repmgrd: fix upstream role check

Only take action if it's confirmed as a standby.

HISTORY

@@ -5,8 +5,9 @@
         repmgr: report unreachable nodes when running "repmgr cluster (matrix|crosscheck);
           GitHub #246 (Ian)
         repmgr: add configuration file parameter "repmgr_bindir"; GitHub #246 (Ian)
-		repmgr: fix "Missing replication slots" label in "node check"; GitHub #507 (Ian)
+        repmgr: fix "Missing replication slots" label in "node check"; GitHub #507 (Ian)
         repmgrd: fix parsing of -d/--daemonize option (Ian)
+        repmgrd: support "pausing" of repmgrd (Ian)
 
 4.1.1   2018-09-05
         logging: explicitly log the text of failed queries as ERRORs to

Makefile.in

@@ -30,13 +30,18 @@ all: \
 PG_CPPFLAGS = -std=gnu89 -I$(includedir_internal) -I$(libpq_srcdir) -Wall -Wmissing-prototypes -Wmissing-declarations $(EXTRA_CFLAGS)
 SHLIB_LINK = $(libpq)
 
-HEADERS = $(wildcard *.h)
+
 
 OBJS = \
 	repmgr.o
 
 include Makefile.global
 
+ifeq ($(vpath_build),yes)
+	HEADERS = $(wildcard *.h)
+else
+	HEADERS_built = $(wildcard *.h)
+endif
 
 $(info Building against PostgreSQL $(MAJORVERSION))
 

configfile.c

@@ -1531,6 +1531,9 @@ parse_bool(const char *s, const char *config_item, ItemList *error_list)
 {
 	PQExpBufferData errors;
 
+	if (s == NULL)
+		return true;
+
 	if (strcasecmp(s, "0") == 0)
 		return false;
 

doc/appendix-release-notes.sgml

@@ -96,10 +96,44 @@
       </para>
     </sect2>
 
+
+    <sect2>
+      <title>repmgrd enhancements</title>
+      <para>
+       <itemizedlist>
+
+          <listitem>
+            <para>
+              <application>repmgrd</application> can now be &quot;paused&quot;, i.e. instructed
+              not to take any action such as a failover, even if the prerequisites for such an
+              action are detected.
+            </para>
+            <para>
+              This removes the need to stop <application>repmgrd</application> on all nodes when
+              performing a planned operation such as a switchover.
+            </para>
+            <para>
+              For further details, see <link linkend="repmgrd-pausing">Pausing repmgrd</link>.
+            </para>
+          </listitem>
+
+       </itemizedlist>
+      </para>
+    </sect2>
+
     <sect2>
       <title>Bug fixes</title>
       <para>
         <itemizedlist>
+
+          <listitem>
+            <para>
+              &repmgr;: fix &quot;Missing replication slots&quot; label in
+              <command><link linkend="repmgr-node-check">repmgr node check</link></command>.  (GitHub #507)
+            </para>
+          </listitem>
+
+
           <listitem>
             <para>
               <application>repmgrd</application>: fix parsing of <option>-d/--daemonize</option> option.

doc/configuration-file-service-commands.sgml

@@ -17,15 +17,15 @@
     <link linkend="repmgr-node-rejoin"><command>repmgr node rejoin</command></link>.
   </para>
   <para>
-    By default, &repmgr; will use PostgreSQL's <command>pg_ctl</command> to control the PostgreSQL
+    By default, &repmgr; will use PostgreSQL's <command>pg_ctl</command> utility to control the PostgreSQL
     server. However this can lead to various problems, particularly when PostgreSQL has been
-    installed from packages, and expecially so if <application>systemd</application> is in use.
+    installed from packages, and especially so if <application>systemd</application> is in use.
   </para>
 
 
   <note>
     <para>
-      If using <application>systemd</application>, ensure you have <varname>RemoteIPC</varname> set to <literal>off</literal>.
+      If using <application>systemd</application>, ensure you have <varname>RemoveIPC</varname> set to <literal>off</literal>.
       See the <ulink url="https://wiki.postgresql.org/wiki/Systemd">systemd</ulink>
       entry in the <ulink url="https://wiki.postgresql.org/wiki/Main_Page">PostgreSQL wiki</ulink> for details.
     </para>

doc/upgrading-repmgr.sgml

@@ -7,9 +7,9 @@
  <title>Upgrading repmgr</title>
 
  <para>
-  &repmgr; is updated regularly with point releases (e.g. 4.0.1 to 4.0.2)
+  &repmgr; is updated regularly with minor releases (e.g. 4.0.1 to 4.0.2)
   containing bugfixes and other minor improvements. Any substantial new
-  functionality will be included in a feature release (e.g. 4.0.x to 4.1.x).
+  functionality will be included in a major release (e.g. 4.0 to 4.1).
  </para>
 
  <sect1 id="upgrading-repmgr-extension" xreflabel="Upgrading repmgr 4.x and later">
@@ -19,83 +19,110 @@
   </indexterm>
   <title>Upgrading repmgr 4.x and later</title>
   <para>
-    &repmgr; 4.x is implemented as a PostgreSQL extension; normally the upgrade consists
+    From version 4, &repmgr; consists of three elements:
-    of the following steps:
+     <itemizedlist spacing="compact" mark="bullet">
-    <orderedlist>
 
