From 8270e8aed46f6e672b94f00fe0bde07cd2b6ddd7 Mon Sep 17 00:00:00 2001

From: Ken Gaillot <kgaillot@redhat.com>

Date: Fri, 13 Dec 2019 11:38:49 -0600

Subject: [PATCH 10/18] Low: controller: don't clear shutdown locks when node

 rejoins

Add new controld_delete_node_state() values for clearing resource history

while preserving shutdown locks. This is accomplished by deleting all

unlocked lrm_resource entries and all lrm_rsc_op entries, instead of the entire

lrm subsection.

---

 daemons/controld/controld_based.c     | 22 +++++++++++++++++++++-

 daemons/controld/controld_join_dc.c   |  7 +++++--

 daemons/controld/controld_remote_ra.c | 16 ++++++++++------

 daemons/controld/controld_utils.h     |  2 ++

 4 files changed, 38 insertions(+), 9 deletions(-)

diff --git a/daemons/controld/controld_based.c b/daemons/controld/controld_based.c

index f3a7c4f..0ffc1e8 100644

--- a/daemons/controld/controld_based.c

+++ b/daemons/controld/controld_based.c

@@ -191,12 +191,21 @@ cib_delete_callback(xmlNode *msg, int call_id, int rc, xmlNode *output,

 // Node's lrm section (name 1x)

 #define XPATH_NODE_LRM          XPATH_NODE_STATE "/" XML_CIB_TAG_LRM

+// Node's lrm_rsc_op entries and lrm_resource entries without lock (name 2x)

+#define XPATH_NODE_LRM_UNLOCKED XPATH_NODE_STATE "//" XML_LRM_TAG_RSC_OP    \

+                                "|" XPATH_NODE_STATE                        \

+                                "//" XML_LRM_TAG_RESOURCE                   \

+                                "[not(@" XML_CONFIG_ATTR_SHUTDOWN_LOCK ")]"

+

 // Node's transient_attributes section (name 1x)

 #define XPATH_NODE_ATTRS        XPATH_NODE_STATE "/" XML_TAG_TRANSIENT_NODEATTRS

 // Everything under node_state (name 1x)

 #define XPATH_NODE_ALL          XPATH_NODE_STATE "/*"

+// Unlocked history + transient attributes (name 3x)

+#define XPATH_NODE_ALL_UNLOCKED XPATH_NODE_LRM_UNLOCKED "|" XPATH_NODE_ATTRS

+

 /*!

  * \internal

  * \brief Delete subsection of a node's CIB node_state

@@ -218,6 +227,11 @@ controld_delete_node_state(const char *uname, enum controld_section_e section,

             xpath = crm_strdup_printf(XPATH_NODE_LRM, uname);

             desc = crm_strdup_printf("resource history for node %s", uname);

             break;

+        case controld_section_lrm_unlocked:

+            xpath = crm_strdup_printf(XPATH_NODE_LRM_UNLOCKED, uname, uname);

+            desc = crm_strdup_printf("resource history (other than shutdown "

+                                     "locks) for node %s", uname);

+            break;

         case controld_section_attrs:

             xpath = crm_strdup_printf(XPATH_NODE_ATTRS, uname);

             desc = crm_strdup_printf("transient attributes for node %s", uname);

@@ -226,6 +240,12 @@ controld_delete_node_state(const char *uname, enum controld_section_e section,

             xpath = crm_strdup_printf(XPATH_NODE_ALL, uname);

             desc = crm_strdup_printf("all state for node %s", uname);

             break;

+        case controld_section_all_unlocked:

+            xpath = crm_strdup_printf(XPATH_NODE_ALL_UNLOCKED,

+                                      uname, uname, uname);

+            desc = crm_strdup_printf("all state (other than shutdown locks) "

+                                     "for node %s", uname);

+            break;

     }

     if (fsa_cib_conn == NULL) {

@@ -234,7 +254,7 @@ controld_delete_node_state(const char *uname, enum controld_section_e section,

     } else {

         int call_id;

