Fix bugs with forced removal

client-cli/pvc/cli_lib/ceph.py

@@ -406,7 +406,7 @@ def format_list_osd(osd_list):
             osd_id_length = _osd_id_length
 
         # Set the OSD node length
-        _osd_node_length = len(osd_information["stats"]["node"]) + 1
+        _osd_node_length = len(osd_information["node"]) + 1
         if _osd_node_length > osd_node_length:
             osd_node_length = _osd_node_length
 
@@ -602,13 +602,6 @@ def format_list_osd(osd_list):
     )
 
     for osd_information in sorted(osd_list, key=lambda x: int(x["id"])):
-        try:
-            # If this happens, the node hasn't checked in fully yet, so just ignore it
-            if osd_information["stats"]["node"] == "|":
-                continue
-        except KeyError:
-            continue
-
         osd_up_flag, osd_up_colour, osd_in_flag, osd_in_colour = getOutputColoursOSD(
             osd_information
         )
@@ -663,7 +656,7 @@ def format_list_osd(osd_list):
                 osd_rdops_length=osd_rdops_length,
                 osd_rddata_length=osd_rddata_length,
                 osd_id=osd_information["id"],
-                osd_node=osd_information["stats"]["node"],
+                osd_node=osd_information["node"],
                 osd_device=osd_information["device"],
                 osd_db_device=osd_db_device,
                 osd_up_colour=osd_up_colour,

daemon-common/ceph.py

@@ -181,6 +181,7 @@ def getClusterOSDList(zkhandler):
 
 def getOSDInformation(zkhandler, osd_id):
     # Get the devices
+    osd_node = zkhandler.read(("osd.node", osd_id))
     osd_device = zkhandler.read(("osd.device", osd_id))
     osd_db_device = zkhandler.read(("osd.db_device", osd_id))
     # Parse the stats data
@@ -189,6 +190,7 @@ def getOSDInformation(zkhandler, osd_id):
 
     osd_information = {
         "id": osd_id,
+        "node": osd_node,
         "device": osd_device,
         "db_device": osd_db_device,
         "stats": osd_stats,
@@ -293,7 +295,7 @@ def remove_osd(zkhandler, osd_id, force_flag):
         )
 
     # Tell the cluster to remove an OSD
-    remove_osd_string = "osd_remove {} {}".format(osd_id, str(force_flag))
+    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

@@ -404,30 +404,47 @@ class CephOSDInstance(object):
                     break
 
             # 4. Determine the block devices
-            retcode, stdout, stderr = common.run_os_command(
+            device_zk = zkhandler.read(("osd.device", osd_id))
-                "readlink /var/lib/ceph/osd/ceph-{}/block".format(osd_id)
+            try:
-            )
+                retcode, stdout, stderr = common.run_os_command(
-            vg_name = stdout.split("/")[-2]  # e.g. /dev/ceph-<uuid>/osd-block-<uuid>
+                    "readlink /var/lib/ceph/osd/ceph-{}/block".format(osd_id)
-            retcode, stdout, stderr = common.run_os_command(
+                )
-                "vgs --separator , --noheadings -o pv_name {}".format(vg_name)
+                vg_name = stdout.split("/")[
-            )
+                    -2
-            pv_block = stdout.strip()
+                ]  # e.g. /dev/ceph-<uuid>/osd-block-<uuid>
+                retcode, stdout, stderr = common.run_os_command(
+                    "vgs --separator , --noheadings -o pv_name {}".format(vg_name)
+                )
+                pv_block = stdout.strip()
+            except Exception as e:
+                print(e)
+                pv_block = device_zk
 
-            # 5. Zap the volumes
+            # 5a. Verify that the blockdev actually has a ceph volume that matches the ID, otherwise don't zap it
             logger.out(
-                "Zapping OSD disk with ID {} on {}".format(osd_id, pv_block), state="i"
+                f"Check OSD disk {pv_block} for OSD signature with ID osd.{osd_id}",
+                state="i",
             )
             retcode, stdout, stderr = common.run_os_command(
-                "ceph-volume lvm zap --destroy {}".format(pv_block)
+                f"ceph-volume lvm list {pv_block}"
             )
-            if retcode:
+            if f"====== osd.{osd_id} =======" in stdout:
-                print("ceph-volume lvm zap")
+                # 5b. Zap the volumes
-                print(stdout)
+                logger.out(
-                print(stderr)
+                    "Zapping OSD disk with ID {} on {}".format(osd_id, pv_block),
-                if force_flag:
+                    state="i",
-                    logger.out("Ignoring error due to force flag", state="i")
+                )
-                else:
+                retcode, stdout, stderr = common.run_os_command(
-                    raise Exception
+                    "ceph-volume lvm zap --destroy {}".format(pv_block)
+                )
+                if retcode:
+                    print("ceph-volume lvm zap")
+                    print(stdout)
+                    print(stderr)
+                    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")
@@ -756,8 +773,8 @@ def ceph_command(zkhandler, logger, this_node, data, d_osd):
 
     # Removing an OSD
     elif command == "osd_remove":
-        osd_id = args[0]
+        osd_id, force = args.split(",")
-        force_flag = bool(strtobool(args[1]))
+        force_flag = bool(strtobool(force))
 
         # Verify osd_id is in the list
         if d_osd[osd_id] and d_osd[osd_id].node == this_node.name: