Merge pull request ceph#62808 from adk3798/mgr-rgw-fix-re-realm-bootstrap-after-deletions

mgr/rgw: don't fail realm bootstrap if system user exists already

Reviewed-by: Redouane Kachach <[email protected]>

Author: adk3798

src/pybind/mgr/rgw/module.py

@@ -210,7 +210,33 @@ def _cmd_rgw_realm_bootstrap(self,
                 RGWAM(self.env).realm_bootstrap(spec, start_radosgw)
         except RGWAMException as e:
             self.log.error('cmd run exception: (%d) %s' % (e.retcode, e.message))
-            return HandleCommandResult(retval=e.retcode, stdout=e.stdout, stderr=e.stderr)
+            # The RGWAM code isn't always consistent about what goes into stdout
+            # and message fields. What is known is there are definitely some cases
+            # where the message field contains useful info and the stdout field
+            # does not, which means just giving the stdout can result in not so useful
+            # return messages, e.g.
+            #
+            # [ceph: root@vm-00 /]# ceph rgw realm bootstrap -i specs/rgw.yaml
+            # Error EEXIST:
+            #
+            # For that reason, we want to include the msg in what we return. Doing so
+            # transformed the same error case that just gave us "Error EEXIST:" into
+            #
+            # [ceph: root@vm-00 /]# ceph rgw realm bootstrap -i specs/rgw.yaml
+            # failed to create system user: Command error (-17): user create --uid sysuser-my_realm_ck
+            # --display-name sysuser-my_realm_ck --system --rgw-zonegroup my_zonegroup_ck --zonegroup-id
+            # 428a28d1-c8a9-4c12-a408-29507ef23842 --rgw-zone my_zone_ck --zone-id 161ee2ab-bc96-4729-a2b5-0d170e347129
+            # stdout:
+            # stderr:could not create user: unable to parse parameters, user: sysuser-my_realm_ck exists
+            # Error EEXIST:
+            #
+            # which is much more useful
+            msg = e.message
+            if msg and e.stdout and (e.stdout not in msg):
+                msg = f'{msg}; {e.stdout}'
+            elif e.stdout:
+                msg = e.stdout
+            return HandleCommandResult(retval=e.retcode, stdout=msg, stderr=e.stderr)
         except PoolCreationError as e:
             self.log.error(f'Pool creation failure: {str(e)}')
             return HandleCommandResult(retval=-errno.EINVAL, stderr=str(e))

src/python-common/ceph/rgw/rgwam_core.py

@@ -503,16 +503,38 @@ def create_zone(self, realm, zg, zone_name, zone_is_master, access_key=None,
 
     def create_system_user(self, realm, zonegroup, zone):
         try:
-            sys_user_info = self.user_op().create(zone,
-                                                  zonegroup,
-                                                  uid=f'sysuser-{realm.name}',
-                                                  uid_prefix='user-sys',
-                                                  is_system=True)
+            sys_user_info = self.user_op().create(
+                zone,
+                zonegroup,
+                uid=f'sysuser-{realm.name}',
+                uid_prefix='user-sys',
+                is_system=True
+            )
             sys_user = RGWUser(sys_user_info)
             logging.info(f'Created system user: {sys_user.uid} on'
-                         '{realm.name}/{zonegroup.name}/{zone.name}')
+                         f'{realm.name}/{zonegroup.name}/{zone.name}')
             return sys_user
         except RGWAMException as e:
+            if e.retcode == -errno.EEXIST:
+                # You get this error (EEXIST) when the user already exists. This
+                # can happen if you delete the zone/zg/realm for the user but not
+                # the user itself and then try to call "rgw realm bootstrap"
+                # with the same zone/zg/realm names again. In this case, let's try
+                # to get the existing user's info
+                try:
+                    sys_user_info = self.user_op().info(
+                        zone,
+                        zonegroup,
+                        uid=f'sysuser-{realm.name}',
+                    )
+                    sys_user = RGWUser(sys_user_info)
+                    logging.info(f'Found existing system user: sysuser-{realm.name}')
+                    return sys_user
+                except RGWAMException as e2:
+                    RGWAMException(
+                        f'System user sysuser-{realm.name} already existed. '
+                        'Failed getting info for user', e2
+                    )
             raise RGWAMException('failed to create system user', e)
 
     def create_normal_user(self, zg, zone, uid=None):