Please describe the problem.
I am establishing running daily tests on annex on NFS mounted partition (one run with NFS as TMPDIR -- passes, another with NFS for HOME).
What version of git-annex are you using? On what operating system?
as of now git-annex-standalone_8.20200810+git47-g27329f0bb-1~ndall+1_amd64.deb
Please provide any additional information below.
...
All 840 tests passed (707.87s)
.t/gpgtmp/7: removePathForcibly:removePathForcibly:removePathForcibly:removeDirectory: unsatisfied constraints (Directory not empty)
sleeping 10 seconds and will retry directory cleanup
git-annex: .t/gpgtmp/7/.nfs0000000000669ff400000a6b: removePathForcibly:removePathForcibly:removePathForcibly:removePathForcibly:removeLink: resource busy (Device or resource busy)
##[error]Process completed with exit code 1.
so it seems to relate to gpg, and not unlikely to be out of git-annex control, but I wondered if it could potentially be some non-clean termination of underlying gpg process or alike which could/should actually be addressed on git-annex side?
This may depend on the version of gpg in use. With latest version, I see:
12 of these build up by the end of the test suite. But here when the test suite exits, they all terminate.
Presumably they have a deleted file they're holding open, which causes NFS to break in the ususal NFS way. Here they have an inotify handle, and some kind of socket. Based on the .nfs filename, which I think usually includes the name of the deleted file but here does not, I suspect it might be for the directory itself, and so due to the inotify handle.
Ah, I think I see.. When I delete the directory, gpg-agent notices and exits.
gpgconf --kill allcan be used to stop them, so I've made the test suide do that after using gpg.