-      <listitem>
+       <listitem>
-        <simpara>
+         <simpara>
-          Stop <application>repmgrd</application> (if in use) on all nodes where it is running
+           the <application>repmgr</application> and <application>repmgrd</application> executables
-        </simpara>
+         </simpara>
-      </listitem>
+       </listitem>
 
-      <listitem>
+       <listitem>
-        <simpara>
+         <simpara>
-          Install the updated package (or compile the updated source)
+           the objects for the &repmgr; PostgreSQL extension (SQL files for creating/updating
-        </simpara>
+           repmgr metadata, and the extension control file)
-      </listitem>
+         </simpara>
+       </listitem>
 
-      <listitem>
+       <listitem>
-        <simpara>
+         <simpara>
-          For major releases, e.g. from <literal>4.0.x</literal> to <literal>4.1</literal>,
+           the shared library module used by <application>repmgrd</application> which
-          where the <literal>repmgr</literal> shared object library has been updated,
+           is resident in the PostgreSQL backend
-          restart PostgreSQL.
+         </simpara>
-        </simpara>
+       </listitem>
-      </listitem>
+     </itemizedlist>
-
+  </para>
-      <listitem>
+  <para>
-        <simpara>
+    With <emphasis>minor releases</emphasis>, usually changes are only made to the <application>repmgr</application>
-          For major releases, e.g. from <literal>4.0.x</literal> to <literal>4.1</literal>,
+    and <application>repmgrd</application> executables. In this case, the upgrade is quite straightforward,
-          execute <command>ALTER EXTENSION repmgr UPDATE</command>
+    and is simply a case of installing the new version, and restarting <application>repmgrd</application>
-          on the primary node in the database where the &repmgr; extension is installed.
+    (if running).
-        </simpara>
-        <simpara>
-          This will update the extension metadata and, if necessary, apply
-          changes to the &repmgr; extension objects.
-        </simpara>
-      </listitem>
-
-      <listitem>
-        <simpara>
-          Start <application>repmgrd</application> (if in use).
-        </simpara>
-      </listitem>
-
-    </orderedlist>
   </para>
 
   <para>
-    Always check the <link linkend="appendix-release-notes">release notes</link> for every
+    For <emphasis>major releases</emphasis>, the &repmgr; PostgreSQL extension will need to be updated
-    release as they may contain upgrade instructions particular to individual versions.
+    to the latest version. Additionally, if the shared library module has been updated (this is sometimes,
+    but not always the case), PostgreSQL itself will need to be restarted on each node.
   </para>
+  <important>
+    <para>
+      Always check the <link linkend="appendix-release-notes">release notes</link> for every
+      release as they may contain upgrade instructions particular to individual versions.
+    </para>
+  </important>
 
-  <para>
+  <sect2 id="upgrading-minor-version" xreflabel="Upgrading a minor version release">
-    Note that it may be necessary to restart the PostgreSQL server if the upgrade contains
-    changes to the shared object file used by <application>repmgrd</application>; check the
-    <link linkend="appendix-release-notes">release notes</link> for details.
-  </para>
-
-  <sect2 id="upgrading-replication-cluster" xreflabel="Upgrading a replication cluster">
 	<indexterm>
 	  <primary>upgrading</primary>
-	  <secondary>repmgr 4.x and later</secondary>
+	  <secondary>minor release</secondary>
 	</indexterm>
-	<title>Upgrading a replication cluster</title>
+	<title>Upgrading a minor version release</title>
+
+    <para>
+      The process for installing minor version upgrades is quite straightforward:
+
+      <itemizedlist spacing="compact" mark="bullet">
+
+        <listitem>
+          <simpara>
+            install the new &repmgr; version
+          </simpara>
+        </listitem>
+
+        <listitem>
+          <simpara>
+            restart <application>repmgrd</application> on all nodes where it is running
+          </simpara>
+        </listitem>
+
+      </itemizedlist>
+
+    </para>
+
+    <note>
+	  <para>
+        Some packaging systems (e.g. <link linkend="packages-debian-ubuntu">Debian/Ubuntu</link>
+        may restart <application>repmgrd</application> as part of the package upgrade process.
+      </para>
+    </note>
+
 	<para>
-	  The same &repmgr; &quot;major version&quot; (e.g. <literal>4.2</literal>) must be
+	  Minor version upgrades can be performed in any order on the nodes in the replication
-	  installed on all nodes in the replication cluster. While it's possible to have differing
+	  cluster.
-	  &repmgr; &quot;minor versions&quot; (e.g. <literal>4.2.1</literal>)  on different nodes,
-	  we strongly recommend updating all nodes to the latest minor version.
 	</para>
-	<note>
+
+	<para>
+	  A PostgreSQL restart is <emphasis>not</emphasis> required for minor version upgrades.
+	</para>
+
+    <note>
 	  <para>
-		Minor version upgrades can be performed in any order on the nodes in the replicaiton
+	    The same &repmgr; &quot;major version&quot; (e.g. <literal>4.2</literal>) must be
-		cluster. In general it makes sense to start on the primary.
+	    installed on all nodes in the replication cluster. While it's possible to have differing
+	    &repmgr; &quot;minor versions&quot; (e.g. <literal>4.2.1</literal>)  on different nodes,
+	    we strongly recommend updating all nodes to the latest minor version.
 	  </para>
-	  <para>
+    </note>
-		A PostgreSQL restart is <emphasis>not</emphasis> required for minor version upgrades.
+
-	  </para>
+  </sect2>
-	</note>
+
+  <sect2 id="upgrading-major-version" xreflabel="Upgrading a major version release">
+	<indexterm>
+	  <primary>upgrading</primary>
+	  <secondary>major release</secondary>
+	</indexterm>
+	<title>Upgrading a major version release</title>
 	<para>
 	  &quot;major version&quot; upgrades need to be planned more carefully, as they may include
 	  changes to the &repmgr; metadata (which need to be propagated from the primary to all
@@ -111,7 +138,14 @@
 
