Merge pull request #108 from zalando/pg_rewind_manual_failover

Call a checkpoint on master before pg_rewind.

patroni/postgresql.py

@@ -342,13 +342,13 @@ class Postgresql:
         ret and not block_callbacks and self.call_nowait(ACTION_ON_START)
         return ret
 
-    def checkpoint(self):
+    def checkpoint(self, connstring=None):
         try:
-            r = parseurl('postgres://{}/postgres'.format(self.local_address))
-            r['options'] = '-c statement_timeout=0'
-            with psycopg2.connect(**r) as conn:
+            connstring = connstring or 'postgres://{}/postgres'.format(self.local_address)
+            with psycopg2.connect(connstring) as conn:
                 conn.autocommit = True
                 with conn.cursor() as cur:
+                    cur.execute("SET statement_timeout = 0")
                     cur.execute('CHECKPOINT')
         except:
             logging.exception('Exception during CHECKPOINT')
@@ -454,6 +454,9 @@ recovery_target_timeline = 'latest'
         r['user'] = r['username']
         env = self.write_pgpass(r)
         pc = "user={user} host={host} port={port} dbname=postgres sslmode=prefer sslcompression=1".format(**r)
+        # first run a checkpoint on a promoted master in order
+        # to make it store the new timeline ([email protected])
+        self.checkpoint(pc)
         logger.info("running pg_rewind from {}".format(pc))
         pg_rewind = ['pg_rewind', '-D', self.data_dir, '--source-server', pc]
         try:

postgres0.yml

@@ -82,10 +82,12 @@ postgresql:
     #use_iam: 1
   #recovery_conf:
     #restore_command: envdir /etc/wal-e.d/env wal-e wal-fetch "%f" "%p" -p 1
+  recovery_conf:
+    restore_command: cp ../wal_archive/%f %p
   parameters:
     archive_mode: "on"
     wal_level: hot_standby
-    archive_command: mkdir -p ../wal_archive && cp %p ../wal_archive/%f
+    archive_command: mkdir -p ../wal_archive && test ! -f ../wal_archive/%f && cp %p ../wal_archive/%f
     max_wal_senders: 5
     wal_keep_segments: 8
     archive_timeout: 1800s

postgres1.yml

@@ -82,10 +82,12 @@ postgresql:
     #use_iam: 1
   #recovery_conf:
     #restore_command: envdir /etc/wal-e.d/env wal-e wal-fetch "%f" "%p" -p 1
+  recovery_conf:
+    restore_command: cp ../wal_archive/%f %p
   parameters:
     archive_mode: "on"
     wal_level: hot_standby
-    archive_command: mkdir -p ../wal_archive && cp %p ../wal_archive/%f
+    archive_command: mkdir -p ../wal_archive && test ! -f ../wal_archive/%f && cp %p ../wal_archive/%f
     max_wal_senders: 5
     wal_keep_segments: 8
     archive_timeout: 1800s