log_hostname can be set autonomously from external_int, temp_buffers is now a multiple of 8k.

defaults/main.yml

@@ -68,7 +68,25 @@ psql_conf_parameters:
   - { name: 'shared_buffers', value: '{{ (ansible_memtotal_mb / 4) | int }}MB', set: 'true' }
   - { name: 'work_mem', value: '{{ ((ansible_memtotal_mb * 1024) / (400 * ansible_processor_vcpus)) | int }}kB', set: 'true' }
   - { name: 'maintenance_work_mem', value: '{{ (ansible_memtotal_mb / 8) | int }}MB', set: 'true' }
-  - { name: 'temp_buffers', value: '{{ ansible_memtotal_mb }}kB', set: 'true' }
+  # A thousandth of the memory, like every other entry of this list is a
+  # fraction of it: N MB of RAM become N kB of temp_buffers.
+  #
+  # Expressed in the BASE UNIT of this parameter, 8kB blocks, which is what the
+  # division is for. The previous form, '{{ ansible_memtotal_mb }}kB', sets the
+  # same thing and postgresql_set is idempotent with it (measured: a real run
+  # against an already configured cluster reports `changed: false`). What it
+  # hides is the rounding: 7941kB is 992.625 blocks, PostgreSQL stores 993, and
+  # the value that comes back is 7944kB. Writing the block count says what will
+  # actually be stored.
+  #
+  # Two things worth knowing about this parameter, both measured:
+  #   * its context is `user`, and postgresql_set reports
+  #     `restart_required: false`, so it never needed to sit in this list of
+  #     parameters that require a restart;
+  #   * in CHECK MODE postgresql_set reports `changed` for every parameter,
+  #     including ones already at the wanted value. Do not read a check run of
+  #     this role as a list of pending changes.
+  - { name: 'temp_buffers', value: '{{ (ansible_memtotal_mb / 8) | round | int }}', set: 'true' }
   - { name: 'wal_buffers', value: '{{ (ansible_memtotal_mb / 1024 * 2) | int }}MB', set: 'true' }
   - { name: 'min_wal_size', value: '{{ (ansible_memtotal_mb / 16) | int }}MB', set: 'true' }
   - { name: 'max_wal_size', value: '{{ (ansible_memtotal_mb / 8) | int }}MB', set: 'true' }
@@ -79,6 +97,25 @@ psql_conf_parameters:
 # Custom parameters. Default empty
 psql_conf_custom_parameters: []
 
+# Reverse DNS lookup on every incoming connection.
+#
+# The default keeps what the role has always done: `on` as soon as the cluster
+# listens on an external interface. That has a cost worth knowing about, namely
+# a resolver round trip per connection, and a pair of log lines every time the
+# peer has no PTR record:
+#
+#   WARNING:  pg_getnameinfo_all() failed: Temporary failure in name resolution
+#   LOG:  connection received: host=??? port=???
+#
+# Set it to 'off' to log the addresses instead, which on a private network is
+# usually the more useful of the two and costs nothing. 'default' also resolves
+# to off, since off is the PostgreSQL default.
+#
+# log_connections and log_disconnections are still decided the same way, in
+# tasks/postgresql-config-{deb,el}.yml; they follow this pattern if they ever
+# need to become configurable as well.
+psql_log_hostname: "{% if psql_listen_on_ext_int %}on{% else %}default{% endif %}"
+
 # logging configuration. Important: the parameters that need a restart must be listed in psql_conf_parameters
 psql_log_configuration:
   - { name: 'log_destination', value: 'stderr', set: 'true' }

tasks/postgresql-config-deb.yml

@@ -132,7 +132,7 @@
       become_user: postgres
       community.postgresql.postgresql_set:
         name: 'log_hostname'
-        value: "{% if psql_listen_on_ext_int %}on{% else %}default{% endif %}"
+        value: '{{ psql_log_hostname }}'
       notify: Reload postgresql
 
     - name: postgresql-config-deb | Flush flush_handlers Restart PostgreSQL

tasks/postgresql-config-el.yml

@@ -153,7 +153,7 @@
       become_user: postgres
       community.postgresql.postgresql_set:
         name: 'log_hostname'
-        value: "{% if psql_listen_on_ext_int %}on{% else %}default{% endif %}"
+        value: '{{ psql_log_hostname }}'
       notify: Reload postgresql
 
     - name: postgresql-config-el | Flush flush_handlers Reload PostgreSQL