[pbs-devel] [PATCH proxmox-backup 08/11] datastore: add manifest locking

[Fabian Grünbichler]

On October 15, 2020 7:25 am, Dietmar Maurer wrote:
> AFAIR flock with atomically replaced files has problems/races.
> 
> Unfortunately, I do not remember why. Somebody remember why? Or is it really safe?

you can obtain the flock a second time after replacing the original 
(still locked) file. using a separate lock file that is not touched 
other than for flocking solves this problem.

$ flock --verbose testfile -c 'sleep 60'
flock: getting lock took 0.000003 seconds
flock: executing /bin/zsh

while that is running and holding the lock:

$ flock --verbose --wait 5 testfile -c 'sleep 60'
flock: timeout while waiting to get lock

$ mv testfile2 testfile
$ flock --verbose --wait 5 testfile -c 'sleep 60'
flock: getting lock took 0.000003 seconds
flock: executing /bin/zsh