[PATCH v3] fix ctdb_mutex_ceph_rados_helper deadlock

[David Disseldorp]

Hi Amitay,

On Thu, 26 Jul 2018 18:50:08 +1000, Amitay Isaacs wrote:

> Is there a way I can test this without a real ceph cluster?  Or is
> there a way to fake a ceph cluster?
> 
> For CTDB, there is a way to run multiple daemons to test cluster set
> up without actually setting a cluster.  Is there something similar to
> create "fake" ceph cluster?

You could use the Ceph vstart.sh script to setup a "fake" Ceph cluster
on your desktop or laptop:
http://docs.ceph.com/docs/master/dev/quick_guide/#running-a-development-deployment

Samba needs to build against Ceph, which can be provided as an explicit
path via:
./configure --with-libcephfs=<Ceph_make_install_DESTDIR>/usr/local/ \
	    --enable-ceph-reclock --with-cluster-support

Once ctdb_mutex_ceph_rados_helper is build, you can run the test script
(test_ceph_rados_reclock.sh) from the Ceph build dir (which includes the
vstart generated keyring and ceph.conf files):
~/ceph/build (mimic)> export LD_LIBRARY_PATH=${PWD}/lib
~/ceph/build (mimic)> export PATH="${PWD}/bin/:~/samba/bin/default/ctdb/:${PATH}
~/ceph/build (mimic)> ~/samba/ctdb/utils/ceph/test_ceph_rados_reclock.sh
...
test_ceph_rados_reclock.sh: all tests passed

Note: you might want to set POOL="cephfs_metadata_a" at the top of
test_ceph_rados_reclock.sh, as vstart no longer creates an "rbd" pool by
default.

Feel free to ping me via irc, etc. if you have any further questions.

Cheers, David