[DPE-6189] Manage passwords with user secrets

actions.yaml

@@ -20,14 +20,6 @@ create-replication:
       default: default
 get-primary:
   description: Get the unit which is the primary/leader in the replication.
-get-password:
-  description: Get the system user's password, which is used by charm.
-    It is for internal charm users and SHOULD NOT be used by applications.
-  params:
-    username:
-      type: string
-      description: The username, the default value 'operator'.
-        Possible values - operator, replication, rewind, patroni.
 list-backups:
   description: Lists backups in s3 storage.
 pre-upgrade-check:
@@ -53,17 +45,6 @@ restore:
     restore-to-time:
       type: string
       description: Point-in-time-recovery target in PSQL format.
-set-password:
-  description: Change the system user's password, which is used by charm.
-    It is for internal charm users and SHOULD NOT be used by applications.
-  params:
-    username:
-      type: string
-      description: The username, the default value 'operator'.
-        Possible values - operator, replication, rewind.
-    password:
-      type: string
-      description: The password will be auto-generated if this option is not specified.
 set-tls-private-key:
   description: Set the private key, which will be used for certificate signing requests (CSR). Run for each unit separately.
   params:

config.yaml

@@ -761,6 +761,14 @@ options:
       Allowed values are: from -1 to 86400.
     type: int
     default: -1
+  system-users:
+    type: secret
+    description: |
+      Configure the internal system users and their passwords. The passwords will
+      be auto-generated if this option is not set. It is for internal use only
+      and SHOULD NOT be used by applications. This needs to be a Juju Secret URI pointing
+      to a secret that contains the following content: `<username>: <password>`.
+      Possible users: backup, monitoring, operator, replication, rewind.
   vacuum_autovacuum_analyze_scale_factor:
     description: |
       Specifies a fraction of the table size to add to autovacuum_vacuum_threshold when

src/charm.py

@@ -41,6 +41,7 @@
     InstallEvent,
     LeaderElectedEvent,
     RelationDepartedEvent,
+    SecretChangedEvent,
     StartEvent,
 )
 from ops.framework import EventBase
@@ -50,6 +51,7 @@
     MaintenanceStatus,
     ModelError,
     Relation,
+    SecretNotFoundError,
     Unit,
     WaitingStatus,
 )
@@ -180,10 +182,10 @@ def __init__(self, *args):
         self.framework.observe(self.on.get_primary_action, self._on_get_primary)
         self.framework.observe(self.on[PEER].relation_changed, self._on_peer_relation_changed)
         self.framework.observe(self.on.secret_changed, self._on_peer_relation_changed)
+        # add specific handler for updated system-user secrets
+        self.framework.observe(self.on.secret_changed, self._on_secret_changed)
         self.framework.observe(self.on[PEER].relation_departed, self._on_peer_relation_departed)
         self.framework.observe(self.on.start, self._on_start)
-        self.framework.observe(self.on.get_password_action, self._on_get_password)
-        self.framework.observe(self.on.set_password_action, self._on_set_password)
         self.framework.observe(self.on.promote_to_primary_action, self._on_promote_to_primary)
         self.framework.observe(self.on.update_status, self._on_update_status)
         self.cluster_name = self.app.name
@@ -328,6 +330,25 @@ def remove_secret(self, scope: Scopes, key: str) -> None:
         secret_key = self._translate_field_to_secret_key(key)
         self.peer_relation_data(scope).delete_relation_data(peers.id, [secret_key])
 
+    def get_secret_from_id(self, secret_id: str) -> dict[str, str]:
+        """Resolve the given id of a Juju secret and return the content as a dict.
+
+        This method can be used to retrieve any secret, not just those used via the peer relation.
+        If the secret is not owned by the charm, it has to be granted access to it.
+
+        Args:
+            secret_id (str): The id of the secret.
+
+        Returns:
+            dict: The content of the secret.
+        """
+        try:
+            secret_content = self.model.get_secret(id=secret_id).get_content(refresh=True)
+        except (SecretNotFoundError, ModelError):
+            raise
+
+        return secret_content
+
     @property
     def is_cluster_initialised(self) -> bool:
         """Returns whether the cluster is already initialised."""
@@ -718,6 +739,17 @@ def _on_peer_relation_changed(self, event: HookEvent):
 
         self._update_new_unit_status()
 
