Add the reported timeline id to the monitor events table. (#753)

hapostgres · Jul 5, 2021 · cf91b7a · cf91b7a
1 parent ca79fba
commit cf91b7a
Show file tree

Hide file tree

Showing 5 changed files with 157 additions and 11 deletions.
diff --git a/src/bin/pg_autoctl/monitor.c b/src/bin/pg_autoctl/monitor.c
@@ -2232,7 +2232,7 @@ monitor_print_last_events(Monitor *monitor, char *formation, int group, int coun
 								   paramCount, paramTypes, paramValues,
 								   &context, &printLastEvents))
 	{
-		log_error("Failed to retrieve current state from the monitor");
+		log_error("Failed to retrieve last events from the monitor");
 		return false;
 	}
 

diff --git a/src/monitor/notifications.c b/src/monitor/notifications.c
@@ -134,6 +134,7 @@ InsertEvent(AutoFailoverNode *node, char *description)
 		replicationStateTypeOid, /* reportedstate */
 		replicationStateTypeOid, /* goalstate */
 		TEXTOID, /* pg_stat_replication.sync_state */
+		INT4OID, /* timeline_id */
 		LSNOID,  /* reportedLSN */
 		INT4OID, /* candidate_priority */
 		BOOLOID, /* replication_quorum */
@@ -150,6 +151,7 @@ InsertEvent(AutoFailoverNode *node, char *description)
 		ObjectIdGetDatum(reportedStateOid), /* reportedstate */
 		ObjectIdGetDatum(goalStateOid),     /* goalstate */
 		CStringGetTextDatum(SyncStateToString(node->pgsrSyncState)), /* sync_state */
+		Int32GetDatum(node->reportedTLI),         /* reportedTLI */
 		LSNGetDatum(node->reportedLSN),           /* reportedLSN */
 		Int32GetDatum(node->candidatePriority),   /* candidate_priority */
 		BoolGetDatum(node->replicationQuorum),    /* replication_quorum */
@@ -162,9 +164,9 @@ InsertEvent(AutoFailoverNode *node, char *description)
 	const char *insertQuery =
 		"INSERT INTO " AUTO_FAILOVER_EVENT_TABLE
 		"(formationid, nodeid, groupid, nodename, nodehost, nodeport,"
-		" reportedstate, goalstate, reportedrepstate, reportedlsn,"
+		" reportedstate, goalstate, reportedrepstate, reportedtli, reportedlsn,"
 		" candidatepriority, replicationquorum, description) "
-		"VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13) "
+		"VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14) "
 		"RETURNING eventid";
 
 	SPI_connect();

diff --git a/src/monitor/pgautofailover--1.5--1.6.sql b/src/monitor/pgautofailover--1.5--1.6.sql
@@ -18,6 +18,10 @@ DROP FUNCTION pgautofailover.get_other_nodes(int);
 DROP FUNCTION pgautofailover.get_other_nodes
               (integer,pgautofailover.replication_state);
 
+DROP FUNCTION pgautofailover.last_events(int);
+DROP FUNCTION pgautofailover.last_events(text,int);
+DROP FUNCTION pgautofailover.last_events(text,int,int);
+
 DROP FUNCTION pgautofailover.current_state(text);
 DROP FUNCTION pgautofailover.current_state(text,int);
 
@@ -158,6 +162,51 @@ INSERT INTO pgautofailover.node
         candidatepriority, replicationquorum, nodecluster
    FROM pgautofailover.node_upgrade_old;
 
