cephadm: implement zap-osds --fsid ... command

Signed-off-by: Sage Weil <sage@newdream.net>

diff --git a/src/cephadm/cephadm b/src/cephadm/cephadm
index 774512ea98aa28cde90ab1dccee2e0968e78c800..7195d599d80d84a8d276c1b89bac36e5e5ce1860 100755 (executable)
--- a/src/cephadm/cephadm
+++ b/src/cephadm/cephadm
@@ -2327,19 +2327,20 @@ def get_container_mounts(ctx, fsid, daemon_type, daemon_id,
             # these do not search for their keyrings in a data directory
             mounts[data_dir + '/keyring'] = '/etc/ceph/ceph.client.%s.%s.keyring' % (daemon_type, daemon_id)
 
-    if daemon_type in ['mon', 'osd']:
+    if daemon_type in ['mon', 'osd', 'clusterless-ceph-volume']:
         mounts['/dev'] = '/dev'  # FIXME: narrow this down?
         mounts['/run/udev'] = '/run/udev'
-    if daemon_type == 'osd':
+    if daemon_type in ['osd', 'clusterless-ceph-volume']:
         mounts['/sys'] = '/sys'  # for numa.cc, pick_address, cgroups, ...
+        mounts['/run/lvm'] = '/run/lvm'
+        mounts['/run/lock/lvm'] = '/run/lock/lvm'
+    if daemon_type == 'osd':
         # selinux-policy in the container may not match the host.
         if HostFacts(ctx).selinux_enabled:
             selinux_folder = '/var/lib/ceph/%s/selinux' % fsid
             if not os.path.exists(selinux_folder):
                 os.makedirs(selinux_folder, mode=0o755)
             mounts[selinux_folder] = '/sys/fs/selinux:ro'
-        mounts['/run/lvm'] = '/run/lvm'
-        mounts['/run/lock/lvm'] = '/run/lock/lvm'
 
     try:
         if ctx.shared_ceph_folder:  # make easy manager modules/ceph-volume development
@@ -5401,6 +5402,68 @@ def command_rm_daemon(ctx):
 ##################################
 
 
+def _zap(ctx, what):
+    mounts = get_container_mounts(ctx, ctx.fsid, 'clusterless-ceph-volume', None)
+    c = CephContainer(
+        ctx,
+        image=ctx.image,
+        entrypoint='/usr/sbin/ceph-volume',
+        envs=ctx.env,
+        args=['lvm', 'zap', '--destroy', what],
+        privileged=True,
+        volume_mounts=mounts,
+    )
+    logger.info(f'Zapping {what}...')
+    out, err, code = call_throws(ctx, c.run_cmd())
+
+
+@infer_image
+def _zap_osds(ctx):
+    # assume fsid lock already held
+
+    # list
+    mounts = get_container_mounts(ctx, ctx.fsid, 'clusterless-ceph-volume', None)
+    c = CephContainer(
+        ctx,
+        image=ctx.image,
+        entrypoint='/usr/sbin/ceph-volume',
+        envs=ctx.env,
+        args=['inventory', '--format', 'json'],
+        privileged=True,
+        volume_mounts=mounts,
+    )
+    out, err, code = call_throws(ctx, c.run_cmd())
+    if code:
+        raise Error('failed to list osd inventory')
+    try:
+        ls = json.loads(out)
+    except ValueError as e:
+        raise Error(f'Invalid JSON in ceph-volume inventory: {e}')
+
+    for i in ls:
+        matches = [lv.get('cluster_fsid') == ctx.fsid for lv in i.get('lvs', [])]
+        if any(matches) and all(matches):
+            _zap(ctx, i.get('path'))
+        elif any(matches):
+            lv_names = [lv['name'] for lv in i.get('lvs', [])]
+            # TODO: we need to map the lv_names back to device paths (the vg
+            # id isn't part of the output here!)
+            logger.warning(f'Not zapping LVs (not implemented): {lv_names}')
+
+
+def command_zap_osds(ctx):
+    if not ctx.force:
+        raise Error('must pass --force to proceed: '
+                    'this command may destroy precious data!')
+
+    lock = FileLock(ctx, ctx.fsid)
+    lock.acquire()
+
+    _zap_osds(ctx)
+
+##################################
+
+
 def command_rm_cluster(ctx):
     # type: (CephadmContext) -> None
     if not ctx.force:
@@ -5469,6 +5532,7 @@ def command_rm_cluster(ctx):
                 if os.path.exists(files[n]):
                     os.remove(files[n])
 
+
 ##################################
 
 
@@ -7634,6 +7698,18 @@ def _get_parser():
         'command', nargs=argparse.REMAINDER,
         help='command')
 
+    parser_zap_osds = subparsers.add_parser(
+        'zap-osds', help='zap all OSDs associated with a particular fsid')
+    parser_zap_osds.set_defaults(func=command_zap_osds)
+    parser_zap_osds.add_argument(
+        '--fsid',
+        required=True,
+        help='cluster FSID')
+    parser_zap_osds.add_argument(
+        '--force',
+        action='store_true',
+        help='proceed, even though this may destroy valuable data')
+
     parser_unit = subparsers.add_parser(
         'unit', help="operate on the daemon's systemd unit")
     parser_unit.set_defaults(func=command_unit)