+    def _on_secret_changed(self, event: SecretChangedEvent) -> None:
+        """Handle the secret_changed event."""
+        if not self.unit.is_leader():
+            return
+
+        if (admin_secret_id := self.config.system_users) and admin_secret_id == event.secret.id:
+            try:
+                self._update_admin_password(admin_secret_id)
+            except PostgreSQLUpdateUserPasswordError:
+                event.defer()
+
     # Split off into separate function, because of complexity _on_peer_relation_changed
     def _start_stop_pgbackrest_service(self, event: HookEvent) -> None:
         # Start or stop the pgBackRest TLS server service when TLS certificate change.
@@ -1048,8 +1080,19 @@ def _on_install(self, event: InstallEvent) -> None:
 
         self.unit.status = WaitingStatus("waiting to start PostgreSQL")
 
-    def _on_leader_elected(self, event: LeaderElectedEvent) -> None:
+    def _on_leader_elected(self, event: LeaderElectedEvent) -> None:  # noqa: C901
         """Handle the leader-elected event."""
+        # consider configured system user passwords
+        system_user_passwords = {}
+        if admin_secret_id := self.config.system_users:
+            try:
+                system_user_passwords = self.get_secret_from_id(secret_id=admin_secret_id)
+            except (ModelError, SecretNotFoundError) as e:
+                # only display the error but don't return to make sure all users have passwords
+                logger.error(f"Error setting internal passwords: {e}")
+                self.unit.status = BlockedStatus("Password setting for system users failed.")
+                event.defer()
+
         # The leader sets the needed passwords if they weren't set before.
         for key in (
             USER_PASSWORD_KEY,
@@ -1060,7 +1103,14 @@ def _on_leader_elected(self, event: LeaderElectedEvent) -> None:
             PATRONI_PASSWORD_KEY,
         ):
             if self.get_secret(APP_SCOPE, key) is None:
-                self.set_secret(APP_SCOPE, key, new_password())
+                if key in system_user_passwords:
+                    # use provided passwords for system-users if available
+                    self.set_secret(APP_SCOPE, key, system_user_passwords[key])
+                    logger.info(f"Using configured password for {key}")
+                else:
+                    # generate a password for this user if not provided
+                    self.set_secret(APP_SCOPE, key, new_password())
+                    logger.info(f"Generated new password for {key}")
 
         if self.has_raft_keys():
             self._raft_reinitialisation()
@@ -1134,6 +1184,12 @@ def _on_config_changed(self, event) -> None:
         # Enable and/or disable the extensions.
         self.enable_disable_extensions()
 
+        if admin_secret_id := self.config.system_users:
+            try:
+                self._update_admin_password(admin_secret_id)
+            except PostgreSQLUpdateUserPasswordError:
+                event.defer()
+
     def enable_disable_extensions(self, database: str | None = None) -> None:
         """Enable/disable PostgreSQL extensions set through config options.
 
@@ -1362,57 +1418,21 @@ def _start_replica(self, event) -> None:
         # Configure Patroni in the replica but don't start it yet.
         self._patroni.configure_patroni_on_unit()
 
