:23.195837+00:00 lcg1662 cvmfs2: (atlas.cern.ch) cache manager disappeared, aborting
2017-02-28T16:55:23.227994+00:00 lcg1662 cvmfs2: (sft.cern.ch) cache manager disappeared, aborting
2017-02-28T16:55:28.810096+00:00 lcg1662 automount[1799]: umount_autofs_indirect: ask umount returned busy /cvmfs
2017-02-28T16:55:53.068270+00:00 lcg1662 cvmfs2: (lhcb.cern.ch) stack trace generation failed
2017-02-28T16:55:53.068583+00:00 lcg1662 cvmfs2: (alice.cern.ch) stack trace generation failed
2017-02-28T16:55:53.068834+00:00 lcg1662 cvmfs2: (lhcb.cern.ch) Signal 6, errno 0
2017-02-28T16:55:53.069071+00:00 lcg1662 cvmfs2: (alice.cern.ch) Signal 6, errno 0
2017-02-28T16:55:53.072047+00:00 lcg1662 cvmfs2: (cms.cern.ch) stack trace generation failed
2017-02-28T16:55:53.072282+00:00 lcg1662 cvmfs2: (cms.cern.ch) Signal 6, errno 0
2017-02-28T16:55:53.157243+00:00 lcg1662 cvmfs2: (grid.cern.ch) stack trace generation failed
2017-02-28T16:55:53.157518+00:00 lcg1662 cvmfs2: (grid.cern.ch) Signal 6, errno 0
2017-02-28T16:55:53.209107+00:00 lcg1662 cvmfs2: (grid.cern.ch) Backtrace (10 symbols):#012/lib64/libcvmfs_fuse.so(+0xb54b3) [0x7f5086ecc4b3]#012/lib64/libpthread.so.0(+0xf370) [0x7f5089609370]#012/lib64/libc.so.6(gsignal+0x37) [0x7f5088a4d1d7]#012/lib64/libc.so.6(abort+0x148) [0x7f5088a4e8c8]#012/lib64/libc.so.6(+0x2e146) [0x7f5088a46146]#012/lib64/libc.so.6(+0x2e1f2) [0x7f5088a461f2]#012/lib64/libcvmfs_fuse.so(+0xd12df) [0x7f5086ee82df]#012/lib64/libcvmfs_fuse.so(+0x1268e3) [0x7f5086f3d8e3]#012/lib64/libpthread.so.0(+0x7dc5) [0x7f5089601dc5]#012/lib64/libc.so.6(clone+0x6d) [0x7f5088b0f73d]
2017-02-28T16:55:53.209716+00:00 lcg1662 cvmfs2: (cms.cern.ch) Backtrace (10 symbols):#012/lib64/libcvmfs_fuse.so(+0xb54b3) [0x7f9cc2b524b3]#012/lib64/libpthread.so.0(+0xf370) [0x7f9cc528f370]#012/lib64/libc.so.6(gsignal+0x37) [0x7f9cc46d31d7]#012/lib64/libc.so.6(abort+0x148) [0x7f9cc46d48c8]#012/lib64/libc.so.6(+0x2e146) [0x7f9cc46cc146]#012/lib64/libc.so.6(+0x2e1f2) [0x7f9cc46cc1f2]#012/lib64/libcvmfs_fuse.so(+0xd12df) [0x7f9cc2b6e2df]#012/lib64/libcvmfs_fuse.so(+0x1268e3) [0x7f9cc2bc38e3]#012/lib64/libpthread.so.0(+0x7dc5) [0x7f9cc5287dc5]#012/lib64/libc.so.6(clone+0x6d) [0x7f9cc479573d]
2017-02-28T16:55:53.210123+00:00 lcg1662 cvmfs2: (grid.cern.ch) address of g_cvmfs_exports: 0x7f50873e0580
2017-02-28T16:55:53.210328+00:00 lcg1662 cvmfs2: (cms.cern.ch) address of g_cvmfs_exports: 0x7f9cc3066580
2017-02-28T16:55:53.210554+00:00 lcg1662 cvmfs2: (lhcb.cern.ch) Backtrace (10 symbols):#012/lib64/libcvmfs_fuse.so(+0xb54b3) [0x7fa2cae3a4b3]#012/lib64/libpthread.so.0(+0xf370) [0x7fa2cd577370]#012/lib64/libc.so.6(gsignal+0x37) [0x7fa2cc9bb1d7]#012/lib64/libc.so.6(abort+0x148) [0x7fa2cc9bc8c8]#012/lib64/libc.so.6(+0x2e146) [0x7fa2cc9b4146]#012/lib64/libc.so.6(+0x2e1f2) [0x7fa2cc9b41f2]#012/lib64/libcvmfs_fuse.so(+0xd12df) [0x7fa2cae562df]#012/lib64/libcvmfs_fuse.so(+0x1268e3) [0x7fa2caeab8e3]#012/lib64/libpthread.so.0(+0x7dc5) [0x7fa2cd56fdc5]#012/lib64/libc.so.6(clone+0x6d) [0x7fa2cca7d73d]
2017-02-28T16:55:53.210966+00:00 lcg1662 cvmfs2: (lhcb.cern.ch) address of g_cvmfs_exports: 0x7fa2cb34e580
2017-02-28T16:55:53.211185+00:00 lcg1662 cvmfs2: (alice.cern.ch) Backtrace (10 symbols):#012/lib64/libcvmfs_fuse.so(+0xb54b3) [0x7ff26fb6a4b3]#012/lib64/libpthread.so.0(+0xf370) [0x7ff2722a7370]#012/lib64/libc.so.6(gsignal+0x37) [0x7ff2716eb1d7]#012/lib64/libc.so.6(abort+0x148) [0x7ff2716ec8c8]#012/lib64/libc.so.6(+0x2e146) [0x7ff2716e4146]#012/lib64/libc.so.6(+0x2e1f2) [0x7ff2716e41f2]#012/lib64/libcvmfs_fuse.so(+0xd12df) [0x7ff26fb862df]#012/lib64/libcvmfs_fuse.so(+0x1268e3) [0x7ff26fbdb8e3]#012/lib64/libpthread.so.0(+0x7dc5) [0x7ff27229fdc5]#012/lib64/libc.so.6(clone+0x6d) [0x7ff2717ad73d]
2017-02-28T16:55:53.211606+00:00 lcg1662 cvmfs2: (alice.cern.ch) address of g_cvmfs_exports: 0x7ff27007e580
2017-02-28T16:55:56.209270+00:00 lcg1662 systemd: Starting Automounts filesystems on demand...
2017-02-28T16:55:56.249835+00:00 lcg1662 systemd: Started Automounts filesystems on demand.