 		<listitem>
           <simpara>
-			Stop <application>repmgrd</application> (if in use) on all nodes where it is running
+			Stop <application>repmgrd</application> (if in use) on all nodes where it is running.
+          </simpara>
+		</listitem>
+
+		<listitem>
+          <simpara>
+			Disable the <application>repmgrd</application> service on all nodes where it is in use;
+            this is to prevent packages from prematurely restarting <application>repmgrd</application>.
           </simpara>
 		</listitem>
 
@@ -121,12 +155,21 @@
           </simpara>
 		</listitem>
 
+        <listitem>
+          <para>
+            If running a <literal>systemd</literal>-based Linux distribution, execute (as <literal>root</literal>,
+            or with appropriate <literal>sudo</literal> permissions):
+            <programlisting>
+systemctl daemon-reload</programlisting>
+          </para>
+        </listitem>
 
 		<listitem>
           <simpara>
-			If necessary, restart PostgreSQL, then <application>repmgrd</application> (if in use)
+			If the &repmgr; shared library module has been updated (check the <link linkend="appendix-release-notes">release notes</link>!),
-			on each node. The order in which this is applied to individual nodes is not critical,
+            restart PostgreSQL, then <application>repmgrd</application> (if in use)	on each node,
-			and it's also fine to restart on all nodes first before starting <application>repmgrd</application>.
+            The order in which this is applied to individual nodes is not critical,
+			and it's also fine to restart PostgreSQL on all nodes first before starting <application>repmgrd</application>.
 		  </simpara>
 		  <simpara>
 			Note that if the upgrade requires a PostgreSQL restart, <application>repmgrd</application>
@@ -138,11 +181,17 @@
           <para>
 			On the primary node, execute
 			<programlisting>
-			  ALTER EXTENSION repmgr UPDATE</programlisting>
+ALTER EXTENSION repmgr UPDATE</programlisting>
 			in the database where &repmgr; is installed.
           </para>
 		</listitem>
 
+		<listitem>
+          <simpara>
+			Reenable the <application>repmgrd</application> service on all nodes where it is in use.
+          </simpara>
+		</listitem>
+
 	  </orderedlist>
 	</para>
 	<tip>
@@ -154,6 +203,17 @@
 	</tip>
   </sect2>
 
+  <sect2 id="upgrading-check-repmgrd" xreflabel="Checking repmgrd status after an upgrade">
+	<indexterm>
+	  <primary>upgrading</primary>
+	  <secondary>checking repmgrd status</secondary>
+	</indexterm>
+	<title>Checking repmgrd status after an upgrade</title>
+	<para>
+      From &repmgr; 4.2, once the upgrade is complete, execute the <command><link linkend="repmgr-daemon-status">repmgr daemon status</link></command>
+      command (on any node) to show an overview of the status of <application>repmgrd</application> on all nodes.
+    </para>
+  </sect2>
  </sect1>
 
  <sect1 id="upgrading-and-pg-upgrade" xreflabel="pg_upgrade and repmgr">

doc/version.sgml

@@ -1 +1 @@
-<!ENTITY repmgrversion "4.2dev">
+<!ENTITY repmgrversion "4.2">

repmgr-action-standby.c

@@ -3718,9 +3718,18 @@ do_standby_switchover(void)
 				i++;
 			}
 		}
-
+		else
+		{
+			/* close all connections - we'll reestablish later */
+			for (cell = all_nodes.head; cell; cell = cell->next)
+			{
+				PQfinish(cell->node_info->conn);
+				cell->node_info->conn = NULL;
+			}
+		}
 	}
 
+
 	/*
 	 * Sanity checks completed - prepare for the switchover
 	 */
@@ -3801,7 +3810,7 @@ do_standby_switchover(void)
 				 shutdown_command);
 
 		clear_node_info_list(&sibling_nodes);
-		clear_node_info_list(&all_nodes);
+
 		key_value_list_free(&remote_config_files);
 
 		return;
@@ -4087,9 +4096,25 @@ do_standby_switchover(void)
 
 			if (sibling_node_record.type == WITNESS)
 			{
+				/* TODO: create "repmgr witness resync" or similar */
 				appendPQExpBuffer(&remote_command_str,
 								  "witness register -d \\'%s\\' --force 2>/dev/null && echo \"1\" || echo \"0\"",
 								  local_node_record.conninfo);
+
+				/*
+				 * Notify the witness repmgrd about the new primary, as at this point it will be assuming
+				 * a failover situation is in place. It will detect the new primary at some point, this
+				 * just speeds up the process.
+				 *
+				 * In the unlikely event repmgrd is not running or not in use, this will have no effect.
+				 */
+				cell->node_info->conn = establish_db_connection_quiet(cell->node_info->conninfo);
+
+				if (PQstatus(cell->node_info->conn) == CONNECTION_OK)
+				{
+					notify_follow_primary(cell->node_info->conn, local_node_record.node_id);
+				}
+				PQfinish(cell->node_info->conn);
 			}
 			else
 			{
@@ -4144,8 +4169,6 @@ do_standby_switchover(void)
 
 	clear_node_info_list(&sibling_nodes);
 
-
-
 	PQfinish(local_conn);
 
 	/*

repmgr_version.h.in

@@ -1,2 +1,2 @@
 #define REPMGR_VERSION_DATE ""
-#define REPMGR_VERSION "4.2dev"
+#define REPMGR_VERSION "4.2"

repmgrd-physical.c

@@ -71,6 +71,8 @@ static void notify_followers(NodeInfoList *standby_nodes, int follow_node_id);
 
 static void check_connection(t_node_info *node_info, PGconn **conn);
 
+static bool check_primary_status(int degraded_monitoring_elapsed);
+
 static bool wait_primary_notification(int *new_primary_id);
 static FailoverState follow_new_primary(int new_primary_id);
 static FailoverState witness_follow_new_primary(int new_primary_id);
@@ -341,6 +343,13 @@ monitor_streaming_primary(void)
 						repmgrd_set_pid(local_conn, getpid(), pid_file);
 					}
 
+					/*
+					 * check that the local node is still primary, otherwise switch
+					 * to standby monitoring
+					 */
+					if (check_primary_status(-1) == false)
+						return;
+
 					goto loop;
 				}
 
@@ -393,143 +402,10 @@ monitor_streaming_primary(void)
 				{
 					local_node_info.node_status = NODE_STATUS_UP;
 
-					/* check to see if the node has been restored as a standby */
+					if (check_primary_status(degraded_monitoring_elapsed) == false)
-					if (get_recovery_type(local_conn) == RECTYPE_STANDBY)
+						return;
-					{
-						PGconn *new_primary_conn;
 
-						initPQExpBuffer(&event_details);
+					goto loop;
-
-						appendPQExpBuffer(&event_details,
-										  _("reconnected to node after %i seconds, node is now a standby, switching to standby monitoring"),
-										  degraded_monitoring_elapsed);
-						log_notice("%s", event_details.data);
-						termPQExpBuffer(&event_details);
-
-						primary_node_id = UNKNOWN_NODE_ID;
-
-						new_primary_conn = get_primary_connection_quiet(local_conn, &primary_node_id, NULL);
-
-						if (PQstatus(new_primary_conn) != CONNECTION_OK)
-						{
-							close_connection(&new_primary_conn);
-							log_warning(_("unable to connect to new primary node %i"), primary_node_id);
-						}
-						else
-						{
-							RecordStatus record_status;
-
-							log_debug("primary node id is now %i", primary_node_id);
-
-							record_status = get_node_record(new_primary_conn, config_file_options.node_id, &local_node_info);
-
-							if (record_status == RECORD_FOUND)
-							{
-								bool resume_monitoring = true;
-
-								log_debug("node %i is registered with type = %s",
-										  config_file_options.node_id,
-										  get_node_type_string(local_node_info.type));
-
-								/*
-								 * node has recovered but metadata not updated - we can do that ourselves,
-								 */
-								if (local_node_info.type == PRIMARY)
-								{
-									log_notice(_("node \"%s\" (ID: %i) still registered as primary, setting to standby"),
-											   config_file_options.node_name,
-											   config_file_options.node_id);
-
-									if (update_node_record_set_active_standby(new_primary_conn, config_file_options.node_id) == false)
-									{
-										resume_monitoring = false;
-									}
-									else
-									{
-										record_status = get_node_record(new_primary_conn, config_file_options.node_id, &local_node_info);
-
-										if (record_status != RECORD_FOUND)
-										{
-											resume_monitoring = false;
-										}
-									}
-								}
-
-								if (resume_monitoring == true)
-								{
-									monitoring_state = MS_NORMAL;
-									log_notice(_("former primary has been restored as standby after %i seconds, updating node record and resuming monitoring"),
-											   degraded_monitoring_elapsed);
-
-									initPQExpBuffer(&event_details);
-
-									appendPQExpBuffer(&event_details,
-													  _("node restored as standby after %i seconds, monitoring connection to upstream node %i"),
-													  degraded_monitoring_elapsed,
-													  local_node_info.upstream_node_id);
-
-									create_event_notification(new_primary_conn,
-															  &config_file_options,
-															  config_file_options.node_id,
-															  "repmgrd_standby_reconnect",
-															  true,
-															  event_details.data);
-
-
-									termPQExpBuffer(&event_details);
-
-									close_connection(&new_primary_conn);
-
-									/* restart monitoring as standby */
-									return;
-								}
-							}
-							else if (record_status == RECORD_NOT_FOUND)
-							{
-								PQExpBufferData event_details;
-								initPQExpBuffer(&event_details);
-
-								appendPQExpBuffer(&event_details,
-												  _("no metadata record found for this node on current primary %i"),
-												  primary_node_id);
-
-								log_error("%s", event_details.data);
-								log_hint(_("check that 'repmgr (primary|standby) register' was executed for this node"));
-
-								close_connection(&new_primary_conn);
-
-								create_event_notification(NULL,
-														  &config_file_options,
-														  config_file_options.node_id,
-														  "repmgrd_shutdown",
-														  false,
-														  event_details.data);
-								termPQExpBuffer(&event_details);
-
-								terminate(ERR_BAD_CONFIG);
-							}
-						}
-					}
-					else
-					{
-						monitoring_state = MS_NORMAL;
-
-						initPQExpBuffer(&event_details);
-						appendPQExpBuffer(&event_details,
-										  _("reconnected to primary node after %i seconds, resuming monitoring"),
-										  degraded_monitoring_elapsed);
-
-						create_event_notification(local_conn,
-												  &config_file_options,
-												  config_file_options.node_id,
-												  "repmgrd_local_reconnect",
-												  true,
-												  event_details.data);
-
-						log_notice("%s", event_details.data);
-						termPQExpBuffer(&event_details);
-						goto loop;
-					}
 				}
 			}
 