-    def _on_get_password(self, event: ActionEvent) -> None:
-        """Returns the password for a user as an action response.
-
-        If no user is provided, the password of the operator user is returned.
-        """
-        username = event.params.get("username", USER)
-        if username not in PASSWORD_USERS:
-            event.fail(
-                f"The action can be run only for users used by the charm or Patroni:"
-                f" {', '.join(PASSWORD_USERS)} not {username}"
-            )
-            return
-        event.set_results({"password": self.get_secret(APP_SCOPE, f"{username}-password")})
-
-    def _on_set_password(self, event: ActionEvent) -> None:
-        """Set the password for the specified user."""
-        # Only leader can write the new password into peer relation.
-        if not self.unit.is_leader():
-            event.fail("The action can be run only on leader unit")
-            return
-
-        username = event.params.get("username", USER)
-        if username not in SYSTEM_USERS:
-            event.fail(
-                f"The action can be run only for users used by the charm:"
-                f" {', '.join(SYSTEM_USERS)} not {username}"
-            )
-            return
-
-        password = event.params.get("password", new_password())
-
-        if password == self.get_secret(APP_SCOPE, f"{username}-password"):
-            event.log("The old and new passwords are equal.")
-            event.set_results({"password": password})
-            return
-
-        # Ensure all members are ready before trying to reload Patroni
-        # configuration to avoid errors (like the API not responding in
-        # one instance because PostgreSQL and/or Patroni are not ready).
+    def _update_admin_password(self, admin_secret_id: str) -> None:
+        """Check if the password of a system user was changed and update it in the database."""
         if not self._patroni.are_all_members_ready():
-            event.fail(
+            # Ensure all members are ready before reloading Patroni configuration to avoid errors
+            # e.g. API not responding in one instance because PostgreSQL / Patroni are not ready
+            raise PostgreSQLUpdateUserPasswordError(
                 "Failed changing the password: Not all members healthy or finished initial sync."
             )
-            return
 
         replication_offer_relation = self.model.get_relation(REPLICATION_OFFER_RELATION)
+        other_cluster_primary_ip = ""
         if (
             replication_offer_relation is not None
             and not self.async_replication.is_primary_cluster()
         ):
-            # Update the password in the other cluster PostgreSQL primary instance.
             other_cluster_endpoints = self.async_replication.get_all_primary_cluster_endpoints()
             other_cluster_primary = self._patroni.get_primary(
                 alternative_endpoints=other_cluster_endpoints
@@ -1422,37 +1442,51 @@ def _on_set_password(self, event: ActionEvent) -> None:
                 for unit in replication_offer_relation.units
                 if unit.name.replace("/", "-") == other_cluster_primary
             )
-            try:
-                self.postgresql.update_user_password(
-                    username, password, database_host=other_cluster_primary_ip
-                )
-            except PostgreSQLUpdateUserPasswordError as e:
-                logger.exception(e)
-                event.fail("Failed changing the password.")
-                return
         elif self.model.get_relation(REPLICATION_CONSUMER_RELATION) is not None:
-            event.fail(
-                "Failed changing the password: This action can be ran only in the cluster from the offer side."
+            logger.error(
+                "Failed changing the password: This can be ran only in the cluster from the offer side."
             )
+            self.unit.status = BlockedStatus("Password update for system users failed.")
             return
-        else:
-            # Update the password in this cluster PostgreSQL primary instance.
-            try:
-                self.postgresql.update_user_password(username, password)
-            except PostgreSQLUpdateUserPasswordError as e:
-                logger.exception(e)
-                event.fail("Failed changing the password.")
-                return
 