+
+ALTER TABLE pgautofailover.event
+	RENAME TO event_upgrade_old;
+
+CREATE TABLE pgautofailover.event
+ (
+    eventid           bigint not null DEFAULT nextval('pgautofailover.event_eventid_seq'::regclass),
+    eventtime         timestamptz not null default now(),
+    formationid       text not null,
+    nodeid            bigint not null,
+    groupid           int not null,
+    nodename          text not null,
+    nodehost          text not null,
+    nodeport          integer not null,
+    reportedstate     pgautofailover.replication_state not null,
+    goalstate         pgautofailover.replication_state not null,
+    reportedrepstate  text,
+    reportedtli       int not null default 1 check (reportedtli > 0),
+    reportedlsn       pg_lsn not null default '0/0',
+    candidatepriority int,
+    replicationquorum bool,
+    description       text,
+
+    PRIMARY KEY (eventid)
+ );
+
+ALTER SEQUENCE pgautofailover.event_eventid_seq
+      OWNED BY pgautofailover.event.eventid;
+
+INSERT INTO pgautofailover.event
+ (
+  eventid, eventtime, formationid, nodeid, groupid,
+  nodename, nodehost, nodeport,
+  reportedstate, goalstate, reportedrepstate,
+  reportedtli, reportedlsn, candidatepriority, replicationquorum,
+  description
+ )
+ SELECT eventid, eventtime, formationid, nodeid, groupid,
+        nodename, nodehost, nodeport,
+        reportedstate, goalstate, reportedrepstate,
+        1 as reportedtli, reportedlsn, candidatepriority, replicationquorum,
+        description
+   FROM pgautofailover.event_upgrade_old as event;
+
+DROP TABLE pgautofailover.event_upgrade_old;
 DROP TABLE pgautofailover.node_upgrade_old;
 DROP TYPE pgautofailover.old_replication_state;
 
@@ -405,6 +454,91 @@ CREATE TRIGGER disable_secondary_check
 	EXECUTE PROCEDURE pgautofailover.update_secondary_check();
 
 
+CREATE FUNCTION pgautofailover.last_events
+ (
+  count int default 10
+ )
+RETURNS SETOF pgautofailover.event LANGUAGE SQL STRICT
+AS $$
+with last_events as
+(
+  select eventid, eventtime, formationid,
+         nodeid, groupid, nodename, nodehost, nodeport,
+         reportedstate, goalstate,
+         reportedrepstate, reportedtli, reportedlsn,
+         candidatepriority, replicationquorum, description
+    from pgautofailover.event
+order by eventid desc
+   limit count
+)
+select * from last_events order by eventtime, eventid;
+$$;
+
+comment on function pgautofailover.last_events(int)
+        is 'retrieve last COUNT events';
+
+grant execute on function pgautofailover.last_events(int)
+   to autoctl_node;
+
+CREATE FUNCTION pgautofailover.last_events
+ (
+  formation_id text default 'default',
+  count        int  default 10
+ )
+RETURNS SETOF pgautofailover.event LANGUAGE SQL STRICT
+AS $$
+with last_events as
+(
+    select eventid, eventtime, formationid,
+           nodeid, groupid, nodename, nodehost, nodeport,
+           reportedstate, goalstate,
+           reportedrepstate, reportedtli, reportedlsn,
+           candidatepriority, replicationquorum, description
+      from pgautofailover.event
+     where formationid = formation_id
+  order by eventid desc
+     limit count
+)
+select * from last_events order by eventtime, eventid;
+$$;
+
+comment on function pgautofailover.last_events(text,int)
+        is 'retrieve last COUNT events for given formation';
+
+grant execute on function pgautofailover.last_events(text,int)
+   to autoctl_node;
+
+CREATE FUNCTION pgautofailover.last_events
+ (
+  formation_id text,
+  group_id     int,
+  count        int default 10
+ )
+RETURNS SETOF pgautofailover.event LANGUAGE SQL STRICT
+AS $$
+with last_events as
+(
+    select eventid, eventtime, formationid,
+           nodeid, groupid, nodename, nodehost, nodeport,
+           reportedstate, goalstate,
+           reportedrepstate, reportedtli, reportedlsn,
+           candidatepriority, replicationquorum, description
+      from pgautofailover.event
+     where formationid = formation_id
+       and groupid = group_id
+  order by eventid desc
+     limit count
+)
+select * from last_events order by eventtime, eventid;
+$$;
+
+comment on function pgautofailover.last_events(text,int,int)
+        is 'retrieve last COUNT events for given formation and group';
+
+grant execute on function pgautofailover.last_events(text,int,int)
+   to autoctl_node;
+
+
 CREATE FUNCTION pgautofailover.current_state
  (
     IN formation_id         text default 'default',

