Improve handling of rounded values

With the OSD LVM information stored in Zookeeper, we can use this to
determine the actual block device to zap rather than relying on runtime
determination and guestimation.

client-cli/pvc/cli_lib/ceph.py

@@ -370,13 +370,22 @@ def format_list_osd(osd_list):
             # If this happens, the node hasn't checked in fully yet, so use some dummy data
             if osd_information["stats"]["node"] == "|":
                 for key in osd_information["stats"].keys():
-                    if osd_information["stats"][key] == "|":
-                        osd_information["stats"][key] = "N/A"
-                    elif osd_information["stats"][key] is None:
+                    if (
+                        osd_information["stats"][key] == "|"
+                        or osd_information["stats"][key] is None
+                    ):
                         osd_information["stats"][key] = "N/A"
                 for key in osd_information.keys():
                     if osd_information[key] is None:
                         osd_information[key] = "N/A"
+            else:
+                for key in osd_information["stats"].keys():
+                    if key in ["utilization", "var"] and isinstance(
+                        osd_information["stats"][key], float
+                    ):
+                        osd_information["stats"][key] = round(
+                            osd_information["stats"][key], 2
+                        )
         except KeyError:
             print(
                 f"Details for OSD {osd_information['id']} missing required keys, skipping."
@@ -449,13 +458,11 @@ def format_list_osd(osd_list):
         if _osd_free_length > osd_free_length:
             osd_free_length = _osd_free_length
 
-        osd_util = round(osd_information["stats"]["utilization"], 2)
-        _osd_util_length = len(str(osd_util)) + 1
+        _osd_util_length = len(str(osd_information["stats"]["utilization"])) + 1
         if _osd_util_length > osd_util_length:
             osd_util_length = _osd_util_length
 
-        osd_var = round(osd_information["stats"]["var"], 2)
-        _osd_var_length = len(str(osd_var)) + 1
+        _osd_var_length = len(str(osd_information["stats"]["var"])) + 1
         if _osd_var_length > osd_var_length:
             osd_var_length = _osd_var_length
 
@@ -605,8 +612,6 @@ def format_list_osd(osd_list):
         osd_up_flag, osd_up_colour, osd_in_flag, osd_in_colour = getOutputColoursOSD(
             osd_information
         )
-        osd_util = round(osd_information["stats"]["utilization"], 2)
-        osd_var = round(osd_information["stats"]["var"], 2)
 
         osd_db_device = osd_information["db_device"]
         if not osd_db_device:
@@ -669,8 +674,8 @@ def format_list_osd(osd_list):
                 osd_reweight=osd_information["stats"]["reweight"],
                 osd_used=osd_information["stats"]["used"],
                 osd_free=osd_information["stats"]["avail"],
-                osd_util=osd_util,
-                osd_var=osd_var,
+                osd_util=osd_information["stats"]["utilization"],
+                osd_var=osd_information["stats"]["var"],
                 osd_wrops=osd_information["stats"]["wr_ops"],
                 osd_wrdata=osd_information["stats"]["wr_data"],
                 osd_rdops=osd_information["stats"]["rd_ops"],

node-daemon/pvcnoded/objects/CephInstance.py

@@ -434,7 +434,7 @@ class CephOSDInstance(object):
                     (("osd.lv", osd_id), osd_lv),
                     (
                         ("osd.stats", osd_id),
-                        f'{"uuid": "|", "up": 0, "in": 0, "primary_affinity": "|", "utilization": "|", "var": "|", "pgs": "|", "kb": "|", "weight": "|", "reweight": "|", "node": "{node}", "used": "|", "avail": "|", "wr_ops": "|", "wr_data": "|", "rd_ops": "|", "rd_data": "|", state="|" }',
+                        '{"uuid": "|", "up": 0, "in": 0, "primary_affinity": "|", "utilization": "|", "var": "|", "pgs": "|", "kb": "|", "weight": "|", "reweight": "|", "node": "|", "used": "|", "avail": "|", "wr_ops": "|", "wr_data": "|", "rd_ops": "|", "rd_data": "|", "state": "|"}',
                     ),
                 ]
             )
@@ -542,47 +542,48 @@ class CephOSDInstance(object):
                     break
 
             # 4. Determine the block devices
-            device_zk = zkhandler.read(("osd.device", osd_id))
-            try:
-                retcode, stdout, stderr = common.run_os_command(
-                    "readlink /var/lib/ceph/osd/ceph-{}/block".format(osd_id)
-                )
-                vg_name = stdout.split("/")[
-                    -2
-                ]  # 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
+            osd_vg = zkhandler.read(("osd.vg", osd_id))
+            osd_lv = zkhandler.read(("osd.lv", osd_id))
+            osd_lvm = f"/dev/{osd_vg}/{osd_lv}"
+            osd_device = None
 
-            # 5a. Verify that the blockdev actually has a ceph volume that matches the ID, otherwise don't zap it
             logger.out(
-                f"Check OSD disk {pv_block} for OSD signature with ID osd.{osd_id}",
+                f"Getting disk info for OSD {osd_id} LV {osd_lvm}",
                 state="i",
             )
             retcode, stdout, stderr = common.run_os_command(
-                f"ceph-volume lvm list {pv_block}"
+                f"ceph-volume lvm list {osd_lvm}"
             )
-            if f"====== osd.{osd_id} =======" in stdout:
-                # 5b. Zap the volumes
-                logger.out(
-                    "Zapping OSD disk with ID {} on {}".format(osd_id, pv_block),
-                    state="i",
-                )
-                retcode, stdout, stderr = common.run_os_command(
-                    "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
+            for line in stdout.split("\n"):
+                if "devices" in line:
+                    osd_device = line.split()[-1]
+
+            if not osd_device:
+                print("ceph-volume lvm list")
+                print("Could not find OSD information in data:")
+                print(stdout)
+                print(stderr)
+                if force_flag:
+                    logger.out("Ignoring error due to force flag", state="i")
+                else:
+                    raise Exception
+
+            # 5. Zap the volumes
+            logger.out(
+                "Zapping OSD {} disk on {}".format(osd_id, osd_device),
+                state="i",
+            )
+            retcode, stdout, stderr = common.run_os_command(
+                "ceph-volume lvm zap --destroy {}".format(osd_device)
+            )
+            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")