-        # Update the password in the secret store.
-        self.set_secret(APP_SCOPE, f"{username}-password", password)
+        try:
+            # get the secret content and check each user configured there
+            # only SYSTEM_USERS with changed passwords are processed, all others ignored
+            updated_passwords = self.get_secret_from_id(secret_id=admin_secret_id)
+            for user, password in list(updated_passwords.items()):
+                if user not in SYSTEM_USERS:
+                    logger.error(
+                        f"Can only update system users: {', '.join(SYSTEM_USERS)} not {user}"
+                    )
+                    updated_passwords.pop(user)
+                    continue
+                if password == self.get_secret(APP_SCOPE, f"{user}-password"):
+                    updated_passwords.pop(user)
+        except (ModelError, SecretNotFoundError) as e:
+            logger.error(f"Error updating internal passwords: {e}")
+            self.unit.status = BlockedStatus("Password update for system users failed.")
+            return
+
+        try:
+            # perform the actual password update for the remaining users
+            for user, password in updated_passwords.items():
+                logger.info(f"Updating password for user {user}")
+                self.postgresql.update_user_password(
+                    user,
+                    password,
+                    database_host=other_cluster_primary_ip if other_cluster_primary_ip else None,
+                )
+                # Update the password in the secret store after updating it in the database
+                self.set_secret(APP_SCOPE, f"{user}-password", password)
+        except PostgreSQLUpdateUserPasswordError as e:
+            logger.exception(e)
+            self.unit.status = BlockedStatus("Password update for system users failed.")
+            return
 
         # Update and reload Patroni configuration in this unit to use the new password.
         # Other units Patroni configuration will be reloaded in the peer relation changed event.
         self.update_config()
 
-        event.set_results({"password": password})
-
     def _on_promote_to_primary(self, event: ActionEvent) -> None:
         if event.params.get("scope") == "cluster":
             return self.async_replication.promote_to_primary(event)

src/config.py

@@ -165,6 +165,7 @@ class CharmConfig(BaseConfigModel):
     storage_default_table_access_method: str | None
     storage_gin_pending_list_limit: int | None
     storage_old_snapshot_threshold: int | None
+    system_users: str | None
     vacuum_autovacuum_analyze_scale_factor: float | None
     vacuum_autovacuum_analyze_threshold: int | None
     vacuum_autovacuum_freeze_max_age: int | None

src/constants.py

@@ -66,6 +66,7 @@
 PATRONI_PASSWORD_KEY = "patroni-password"  # noqa: S105
 SECRET_INTERNAL_LABEL = "internal-secret"  # noqa: S105
 SECRET_DELETED_LABEL = "None"  # noqa: S105
+SYSTEM_USERS_PASSWORD_CONFIG = "system-users"  # noqa: S105
 
 APP_SCOPE = "app"
 UNIT_SCOPE = "unit"

tests/integration/ha_tests/conftest.py

@@ -44,10 +44,7 @@ async def loop_wait(ops_test: OpsTest) -> None:
     initial_loop_wait = await get_patroni_setting(ops_test, "loop_wait")
     yield
     # Rollback to the initial configuration.
-    app = await app_name(ops_test)
-    patroni_password = await get_password(
-        ops_test, ops_test.model.applications[app].units[0].name, "patroni"
-    )
+    patroni_password = await get_password(ops_test, "patroni")
     await change_patroni_setting(
         ops_test, "loop_wait", initial_loop_wait, patroni_password, use_random_unit=True
     )
@@ -57,10 +54,7 @@ async def loop_wait(ops_test: OpsTest) -> None:
 async def primary_start_timeout(ops_test: OpsTest) -> None:
     """Temporary change the primary start timeout configuration."""
     # Change the parameter that makes the primary reelection faster.
-    app = await app_name(ops_test)
-    patroni_password = await get_password(
-        ops_test, ops_test.model.applications[app].units[0].name, "patroni"
-    )
+    patroni_password = await get_password(ops_test, "patroni")
     initial_primary_start_timeout = await get_patroni_setting(ops_test, "primary_start_timeout")
     await change_patroni_setting(ops_test, "primary_start_timeout", 0, patroni_password)
     yield
@@ -104,7 +98,7 @@ async def wal_settings(ops_test: OpsTest) -> None:
     for unit in ops_test.model.applications[app].units:
         # Start Patroni if it was previously stopped.
         await run_command_on_unit(ops_test, unit.name, "snap start charmed-postgresql.patroni")
-        patroni_password = await get_password(ops_test, unit.name, "patroni")
+        patroni_password = await get_password(ops_test, "patroni")
 
         # Rollback to the initial settings.
         await change_wal_settings(