-        options |= cib_quorum_override|cib_xpath;

+        options |= cib_quorum_override|cib_xpath|cib_multiple;

         call_id = fsa_cib_conn->cmds->remove(fsa_cib_conn, xpath, NULL, options);

         crm_info("Deleting %s (via CIB call %d) " CRM_XS " xpath=%s",

                  desc, call_id, xpath);

diff --git a/daemons/controld/controld_join_dc.c b/daemons/controld/controld_join_dc.c

index 885b2a9..f0eb2a2 100644

--- a/daemons/controld/controld_join_dc.c

+++ b/daemons/controld/controld_join_dc.c

@@ -534,6 +534,7 @@ do_dc_join_ack(long long action,

     int join_id = -1;

     int call_id = 0;

     ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg);

+    enum controld_section_e section = controld_section_lrm;

     const char *op = crm_element_value(join_ack->msg, F_CRM_TASK);

     const char *join_from = crm_element_value(join_ack->msg, F_CRM_HOST_FROM);

@@ -583,8 +584,10 @@ do_dc_join_ack(long long action,

     /* Update CIB with node's current executor state. A new transition will be

      * triggered later, when the CIB notifies us of the change.

      */

-    controld_delete_node_state(join_from, controld_section_lrm,

-                               cib_scope_local);

+    if (controld_shutdown_lock_enabled) {

+        section = controld_section_lrm_unlocked;

+    }

+    controld_delete_node_state(join_from, section, cib_scope_local);

     if (safe_str_eq(join_from, fsa_our_uname)) {

         xmlNode *now_dc_lrmd_state = controld_query_executor_state(fsa_our_uname);

diff --git a/daemons/controld/controld_remote_ra.c b/daemons/controld/controld_remote_ra.c

index 2d3dfa7..a81c354 100644

--- a/daemons/controld/controld_remote_ra.c

+++ b/daemons/controld/controld_remote_ra.c

@@ -1,5 +1,5 @@

 /*

- * Copyright 2013-2019 the Pacemaker project contributors

+ * Copyright 2013-2020 the Pacemaker project contributors

  *

  * The version control history for this file may have further details.

  *

@@ -177,17 +177,21 @@ remote_node_up(const char *node_name)

     int call_opt, call_id = 0;

     xmlNode *update, *state;

     crm_node_t *node;

+    enum controld_section_e section = controld_section_all;

     CRM_CHECK(node_name != NULL, return);

     crm_info("Announcing pacemaker_remote node %s", node_name);

-    /* Clear node's entire state (resource history and transient attributes).

-     * The transient attributes should and normally will be cleared when the

-     * node leaves, but since remote node state has a number of corner cases,

-     * clear them here as well, to be sure.

+    /* Clear node's entire state (resource history and transient attributes)

+     * other than shutdown locks. The transient attributes should and normally

+     * will be cleared when the node leaves, but since remote node state has a

+     * number of corner cases, clear them here as well, to be sure.

      */

     call_opt = crmd_cib_smart_opt();

-    controld_delete_node_state(node_name, controld_section_all, call_opt);

+    if (controld_shutdown_lock_enabled) {

+        section = controld_section_all_unlocked;

+    }

+    controld_delete_node_state(node_name, section, call_opt);

     /* Clear node's probed attribute */

     update_attrd(node_name, CRM_OP_PROBED, NULL, NULL, TRUE);

diff --git a/daemons/controld/controld_utils.h b/daemons/controld/controld_utils.h

index 8e31007..5549636 100644

--- a/daemons/controld/controld_utils.h

+++ b/daemons/controld/controld_utils.h

@@ -90,8 +90,10 @@ bool controld_action_is_recordable(const char *action);

 // Subsections of node_state

 enum controld_section_e {

     controld_section_lrm,

+    controld_section_lrm_unlocked,

     controld_section_attrs,

     controld_section_all,

+    controld_section_all_unlocked

 };

 void controld_delete_node_state(const char *uname,

-- 

1.8.3.1