From e13624ef5b2171516979827dcbe7ff03eb8247e5 Mon Sep 17 00:00:00 2001

From: Tomas Jelinek <tojeline@redhat.com>

Date: Wed, 31 May 2017 07:39:23 +0200

Subject: [PATCH] squash 1176018 remote guest nodes crashes fixed

c7a24e6 fix adding a node to a stopped cluster

88ad6e6 fix 'pcs cluster restore' command for pcmk authkey

e4b768c fix crash of 'pcs cluster destroy --all'

6b15785 fix crash of 'pcs cluster setup --force'

---

 pcs/cluster.py | 33 +++++++++++++++++++++++++--------

 pcs/config.py  |  4 ++--

 2 files changed, 27 insertions(+), 10 deletions(-)

diff --git a/pcs/cluster.py b/pcs/cluster.py

index d64194d..b47db4a 100644

--- a/pcs/cluster.py

+++ b/pcs/cluster.py

@@ -425,9 +425,9 @@ def cluster_setup(argv):

     else:

         # verify and ensure no cluster is set up on the nodes

         # checks that nodes are authenticated as well

+        lib_env = utils.get_lib_env()

         if "--force" not in utils.pcs_options:

             all_nodes_available = True

-            lib_env = utils.get_lib_env()

             for node in primary_addr_list:

                 available, message = utils.canAddNodeToCluster(

                     lib_env.node_communicator(),

@@ -1757,9 +1757,12 @@ def node_add(lib_env, node0, node1, modifiers):

                 NodeAddressesList([node_addr]),

             )

+        # do not send pcmk authkey to guest and remote nodes, they either have

+        # it or are not working anyway

+        # if the cluster is stopped, we cannot get the cib anyway

         _share_authkey(

             lib_env,

-            get_nodes(lib_env.get_corosync_conf(), lib_env.get_cib()),

+            get_nodes(lib_env.get_corosync_conf()),

             node_addr,

             allow_incomplete_distribution=modifiers["skip_offline_nodes"]

         )

@@ -2112,15 +2115,29 @@ def cluster_reload(argv):

 # Code taken from cluster-clean script in pacemaker

 def cluster_destroy(argv):

     if "--all" in utils.pcs_options:

+        # destroy remote and guest nodes

+        cib = None

         lib_env = utils.get_lib_env()

-        all_remote_nodes = get_nodes(tree=lib_env.get_cib())

-        if len(all_remote_nodes) > 0:

-            _destroy_pcmk_remote_env(

-                lib_env,

-                all_remote_nodes,

-                allow_fails=True

+        try:

+            cib = lib_env.get_cib()

+        except LibraryError as e:

+            warn(

+                "Unable to load CIB to get guest and remote nodes from it, "

+                "those nodes will not be deconfigured."

             )

+        if cib is not None:

+            try:

+                all_remote_nodes = get_nodes(tree=cib)

+                if len(all_remote_nodes) > 0:

+                    _destroy_pcmk_remote_env(

+                        lib_env,

+                        all_remote_nodes,

+                        allow_fails=True

+                    )

+            except LibraryError as e:

+                utils.process_library_reports(e.args)

+        # destroy full-stack nodes

         destroy_cluster(utils.getNodesFromCorosyncConf())

     else:

         print("Shutting down pacemaker/corosync services...")

diff --git a/pcs/config.py b/pcs/config.py

index 94191e1..5526eb5 100644

--- a/pcs/config.py

+++ b/pcs/config.py

@@ -446,12 +446,12 @@ def config_backup_path_list(with_uid_gid=False, force_rhel6=None):

         "uname": settings.pacemaker_uname,

         "gname": settings.pacemaker_gname,

     }

-    pcmk_authkey_attrs = dict(cib_attrs)

-    pcmk_authkey_attrs["mode"] = 0o440

     if with_uid_gid:

         cib_attrs["uid"] = _get_uid(cib_attrs["uname"])

         cib_attrs["gid"] = _get_gid(cib_attrs["gname"])

+    pcmk_authkey_attrs = dict(cib_attrs)

+    pcmk_authkey_attrs["mode"] = 0o440

     file_list = {

         "cib.xml": {

             "path": os.path.join(settings.cib_dir, "cib.xml"),

-- 

1.8.3.1