diff --git a/src/monitor/pgautofailover.sql b/src/monitor/pgautofailover.sql
@@ -176,6 +176,7 @@ CREATE TABLE pgautofailover.event
     reportedstate     pgautofailover.replication_state not null,
     goalstate         pgautofailover.replication_state not null,
     reportedrepstate  text,
+    reportedtli       int not null default 1 check (reportedtli > 0),
     reportedlsn       pg_lsn not null default '0/0',
     candidatepriority int,
     replicationquorum bool,
@@ -510,7 +511,7 @@ with last_events as
   select eventid, eventtime, formationid,
          nodeid, groupid, nodename, nodehost, nodeport,
          reportedstate, goalstate,
-         reportedrepstate, reportedlsn,
+         reportedrepstate, reportedtli, reportedlsn,
          candidatepriority, replicationquorum, description
     from pgautofailover.event
 order by eventid desc
@@ -522,6 +523,9 @@ $$;
 comment on function pgautofailover.last_events(int)
         is 'retrieve last COUNT events';
 
+grant execute on function pgautofailover.last_events(int)
+   to autoctl_node;
+
 CREATE FUNCTION pgautofailover.last_events
  (
   formation_id text default 'default',
@@ -534,7 +538,7 @@ with last_events as
     select eventid, eventtime, formationid,
            nodeid, groupid, nodename, nodehost, nodeport,
            reportedstate, goalstate,
-           reportedrepstate, reportedlsn,
+           reportedrepstate, reportedtli, reportedlsn,
            candidatepriority, replicationquorum, description
       from pgautofailover.event
      where formationid = formation_id
@@ -547,6 +551,9 @@ $$;
 comment on function pgautofailover.last_events(text,int)
         is 'retrieve last COUNT events for given formation';
 
+grant execute on function pgautofailover.last_events(text,int)
+   to autoctl_node;
+
 CREATE FUNCTION pgautofailover.last_events
  (
   formation_id text,
@@ -560,7 +567,7 @@ with last_events as
     select eventid, eventtime, formationid,
            nodeid, groupid, nodename, nodehost, nodeport,
            reportedstate, goalstate,
-           reportedrepstate, reportedlsn,
+           reportedrepstate, reportedtli, reportedlsn,
            candidatepriority, replicationquorum, description
       from pgautofailover.event
      where formationid = formation_id
@@ -574,6 +581,9 @@ $$;
 comment on function pgautofailover.last_events(text,int,int)
         is 'retrieve last COUNT events for given formation and group';
 
+grant execute on function pgautofailover.last_events(text,int,int)
+   to autoctl_node;
+
 CREATE FUNCTION pgautofailover.current_state
  (
     IN formation_id         text default 'default',

diff --git a/tests/pgautofailover_utils.py b/tests/pgautofailover_utils.py
@@ -669,19 +669,19 @@ def get_events_str(self):
         if events:
             return "\n".join(
                 [
-                    "%32s %8s %17s/%-17s %7s %10s %s"
+                    "%32s %8s %17s/%-17s %10s %10s %s"
                     % (
                         "eventtime",
                         "name",
                         "state",
                         "goal state",
                         "repl st",
-                        "lsn",
+                        "tli:lsn",
                         "event",
                     )
                 ]
                 + [
-                    "%32s %8s %17s/%-17s %7s %10s %s" % result
+                    "%32s %8s %17s/%-17s %3d:%7s %10s %s" % result
                     for result in events
                 ]
             )
@@ -1274,7 +1274,7 @@ def get_events(self):
             last_events_query = (
                 "select eventtime, nodename, "
                 "reportedstate, goalstate, "
-                "reportedrepstate, reportedlsn, description "
+                "reportedrepstate, reportedtli, reportedlsn, description "
                 "from pgautofailover.last_events('default', count => 20)"
             )
             return self.monitor.get_events()
@@ -1842,7 +1842,7 @@ def get_events(self):
         last_events_query = (
             "select eventtime, nodename, "
             "reportedstate, goalstate, "
-            "reportedrepstate, reportedlsn, description "
+            "reportedrepstate, reportedtli, reportedlsn, description "
             "from pgautofailover.last_events('default', count => 20)"
         )