@@ -541,6 +417,11 @@ monitor_streaming_primary(void)
 			 */
 		}
 loop:
+
+		/* check node is still primary, if not restart monitoring */
+		if (check_primary_status(-1) == false)
+			return;
+
 		/* emit "still alive" log message at regular intervals, if requested */
 		if (config_file_options.log_status_interval > 0)
 		{
@@ -575,6 +456,181 @@ loop:
 }
 
 
+/*
+ * If monitoring a primary, it's possible that after an outage of the local node
+ * (due to e.g. a switchover), the node has come back as a standby. We therefore
+ * need to verify its status and if everything looks OK, restart monitoring in
+ * standby mode.
+ */
+bool
+check_primary_status(int degraded_monitoring_elapsed)
+{
+	PQExpBufferData event_details;
+
+	/* check to see if the node has been restored as a standby */
+	if (get_recovery_type(local_conn) == RECTYPE_STANDBY)
+	{
+		PGconn *new_primary_conn;
+
+		initPQExpBuffer(&event_details);
+
+		if (degraded_monitoring_elapsed > 0)
+		{
+			appendPQExpBuffer(&event_details,
+							  _("reconnected to node after %i seconds, node is now a standby, switching to standby monitoring"),
+							  degraded_monitoring_elapsed);
+		}
+		else
+		{
+			appendPQExpBufferStr(&event_details,
+								 _("node is now a standby, switching to standby monitoring"));
+		}
+
+		log_notice("%s", event_details.data);
+		termPQExpBuffer(&event_details);
+
+		primary_node_id = UNKNOWN_NODE_ID;
+
+		new_primary_conn = get_primary_connection_quiet(local_conn, &primary_node_id, NULL);
+
+		if (PQstatus(new_primary_conn) != CONNECTION_OK)
+		{
+			close_connection(&new_primary_conn);
+			log_warning(_("unable to connect to new primary node %i"), primary_node_id);
+		}
+		else
+		{
+			RecordStatus record_status;
+
+			log_debug("primary node id is now %i", primary_node_id);
+
+			record_status = get_node_record(new_primary_conn, config_file_options.node_id, &local_node_info);
+
+			if (record_status == RECORD_FOUND)
+			{
+				bool resume_monitoring = true;
+
+				log_debug("node %i is registered with type = %s",
+						  config_file_options.node_id,
+						  get_node_type_string(local_node_info.type));
+
+				/*
+				 * node has recovered but metadata not updated - we can do that ourselves,
+				 */
+				if (local_node_info.type == PRIMARY)
+				{
+					log_notice(_("node \"%s\" (ID: %i) still registered as primary, setting to standby"),
+											   config_file_options.node_name,
+											   config_file_options.node_id);
+
+					if (update_node_record_set_active_standby(new_primary_conn, config_file_options.node_id) == false)
+					{
+						resume_monitoring = false;
+					}
+					else
+					{
+						record_status = get_node_record(new_primary_conn, config_file_options.node_id, &local_node_info);
+
+						if (record_status != RECORD_FOUND)
+						{
+							resume_monitoring = false;
+						}
+					}
+				}
+
+				if (resume_monitoring == true)
+				{
+					initPQExpBuffer(&event_details);
+
+					if (degraded_monitoring_elapsed > 0)
+					{
+						monitoring_state = MS_NORMAL;
+
+						log_notice(_("former primary has been restored as standby after %i seconds, updating node record and resuming monitoring"),
+								   degraded_monitoring_elapsed);
+
+
+						appendPQExpBuffer(&event_details,
+										  _("node restored as standby after %i seconds, monitoring connection to upstream node %i"),
+										  degraded_monitoring_elapsed,
+										  local_node_info.upstream_node_id);
+					}
+					else
+					{
+						appendPQExpBuffer(&event_details,
+										  _("node has become a standby, monitoring connection to upstream node %i"),
+										  local_node_info.upstream_node_id);
+					}
+
+					create_event_notification(new_primary_conn,
+											  &config_file_options,
+											  config_file_options.node_id,
+											  "repmgrd_standby_reconnect",
+											  true,
+											  event_details.data);
+
+
+					termPQExpBuffer(&event_details);
+
+					close_connection(&new_primary_conn);
+
+					/* restart monitoring as standby */
+					return false;
+				}
+			}
+			else if (record_status == RECORD_NOT_FOUND)
+			{
+				initPQExpBuffer(&event_details);
+
+				appendPQExpBuffer(&event_details,
+								  _("no metadata record found for this node on current primary %i"),
+								  primary_node_id);
+
+				log_error("%s", event_details.data);
+				log_hint(_("check that 'repmgr (primary|standby) register' was executed for this node"));
+
+				close_connection(&new_primary_conn);
+
+				create_event_notification(NULL,
+										  &config_file_options,
+										  config_file_options.node_id,
+										  "repmgrd_shutdown",
+										  false,
+										  event_details.data);
+				termPQExpBuffer(&event_details);
+
+				terminate(ERR_BAD_CONFIG);
+			}
+		}
+	}
+	else
+	{
+		if (degraded_monitoring_elapsed > 0)
+		{
+			monitoring_state = MS_NORMAL;
+
+			initPQExpBuffer(&event_details);
+			appendPQExpBuffer(&event_details,
+							  _("reconnected to primary node after %i seconds, resuming monitoring"),
+							  degraded_monitoring_elapsed);
+
+			create_event_notification(local_conn,
+									  &config_file_options,
+									  config_file_options.node_id,
+									  "repmgrd_local_reconnect",
+									  true,
+									  event_details.data);
+
+			log_notice("%s", event_details.data);
+			termPQExpBuffer(&event_details);
+		}
+	}
+
+	return true;
+}
+
+
+
 void
 monitor_streaming_standby(void)
 {
@@ -595,7 +651,7 @@ monitor_streaming_standby(void)
 	 */
 	if (local_node_info.upstream_node_id == UNKNOWN_NODE_ID)
 	{
-		local_node_info.upstream_node_id = get_primary_node_id(local_conn);
+		upstream_conn = get_primary_connection(local_conn, &local_node_info.upstream_node_id, NULL);
 
 		/*
 		 * Terminate if there doesn't appear to be an active cluster primary.
@@ -608,34 +664,40 @@ monitor_streaming_standby(void)
 			log_error(_("unable to determine an active primary for this cluster, terminating"));
 			terminate(ERR_BAD_CONFIG);
 		}
+
+		(void) get_node_record(upstream_conn, local_node_info.upstream_node_id, &upstream_node_info);
+
 	}
-
+	else
-	record_status = get_node_record(local_conn, local_node_info.upstream_node_id, &upstream_node_info);
-
-	/*
-	 * Terminate if we can't find the record for the node we're supposed to
-	 * monitor. This is a "fix-the-config" situation, not a lot else we can
-	 * do.
-	 */
-	if (record_status == RECORD_NOT_FOUND)
 	{
-		log_error(_("no record found for upstream node (ID: %i), terminating"),
+		record_status = get_node_record(local_conn, local_node_info.upstream_node_id, &upstream_node_info);
-				  local_node_info.upstream_node_id);
-		log_hint(_("ensure the upstream node is registered correctly"));
 
-		terminate(ERR_DB_CONN);
+		/*
-	}
+		 * Terminate if we can't find the record for the node we're supposed to
-	else if (record_status == RECORD_ERROR)
+		 * monitor. This is a "fix-the-config" situation, not a lot else we can
-	{
+		 * do.
-		log_error(_("unable to retrieve record for upstream node (ID: %i), terminating"),
+		 */
-				  local_node_info.upstream_node_id);
+		if (record_status == RECORD_NOT_FOUND)
+		{
+			log_error(_("no record found for upstream node (ID: %i), terminating"),
+					  local_node_info.upstream_node_id);
+			log_hint(_("ensure the upstream node is registered correctly"));
 
-		terminate(ERR_DB_CONN);
+			terminate(ERR_DB_CONN);
+		}
+		else if (record_status == RECORD_ERROR)
+		{
+			log_error(_("unable to retrieve record for upstream node (ID: %i), terminating"),
+					  local_node_info.upstream_node_id);
+
+			terminate(ERR_DB_CONN);
+		}
+
+		log_debug("connecting to upstream node %i: \"%s\"", upstream_node_info.node_id, upstream_node_info.conninfo);
+
+		upstream_conn = establish_db_connection(upstream_node_info.conninfo, false);
 	}
 
-	log_debug("connecting to upstream node %i: \"%s\"", upstream_node_info.node_id, upstream_node_info.conninfo);
-
-	upstream_conn = establish_db_connection(upstream_node_info.conninfo, false);
 
 	/*
 	 * Upstream node must be running at repmgrd startup.
@@ -652,6 +714,15 @@ monitor_streaming_standby(void)
 		terminate(ERR_DB_CONN);
 	}
 
+	record_status = get_node_record(upstream_conn, local_node_info.node_id, &local_node_info);
+
+	if (upstream_node_info.node_id == local_node_info.node_id)
+	{
+		PQfinish(upstream_conn);
+		upstream_conn = NULL;
+		return;
+	}
+
 	/*
 	 * refresh upstream node record from upstream node, so it's as up-to-date
 	 * as possible
@@ -682,6 +753,23 @@ monitor_streaming_standby(void)
 		primary_conn = upstream_conn;
 	}
 
+	/*
+	 * It's possible monitoring has been restarted after some outage which
+	 * resulted in the local node being marked as inactive; if so mark it
+	 * as active again.
+	 */
+	if (local_node_info.active == false)
+	{
+		if (update_node_record_set_active(primary_conn, local_node_info.node_id, true) == true)
+		{
+			PQExpBufferData event_details;
+
+			initPQExpBuffer(&event_details);
+
+			local_node_info.active = true;
+		}
+	}
+
 	primary_node_id = get_primary_node_id(primary_conn);
 
 	/* Log startup event */
@@ -766,6 +854,7 @@ monitor_streaming_standby(void)
 				if (PQstatus(local_conn) != CONNECTION_OK)
 				{
 					check_connection(&local_node_info, &local_conn);
+					log_debug("YYY here");
 				}
 
 				try_reconnect(&upstream_conn, &upstream_node_info);
@@ -778,6 +867,43 @@ monitor_streaming_standby(void)
 					if (upstream_node_info.type == PRIMARY)
 					{
 						primary_conn = upstream_conn;
+
+						if (get_recovery_type(primary_conn) == RECTYPE_STANDBY)
+						{
+							ExecStatusType ping_result;
+
+							log_notice(_("current upstream node \"%s\" (node ID: %i) is not primary, restarting monitoring"),
+									   upstream_node_info.node_name, upstream_node_info.node_id);
+							PQfinish(upstream_conn);
+							upstream_conn = NULL;
+							termPQExpBuffer(&event_details);
+							local_node_info.upstream_node_id = UNKNOWN_NODE_ID;
+
+							/* check local connection */
+							ping_result = connection_ping(local_conn);
+
+							if (ping_result != PGRES_TUPLES_OK)
+							{
+								int i;
+
+								PQfinish(local_conn);
+
+								for (i = 0; i < config_file_options.repmgrd_standby_startup_timeout; i++)
+								{
+									local_conn = establish_db_connection(local_node_info.conninfo, false);
+
+									if (PQstatus(local_conn) == CONNECTION_OK)
+										break;
+
+									log_debug("sleeping 1 second; %i of %i attempts to reconnect to local node",
+											  i + 1,
+											  config_file_options.repmgrd_standby_startup_timeout);
+									sleep(1);
+								}
+							}
+
+							return;
+						}
 					}
 
 					initPQExpBuffer(&event_details);
@@ -1140,6 +1266,7 @@ loop:
 
 		check_connection(&local_node_info, &local_conn);
 
+
 		if (PQstatus(local_conn) != CONNECTION_OK)
 		{
 			if (local_node_info.active == true)
@@ -1180,11 +1307,37 @@ loop:
 		}
 		else
 		{
+			int stored_local_node_id = repmgrd_get_local_node_id(local_conn);
+
+			/*
+			 * If the local node was restarted, we'll need to reinitialise values
+			 * stored in shared memory.
+			 */
+
+			if (stored_local_node_id == UNKNOWN_NODE_ID)
+			{
+				repmgrd_set_local_node_id(local_conn, config_file_options.node_id);
+				repmgrd_set_pid(local_conn, getpid(), pid_file);
+			}
+
+			if (PQstatus(primary_conn) == CONNECTION_OK)
+			{
+				if (get_recovery_type(primary_conn) == RECTYPE_STANDBY)
+				{
+					log_notice(_("current upstream node \"%s\" (node ID: %i) is not primary, restarting monitoring"),
+							   upstream_node_info.node_name, upstream_node_info.node_id);
+					PQfinish(primary_conn);
+					primary_conn = NULL;
+					termPQExpBuffer(&event_details);
+
+					local_node_info.upstream_node_id = UNKNOWN_NODE_ID;
+					return;
+				}
+			}
+
 			/* we've reconnected to the local node after an outage */
 			if (local_node_info.active == false)
 			{
-				int stored_local_node_id = UNKNOWN_NODE_ID;
-
 				if (PQstatus(primary_conn) == CONNECTION_OK)
 				{
 					if (update_node_record_set_active(primary_conn, local_node_info.node_id, true) == true)
@@ -1194,7 +1347,6 @@ loop:
 						initPQExpBuffer(&event_details);
 
 						local_node_info.active = true;
-
 						appendPQExpBuffer(&event_details,
 										  _("reconnected to local node \"%s\" (ID: %i), marking active"),
 										  local_node_info.node_name,
@@ -1212,18 +1364,6 @@ loop:
 						termPQExpBuffer(&event_details);
 					}
 				}
-
-				/*
-				 * If the local node was restarted, we'll need to reinitialise values
-				 * stored in shared memory.
-				 */
-
-				stored_local_node_id = repmgrd_get_local_node_id(local_conn);
-				if (stored_local_node_id == UNKNOWN_NODE_ID)
-				{
-					repmgrd_set_local_node_id(local_conn, config_file_options.node_id);
-					repmgrd_set_pid(local_conn, getpid(), pid_file);
-				}
 			}
 		}
 
