Add OSD removal force option

Ensures a removal can continue even in situations where some step(s)
might fail, for instance removing an obsolete OSD from a replaced node.

api-daemon/pvcapid/flaskapi.py

@@ -4099,11 +4099,16 @@ class API_Storage_Ceph_OSD_Element(Resource):
 
     @RequestParser(
         [
+            {
+                "name": "force",
+                "required": False,
+                "helptext": "Force removal even if steps fail.",
+            },
             {
                 "name": "yes-i-really-mean-it",
                 "required": True,
                 "helptext": "Please confirm that 'yes-i-really-mean-it'.",
-            }
+            },
         ]
     )
     @Authenticator
@@ -4116,6 +4121,11 @@ class API_Storage_Ceph_OSD_Element(Resource):
         tags:
           - storage / ceph
         parameters:
+          - in: query
+            name: force
+            type: boolean
+            required: flase
+            description: Force removal even if some step(s) fail
           - in: query
             name: yes-i-really-mean-it
             type: string
@@ -4138,7 +4148,7 @@ class API_Storage_Ceph_OSD_Element(Resource):
               type: object
               id: Message
         """
-        return api_helper.ceph_osd_remove(osdid)
+        return api_helper.ceph_osd_remove(osdid, reqargs.get("force", False))
 
 
 api.add_resource(API_Storage_Ceph_OSD_Element, "/storage/ceph/osd/<osdid>")

api-daemon/pvcapid/helper.py

@@ -1302,11 +1302,11 @@ def ceph_osd_add(zkhandler, node, device, weight, ext_db_flag=False, ext_db_rati
 
 
 @ZKConnection(config)
-def ceph_osd_remove(zkhandler, osd_id):
+def ceph_osd_remove(zkhandler, osd_id, force_flag):
     """
     Remove a Ceph OSD from the PVC Ceph storage cluster.
     """
-    retflag, retdata = pvc_ceph.remove_osd(zkhandler, osd_id)
+    retflag, retdata = pvc_ceph.remove_osd(zkhandler, osd_id, force_flag)
 
     if retflag:
         retcode = 200

client-cli/pvc/cli_lib/ceph.py

@@ -255,7 +255,7 @@ def ceph_osd_add(config, node, device, weight, ext_db_flag, ext_db_ratio):
     return retstatus, response.json().get("message", "")
 
 
-def ceph_osd_remove(config, osdid):
+def ceph_osd_remove(config, osdid, force_flag):
     """
     Remove Ceph OSD
 
@@ -263,7 +263,7 @@ def ceph_osd_remove(config, osdid):
     API arguments:
     API schema: {"message":"{data}"}
     """
-    params = {"yes-i-really-mean-it": "yes"}
+    params = {"force": force_flag, "yes-i-really-mean-it": "yes"}
     response = call_api(
         config, "delete", "/storage/ceph/osd/{osdid}".format(osdid=osdid), params=params
     )

client-cli/pvc/pvc.py

@@ -3376,6 +3376,14 @@ def ceph_osd_add(node, device, weight, ext_db_flag, ext_db_ratio, confirm_flag):
 ###############################################################################
 @click.command(name="remove", short_help="Remove OSD.")
 @click.argument("osdid")
+@click.option(
+    "-f",
+    "--force",
+    "force_flag",
+    is_flag=True,
+    default=False,
+    help="Force removal even if steps fail",
+)
 @click.option(
     "-y",
     "--yes",
@@ -3385,11 +3393,13 @@ def ceph_osd_add(node, device, weight, ext_db_flag, ext_db_ratio, confirm_flag):
     help="Confirm the removal",
 )
 @cluster_req
-def ceph_osd_remove(osdid, confirm_flag):
+def ceph_osd_remove(osdid, force_flag, confirm_flag):
     """
     Remove a Ceph OSD with ID OSDID.
 
     DANGER: This will completely remove the OSD from the cluster. OSDs will rebalance which will negatively affect performance and available space. It is STRONGLY RECOMMENDED to set an OSD out (using 'pvc storage osd out') and allow the cluster to fully rebalance (verified with 'pvc storage status') before removing an OSD.
+
+    NOTE: The "-f"/"--force" option is useful after replacing a failed node, to ensure the OSD is removed even if the OSD in question does not properly exist on the node after a rebuild.
     """
     if not confirm_flag and not config["unsafe"]:
         try:
@@ -3397,7 +3407,7 @@ def ceph_osd_remove(osdid, confirm_flag):
         except Exception:
             exit(0)
 
-    retcode, retmsg = pvc_ceph.ceph_osd_remove(config, osdid)
+    retcode, retmsg = pvc_ceph.ceph_osd_remove(config, osdid, force_flag)
     cleanup(retcode, retmsg)
 
 

daemon-common/ceph.py

@@ -286,14 +286,14 @@ def add_osd(zkhandler, node, device, weight, ext_db_flag=False, ext_db_ratio=0.0
     return success, message
 
 
-def remove_osd(zkhandler, osd_id):
+def remove_osd(zkhandler, osd_id, force_flag):
     if not verifyOSD(zkhandler, osd_id):
         return False, 'ERROR: No OSD with ID "{}" is present in the cluster.'.format(
             osd_id
         )
 
     # Tell the cluster to remove an OSD
-    remove_osd_string = "osd_remove {}".format(osd_id)
+    remove_osd_string = "osd_remove {} {}".format(osd_id, str(force_flag))
     zkhandler.write([("base.cmd.ceph", remove_osd_string)])
     # Wait 1/2 second for the cluster to get the message and start working
     time.sleep(0.5)

node-daemon/pvcnoded/objects/CephInstance.py

@@ -310,7 +310,7 @@ class CephOSDInstance(object):
             return False
 
     @staticmethod
-    def remove_osd(zkhandler, logger, osd_id, osd_obj):
+    def remove_osd(zkhandler, logger, osd_id, osd_obj, force_flag):
         logger.out("Removing OSD disk {}".format(osd_id), state="i")
         try:
             # Verify the OSD is present
@@ -320,7 +320,10 @@ class CephOSDInstance(object):
                 logger.out(
                     "Could not find OSD {} in the cluster".format(osd_id), state="e"
                 )
-                return True
+                if force_flag:
+                    logger.out("Ignoring error due to force flag", state="i")
+                else:
+                    return True
 
             # 1. Set the OSD down and out so it will flush
             logger.out("Setting down OSD disk with ID {}".format(osd_id), state="i")
@@ -331,7 +334,10 @@ class CephOSDInstance(object):
                 print("ceph osd down")
                 print(stdout)
                 print(stderr)
-                raise Exception
+                if force_flag:
+                    logger.out("Ignoring error due to force flag", state="i")
+                else:
+                    raise Exception
 
             logger.out("Setting out OSD disk with ID {}".format(osd_id), state="i")
             retcode, stdout, stderr = common.run_os_command(
@@ -341,7 +347,10 @@ class CephOSDInstance(object):
                 print("ceph osd out")
                 print(stdout)
                 print(stderr)
-                raise Exception
+                if force_flag:
+                    logger.out("Ignoring error due to force flag", state="i")
+                else:
+                    raise Exception
 
             # 2. Wait for the OSD to flush
             logger.out("Flushing OSD disk with ID {}".format(osd_id), state="i")
@@ -359,7 +368,11 @@ class CephOSDInstance(object):
                     if num_pgs > 0:
                         time.sleep(5)
                     else:
-                        raise Exception
+                        if force_flag:
+                            logger.out("Ignoring error due to force flag", state="i")
+                        else:
+                            raise Exception
+
                 except Exception:
                     break
 
@@ -372,7 +385,10 @@ class CephOSDInstance(object):
                 print("systemctl stop")
                 print(stdout)
                 print(stderr)
-                raise Exception
+                if force_flag:
+                    logger.out("Ignoring error due to force flag", state="i")
+                else:
+                    raise Exception
 
             # FIXME: There has to be a better way to do this /shrug
             while True:
@@ -408,7 +424,10 @@ class CephOSDInstance(object):
                 print("ceph-volume lvm zap")
                 print(stdout)
                 print(stderr)
-                raise Exception
+                if force_flag:
+                    logger.out("Ignoring error due to force flag", state="i")
+                else:
+                    raise Exception
 
             # 6. Purge the OSD from Ceph
             logger.out("Purging OSD disk with ID {}".format(osd_id), state="i")
@@ -419,7 +438,10 @@ class CephOSDInstance(object):
                 print("ceph osd purge")
                 print(stdout)
                 print(stderr)
-                raise Exception
+                if force_flag:
+                    logger.out("Ignoring error due to force flag", state="i")
+                else:
+                    raise Exception
 
             # 7. Remove the DB device
             if zkhandler.exists(("osd.db_device", osd_id)):
@@ -734,7 +756,8 @@ def ceph_command(zkhandler, logger, this_node, data, d_osd):
 
     # Removing an OSD
     elif command == "osd_remove":
-        osd_id = args
+        osd_id = args[0]
+        force_flag = bool(strtobool(args[1]))
 
         # Verify osd_id is in the list
         if d_osd[osd_id] and d_osd[osd_id].node == this_node.name:
@@ -743,7 +766,7 @@ def ceph_command(zkhandler, logger, this_node, data, d_osd):
             with zk_lock:
                 # Remove the OSD
                 result = CephOSDInstance.remove_osd(
-                    zkhandler, logger, osd_id, d_osd[osd_id]
+                    zkhandler, logger, osd_id, d_osd[osd_id], force_flag
                 )
                 # Command succeeded
                 if result: