File pacemaker-crmd-dont-fence-old-DC-if-shutting-do... of Package pacemaker

Overview Repositories Revisions Requests Users Attributes Meta

File pacemaker-crmd-dont-fence-old-DC-if-shutting-down-as-soon-to-be-DC-joins.patch of Package pacemaker

commit 249c7c9a83371a67e573276a285697994fe66fed
Author: Ken Gaillot <kgaillot@redhat.com>
Date:   Mon Apr 17 15:41:18 2017 -0500

Fix: crmd: don't fence old DC if it's shutting down as soon-to-be DC joins
    
    Existing peers of a DC that is shutting down can avoid fencing it (by updating
    its expected state) because it broadcasts its shutdown request. However, a
    newly joining node won't get that broadcast.
    
    Previously, if the joining node became the new DC, it would fence the old one.
    Now, the DC notifies joining nodes (via a join message field) whether it is
    shutting down, and joining nodes update its expected state accordingly.

diff --git a/crmd/join_client.c b/crmd/join_client.c
index 319272d0f..45104831b 100644
--- a/crmd/join_client.c
+++ b/crmd/join_client.c
@@ -30,6 +30,26 @@ void join_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, v
 
 extern ha_msg_input_t *copy_ha_msg_input(ha_msg_input_t * orig);
 
+/*!
+ * \internal
+ * \brief Remember if DC is shutting down as we join
+ *
+ * If we're joining while the current DC is shutting down, update its expected
+ * state, so we don't fence it if we become the new DC. (We weren't a peer
+ * when it broadcast its shutdown request.)
+ *
+ * \param[in] msg  A join message from the DC
+ */
+static void
+update_dc_expected(xmlNode *msg)
+{
+    if (fsa_our_dc && crm_is_true(crm_element_value(msg, F_CRM_DC_LEAVING))) {
+        crm_node_t *dc_node = crm_get_peer(0, fsa_our_dc);
+
+        crm_update_peer_expected(__FUNCTION__, dc_node, CRMD_JOINSTATE_DOWN);
+    }
+}
+
 /*	A_CL_JOIN_QUERY		*/
 /* is there a DC out there? */
 void
@@ -128,6 +148,8 @@ do_cl_join_offer_respond(long long action,
         return;
     }
 
+    update_dc_expected(input->msg);
+
     CRM_LOG_ASSERT(input != NULL);
     query_call_id =
         fsa_cib_conn->cmds->query(fsa_cib_conn, NULL, NULL, cib_scope_local | cib_no_children);
@@ -250,6 +272,8 @@ do_cl_join_finalize_respond(long long action,
         return;
     }
 
+    update_dc_expected(input->msg);
+
     /* send our status section to the DC */
     tmp1 = do_lrm_query(TRUE, fsa_our_uname);
     if (tmp1 != NULL) {
diff --git a/crmd/join_dc.c b/crmd/join_dc.c
index 999996d6c..ebb5059b6 100644
--- a/crmd/join_dc.c
+++ b/crmd/join_dc.c
@@ -106,6 +106,30 @@ initialize_join(gboolean before)
     }
 }
 
+/*!
+ * \internal
+ * \brief Create a join message from the DC
+ *
+ * \param[in] join_op  Join operation name
+ * \param[in] host_to  Recipient of message
+ */
+static xmlNode *
+create_dc_message(const char *join_op, const char *host_to)
+{
+    xmlNode *msg = create_request(join_op, NULL, host_to, CRM_SYSTEM_CRMD,
+                                  CRM_SYSTEM_DC, NULL);
+
+    /* Identify which election this is a part of */
+    crm_xml_add_int(msg, F_CRM_JOIN_ID, current_join_id);
+
+    /* Add a field specifying whether the DC is shutting down. This keeps the
+     * joining node from fencing the old DC if it becomes the new DC.
+     */
+    crm_xml_add_boolean(msg, F_CRM_DC_LEAVING,
+                        is_set(fsa_input_register, R_SHUTDOWN));
+    return msg;
+}
+
 static void
 join_make_offer(gpointer key, gpointer value, gpointer user_data)
 {
@@ -147,10 +171,8 @@ join_make_offer(gpointer key, gpointer value, gpointer user_data)
 
     crm_update_peer_join(__FUNCTION__, (crm_node_t*)member, crm_join_none);
 
-    offer = create_request(CRM_OP_JOIN_OFFER, NULL, member->uname,
-                           CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL);
+    offer = create_dc_message(CRM_OP_JOIN_OFFER, member->uname);
 
-    crm_xml_add_int(offer, F_CRM_JOIN_ID, current_join_id);
     /* send the welcome */
     crm_info("join-%d: Sending offer to %s", current_join_id, member->uname);
 
@@ -588,9 +610,7 @@ finalize_join_for(gpointer key, gpointer value, gpointer user_data)
     }
 
     /* send the ack/nack to the node */
-    acknak = create_request(CRM_OP_JOIN_ACKNAK, NULL, join_to,
-                            CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL);
-    crm_xml_add_int(acknak, F_CRM_JOIN_ID, current_join_id);
+    acknak = create_dc_message(CRM_OP_JOIN_ACKNAK, join_to);
 
     crm_debug("join-%d: ACK'ing join request from %s",
               current_join_id, join_to);
diff --git a/include/crm/msg_xml.h b/include/crm/msg_xml.h
index 7198fe506..750474415 100644
--- a/include/crm/msg_xml.h
+++ b/include/crm/msg_xml.h
@@ -64,6 +64,7 @@
 #  define F_CRM_ORIGIN			"origin"
 #  define F_CRM_USER			"crm_user"
 #  define F_CRM_JOIN_ID			"join_id"
+#  define F_CRM_DC_LEAVING      "dc-leaving"
 #  define F_CRM_ELECTION_ID		"election-id"
 #  define F_CRM_ELECTION_AGE_S		"election-age-sec"
 #  define F_CRM_ELECTION_AGE_US		"election-age-nano-sec"

Places

File pacemaker-crmd-dont-fence-old-DC-if-shutting-down-as-soon-to-be-DC-joins.patch of Package pacemaker

Places