Merge lp:~stub/charms/trusty/postgresql/standby-service into lp:~stub/charms/trusty/postgresql/trunk

[Stuart Bishop]

The last landing broke PostgreSQL < 9.4 installs using the default configuration options.

Fix this by downgrading unsupported wal_level's when necessary. This seems preferable to changing the default, which would mean requiring the wal_level to be set before the new relation type would actually work.

[Stuart Bishop (stub)]

review: Abstain

Unmerged revisions

126. By Stuart Bishop on 2015-07-10

Handle unsupported wal_level with PG < 9.4

125. By Stuart Bishop on 2015-07-10

Merge trunk

124. By Stuart Bishop on 2015-05-26

Bug #1458754 workaround and tweaks

123. By Stuart Bishop on 2015-05-25

Run config_changed *after* creating the user and db it needs

122. By Stuart Bishop on 2015-05-25

External logical replication

121. By Stuart Bishop on 2015-05-21

master relation hooks

120. By Stuart Bishop on 2015-05-21

Correct relation name

119. By Stuart Bishop on 2015-05-01

Replica relation for replicas external to the service

=== modified file 'config.yaml'
--- config.yaml	2015-03-12 10:18:22 +0000
+++ config.yaml	2015-07-10 11:09:27 +0000
@@ -179,13 +179,13 @@
         transactions on a streaming hot standby and allowing the master to
         defer cleanup and avoid query cancelations on the hot standby.
   wal_level:
-    default: minimal
+    default: logical
     type: string
     description: >
-        'minimal', 'archive' or 'hot_standby'. Defines how much information
-        is written to the WAL. Set to 'minimal' for stand alone databases
-        and 'hot_standby' for replicated setups. Overridden by juju when
-        replication is used.
+        'minimal', 'archive', 'hot_standby' or 'logical'. Defines how much
+        information is written to the WAL. Set to 'minimal' for stand alone
+        databases and 'hot_standby' for replicated setups. Overridden by
+        juju when replication is used.
   max_wal_senders:
     default: 0
     type: int

=== modified file 'hooks/hooks.py' (properties changed: -x to +x)
--- hooks/hooks.py	2015-07-03 06:19:42 +0000
+++ hooks/hooks.py	2015-07-10 11:09:27 +0000
@@ -158,6 +158,9 @@
         for relid in hookenv.relation_ids('replication'):
             hookenv.relation_set(relid, replication_state)
 
+        for relid in hookenv.relation_ids('master'):
+            hookenv.relation_set(relid, state=self.get('state'))
+
         log('saving local state', DEBUG)
         self.save()
 
@@ -411,6 +414,12 @@
     host.write_file(postgresql_sysctl, ''.join(lines), perms=0600)
     _run_sysctl(postgresql_sysctl)
 
+    # Our config file specifies a default wal_level that only works
+    # with PostgreSQL 9.4. Downgrade this for earlier versions of
+    # PostgreSQL. We have this default so more things Just Work.
+    if pg_version() < '9.4' and config_data['wal_level'] == 'logical':
+        config_data['wal_level'] = 'hot_standby'
+
     # If we are replicating, some settings may need to be overridden to
     # certain minimum levels.
     num_slaves = slave_count()
@@ -418,7 +427,8 @@
         log('{} hot standbys in peer relation.'.format(num_slaves))
         log('Ensuring minimal replication settings')
         config_data['hot_standby'] = True
-        config_data['wal_level'] = 'hot_standby'
+        if config_data['wal_level'] != 'logical':
+            config_data['wal_level'] = 'hot_standby'
         config_data['wal_keep_segments'] = max(
             config_data['wal_keep_segments'],
             config_data['replicated_wal_keep_segments'])
@@ -432,12 +442,14 @@
     # non-streaming replication, or for PITR.
     if config_data.get('swiftwal_log_shipping', None):
         config_data['archive_mode'] = True
-        config_data['wal_level'] = 'hot_standby'
+        if config_data['wal_level'] != 'logical':
+            config_data['wal_level'] = 'hot_standby'
         config_data['archive_command'] = swiftwal_archive_command()
 
     if config_data.get('wal_e_storage_uri', None):
         config_data['archive_mode'] = True
-        config_data['wal_level'] = 'hot_standby'
+        if config_data['wal_level'] != 'logical':
+            config_data['wal_level'] = 'hot_standby'
         config_data['archive_command'] = wal_e_archive_command()
 
     # Send config data to the template
@@ -565,7 +577,8 @@
     # every other unit's postgres database and the magic replication
     # database. It also needs to be able to connect to its own postgres
     # database.
-    for relid in hookenv.relation_ids('replication'):
+    relids = hookenv.relation_ids('replication')
+    for relid in relids:
         for unit in hookenv.related_units(relid):
             relation = hookenv.relation_get(unit=unit, rid=relid)
             remote_addr = munge_address(relation['private-address'])
@@ -584,7 +597,34 @@
                            }
             relation_data.append(remote_pgdb)
 
-    # Hooks need permissions too to setup replication.
+    # More replication connections, this time from external services.
+    # Somewhat different than before, as we do not share credentials
+    # and services using 9.4's logical replication feature will want
+    # to specify the database name.
+    relids = hookenv.relation_ids('master')
+    for relid in relids:
+        for unit in hookenv.related_units(relid):
+            remote_rel = hookenv.relation_get(unit=unit, rid=relid)
+            local_rel = hookenv.relation_get(unit=hookenv.local_unit(),
+                                             rid=relid)
+            remote_addr = munge_address(remote_rel['private-address'])
+            remote_replication = {'database': 'replication',
+                                  'user': local_rel['user'],
+                                  'private-address': remote_addr,
+                                  'relation-id': relid,
+                                  'unit': unit,
+                                  }
+            relation_data.append(remote_replication)
+            if 'database' in local_rel:
+                remote_pgdb = {'database': local_rel['database'],
+                            'user': local_rel['user'],
+                            'private-address': remote_addr,
+                            'relation-id': relid,
+                            'unit': unit,
+                            }
+                relation_data.append(remote_pgdb)
+
+    # Local hooks also need permissions to setup replication.
     for relid in hookenv.relation_ids('replication'):
         local_replication = {'database': 'postgres',
                              'user': 'juju_replication',
@@ -2176,16 +2216,14 @@
     # Build the set of client relations that both the master and this
     # unit have joined.
     possible_client_relations = set(hookenv.relation_ids('db') +
-                                    hookenv.relation_ids('db-admin'))
+                                    hookenv.relation_ids('db-admin') +
+                                    hookenv.relation_ids('master'))
     active_client_relations = possible_client_relations.intersection(
         set(client_relations.split()))
 
     for client_relation in active_client_relations:
         # We need to pull the credentials from the master unit's
-        # end of the client relation. This is problematic as we
-        # have no way of knowing if the master unit has joined
-        # the relation yet. We use the exception handler to detect
-        # this case per Bug #1192803.
+        # end of the client relation.
         log('Hot standby republishing credentials from {} to {}'.format(
             master, client_relation))
 
@@ -2403,6 +2441,8 @@
     num_slaves = 0
     for relid in hookenv.relation_ids('replication'):
         num_slaves += len(hookenv.related_units(relid))
+    for relid in hookenv.relation_ids('master'):
+        num_slaves += len(hookenv.related_units(relid))
     return num_slaves
 
 
@@ -2480,7 +2520,7 @@
             or not metrics_sample_interval):
         log("Required config not found or invalid "
             "(metrics_target, metrics_sample_interval), "
-            "disabling metrics", WARNING)
+            "disabling statsd metrics", DEBUG)
         delete_metrics_cronjob(cron_path)
         return
 
@@ -2691,6 +2731,58 @@
     postgresql_stop()
 
 
[e-mail]('master-relation-joined', 'master-relation-changed')
+def master_relation_joined_changed():
+    local_relation = hookenv.relation_get(unit=hookenv.local_unit())
+
+    # Relation settings both master and standbys can set now.
+    allowed_units = sorted(hookenv.related_units())  # Bug #1458754
+    hookenv.relation_set(
+        relation_settings={'allowed-units': ' '.join(allowed_units),
+                           'host': hookenv.unit_private_ip(),
+                           'port': get_service_port(),
+                           'state': local_state['state'],
+                           'version': pg_version()})
+
+    if local_state['state'] == 'hot standby':
+        # Hot standbys cannot create credentials. Publish them from the
+        # master if they are available, or defer until a peer-relation-changed
+        # hook when they are.
+        publish_hot_standby_credentials()
+        config_changed()
+        return
+
+    user = local_relation.get('user') or user_name(hookenv.relation_id(),
+                                                   hookenv.remote_unit())
+    password = local_relation.get('password') or create_user(user,
+                                                             admin=True,
+                                                             replication=True)
+    hookenv.relation_set(user=user, password=password)
+
+    # For logical replication, the standby service may request an explicit
+    # database.
+    database = hookenv.relation_get('database')
+    if database:
+        ensure_database(user, user, database)
+        hookenv.relation_set(database=database)  # Signal database is ready
+
+    # We may need to bump the number of replication connections and
+    # restart, and we will certainly need to regenerate pg_hba.conf
+    # and reload.
+    config_changed()  # Must be called after db & user are created.
+
+
[e-mail]()
+def master_relation_departed():
+    config_changed()
+    allowed_units = hookenv.relation_get('allowed-units',
+                                         hookenv.local_unit()).split()
+    if hookenv.remote_unit() in allowed_units:
+        allowed_units.remove(hookenv.remote_unit())
+    hookenv.relation_set(relation_settings={
+        'allowed-units': ' '.join(allowed_units)})
+
+
 def _get_postgresql_config_dir(config_data=None):
     """ Return the directory path of the postgresql configuration files. """
     if config_data is None:

=== removed symlink 'hooks/master-relation-broken'
=== target was u'hooks.py'
=== added symlink 'hooks/master-relation-departed'
=== target is u'hooks.py'
=== modified file 'metadata.yaml'
--- metadata.yaml	2014-05-29 13:08:35 +0000
+++ metadata.yaml	2015-07-10 11:09:27 +0000
@@ -18,8 +18,10 @@
 provides:
   db:
     interface: pgsql
+    optional: true
   db-admin:
     interface: pgsql
+    optional: true
   nrpe-external-master:
     interface: nrpe-external-master
     scope: container
@@ -27,6 +29,14 @@
     interface: block-storage
     scope: container
     optional: true
+  master:
+    interface: pgreplication
+    optional: true
+  # We do not yet support slaving a PostgreSQL service to another
+  # PostgreSQL service.
+  # replica:
+  #   interface: pgreplication
+  #   optional: true
 requires:
   persistent-storage:
     interface: directory-path
@@ -34,4 +44,4 @@
     interface: syslog
 peers:
   replication:
-    interface: pgreplication
+    interface: pgpeer

=== modified file 'templates/postgresql.conf.tmpl'
--- templates/postgresql.conf.tmpl	2015-03-12 10:18:22 +0000
+++ templates/postgresql.conf.tmpl	2015-07-10 11:09:27 +0000
@@ -184,7 +184,6 @@
 standard_conforming_strings = {{standard_conforming_strings}}
 {% endif -%}
 
-{% if version >= "9.0" -%}
 #------------------------------------------------------------------------------
 # Replication
 #------------------------------------------------------------------------------
@@ -199,6 +198,9 @@
 {% endif -%}
 {% if max_wal_senders  != "" -%}
 max_wal_senders = {{max_wal_senders}}
+{% if version >= "9.4" -%}
+max_replication_slots = {{max_wal_senders}}
+{% endif -%}
 {% endif -%}
 {% if wal_keep_segments  != "" -%}
 wal_keep_segments = {{wal_keep_segments}}
@@ -209,5 +211,3 @@
 {% if archive_command != "" -%}
 archive_command = '{{archive_command}}'
 {% endif -%}
-
-{% endif -%}