@@ -1299,12 +1439,13 @@ monitor_streaming_witness(void)
 		log_warning(_("unable to retrieve node record from primary"));
 	}
 
-
-	/* Log startup event */
-	if (startup_event_logged == false)
 	{
 		PQExpBufferData event_details;
 
+		char *event_type = startup_event_logged == false
+			? "repmgrd_start"
+			: "repmgrd_upstream_reconnect";
+
 		initPQExpBuffer(&event_details);
 
 		appendPQExpBuffer(&event_details,
@@ -1312,16 +1453,17 @@ monitor_streaming_witness(void)
 						  upstream_node_info.node_name,
 						  upstream_node_info.node_id);
 
+		log_info("%s", event_details.data);
+
 		create_event_notification(primary_conn,
 								  &config_file_options,
 								  config_file_options.node_id,
-								  "repmgrd_start",
+								  event_type,
 								  true,
 								  event_details.data);
 
-		startup_event_logged = true;
+		if (startup_event_logged == false)
-
+			startup_event_logged = true;
-		log_info("%s", event_details.data);
 
 		termPQExpBuffer(&event_details);
 	}
@@ -1371,6 +1513,17 @@ monitor_streaming_witness(void)
 									  upstream_node_unreachable_elapsed);
 					log_notice("%s", event_details.data);
 
+					/* check upstream is still primary */
+					if (get_recovery_type(primary_conn) != RECTYPE_PRIMARY)
+					{
+						log_notice(_("current upstream node \"%s\" (node ID: %i) is not primary, restarting monitoring"),
+								   upstream_node_info.node_name, upstream_node_info.node_id);
+						PQfinish(primary_conn);
+						primary_conn = NULL;
+						termPQExpBuffer(&event_details);
+						return;
+					}
+
 					create_event_notification(primary_conn,
 											  &config_file_options,
 											  config_file_options.node_id,
@@ -1428,14 +1581,25 @@ monitor_streaming_witness(void)
 									  upstream_node_info.node_id,
 									  degraded_monitoring_elapsed);
 
+					log_notice("%s", event_details.data);
+
+					/* check upstream is still primary */
+					if (get_recovery_type(primary_conn) != RECTYPE_PRIMARY)
+					{
+						log_notice(_("current upstream node \"%s\" (node ID: %i) is not primary, restarting monitoring"),
+								   upstream_node_info.node_name, upstream_node_info.node_id);
+						PQfinish(primary_conn);
+						primary_conn = NULL;
+						termPQExpBuffer(&event_details);
+						return;
+					}
+
 					create_event_notification(primary_conn,
 											  &config_file_options,
 											  config_file_options.node_id,
 											  "repmgrd_upstream_reconnect",
 											  true,
 											  event_details.data);
-
-					log_notice("%s", event_details.data);
 					termPQExpBuffer(&event_details);
 
 					goto loop;
@@ -1467,6 +1631,12 @@ monitor_streaming_witness(void)
 							continue;
 						}
 
+						/* skip node if configured as a witness node - we can't possibly "follow" that */
+						if (cell->node_info->type == WITNESS)
+						{
+							continue;
+						}
+
 						cell->node_info->conn = establish_db_connection(cell->node_info->conninfo, false);
 
 						if (PQstatus(cell->node_info->conn) != CONNECTION_OK)
@@ -1594,14 +1764,28 @@ loop:
 		}
 
 
-		/* refresh repmgr.nodes after "witness_sync_interval" seconds */
+		/*
+		 * Refresh repmgr.nodes after "witness_sync_interval" seconds, and check if primary
+		 * has changed
+		 */
 
 		{
 			int witness_sync_interval_elapsed = calculate_elapsed(witness_sync_interval_start);
+
 			if (witness_sync_interval_elapsed >= config_file_options.witness_sync_interval)
 			{
+				if (get_recovery_type(primary_conn) != RECTYPE_PRIMARY)
+				{
+					log_notice(_("current upstream node \"%s\" (node ID: %i) is not primary, restarting monitoring"),
+							   upstream_node_info.node_name, upstream_node_info.node_id);
+					PQfinish(primary_conn);
+					primary_conn = NULL;
+					return;
+				}
+
 				log_debug("synchronising witness node records");
 				witness_copy_node_records(primary_conn, local_conn);
+
 				INSTR_TIME_SET_CURRENT(witness_sync_interval_start);
 			}
 		}
@@ -3092,6 +3276,8 @@ check_connection(t_node_info *node_info, PGconn **conn)
 	if (is_server_available(node_info->conninfo) == false)
 	{
 		log_warning(_("connection to node %i lost"), node_info->node_id);
+		PQfinish(*conn);
+		*conn = NULL;
 	}
 
 	if (PQstatus(*conn) != CONNECTION_OK)

repmgrd.c

@@ -175,7 +175,6 @@ main(int argc, char **argv)
 
 				/* daemon options */
 
-
 			case 'd':
 				daemonize = true;
 				break;
@@ -184,7 +183,6 @@ main(int argc, char **argv)
 				daemonize = parse_bool(optarg, "-d/--daemonize", &cli_errors);
 				break;
 
-
 			case 'p':
 				strncpy(pid_file, optarg, MAXPGPATH);
 				break;