If the remote server in the middle of a backup crashes, will leave the repository in a state, so that it is not accessible any more.

Backtrace of a failed attempt shows this line:
Remote: borg.locking.LockTimeout: <path_to_repo>/lock.exclusive

Background: I'm currently evaluating borgbackup for a remote repository scenario. The remote server is running "borg serve". #One of my standard test cases is to crash or simulate a server crash and look how the systems recover from this failure.

Steps to reproduce:
Start backup run:
$ borg create someuser@localhost://...

On the same machine, but in another shell, while backup is still running:
$ ps ax | grep "borg serve" | grep python
$ kill -9

Any access to the repo after that will fail.

Manually removing the lock (rm -r <path_to_repo>/lock.exclusive) makes the repo accessible again. If access on the remote server is limited to "borg serve" only, so "rm" cannot be executed, this makes the repo completely inaccessible until the remote admin manually fixes the bug.