Because the replication's purge_old_state() function now
fails if cfs_update() fails, and since it tries to access
the actual local pmxcfs it should never have been called by
tests anyway.
Commit 3385399339c94 ("replication: keep retrying every 30 minutes in
error state") changed the retry behavior to not stop after the 3rd error
and then stick to half-hour intervals. This needs to be reflected in the
tests. The numbers here match. (1900 + 30*60 = 3700).
Commit fd844180a7efa ("replication: don't sync to offline targets on
error states) changed the retry behavior to check whether the target
node is online. If this is not the case we fail right away. This
introduced a dependency on PVE::Cluster::get_members which we now need
to mock. Tests currently use node names "node{1,2,3}", so I just mock
those 3.
Otherwise the user running the tests may either create (and
end up being the ownderof) the system wide
/var/lock/pve-manager/* files, or the tests will fail (or
loop endlessly) if the user doesn't have access to them.
