Add warnings about snapshot consistency

We supported creating snapshots, but not doing anything with them. This
removes the manual task of restoring a snapshot and replace it with a
PVC abstraction of rolling back to a snapshot.

While Ceph recommends cloning a snapshot instead of rolling back, due to
the time taken, in our usecase I don't think that is an optimal
strategy, as it will leave dangling clones that we'd then have to
manage.

Closes #183

api-daemon/pvcapid/flaskapi.py

@@ -6362,6 +6362,59 @@ api.add_resource(
 )
 
 
+# /storage/ceph/snapshot/<pool>/<volume>/<snapshot>/rollback
+class API_Storage_Ceph_Snapshot_Rollback_Element(Resource):
+    @Authenticator
+    def post(self, pool, volume, snapshot):
+        """
+        Roll back an RBD volume {volume} in pool {pool} to snapshot {snapshot}
+
+        WARNING: This action cannot be done on an active RBD volume. All IO MUST be stopped first.
+        ---
+        tags:
+          - storage / ceph
+        parameters:
+          - in: query
+            name: snapshot
+            type: string
+            required: true
+            description: The name of the snapshot
+          - in: query
+            name: volume
+            type: string
+            required: true
+            description: The name of the volume
+          - in: query
+            name: pool
+            type: integer
+            required: true
+            description: The name of the pool
+        responses:
+          200:
+            description: OK
+            schema:
+              type: object
+              id: Message
+          404:
+            description: Not found
+            schema:
+              type: object
+              id: Message
+          400:
+            description: Bad request
+            schema:
+              type: object
+              id: Message
+        """
+        return api_helper.ceph_volume_snapshot_rollback(pool, volume, snapshot)
+
+
+api.add_resource(
+    API_Storage_Ceph_Snapshot_Rollback_Element,
+    "/storage/ceph/snapshot/<pool>/<volume>/<snapshot>/rollback",
+)
+
+
 ##########################################################
 # Provisioner API
 ##########################################################

api-daemon/pvcapid/helper.py

@@ -2183,6 +2183,22 @@ def ceph_volume_snapshot_rename(zkhandler, pool, volume, name, new_name):
     return output, retcode
 
 
+@ZKConnection(config)
+def ceph_volume_snapshot_rollback(zkhandler, pool, volume, name):
+    """
+    Roll back a Ceph RBD volume to a given snapshot in the PVC Ceph storage cluster.
+    """
+    retflag, retdata = pvc_ceph.rollback_snapshot(zkhandler, pool, volume, name)
+
+    if retflag:
+        retcode = 200
+    else:
+        retcode = 400
+
+    output = {"message": retdata.replace('"', "'")}
+    return output, retcode
+
+
 @ZKConnection(config)
 def ceph_volume_snapshot_remove(zkhandler, pool, volume, name):
     """

client-cli/pvc/cli/cli.py

@@ -4325,6 +4325,10 @@ def cli_storage_volume_snapshot():
 def cli_storage_volume_snapshot_add(pool, volume, name):
     """
     Add a snapshot with name NAME of Ceph RBD volume VOLUME in pool POOL.
+
+    WARNING: RBD snapshots are crash-consistent but not filesystem-aware. If a snapshot was taken
+    of a running VM, restoring that snapshot will be equivalent to having forcibly restarted the
+    VM at the moment of the snapshot.
     """
 
     retcode, retmsg = pvc.lib.storage.ceph_snapshot_add(CLI_CONFIG, pool, volume, name)
@@ -4372,6 +4376,36 @@ def cli_storage_volume_snapshot_remove(pool, volume, name):
     finish(retcode, retmsg)
 
 
+###############################################################################
+# > pvc storage volume snapshot rollback
+###############################################################################
+@click.command(name="rollback", short_help="Roll back RBD volume to snapshot.")
+@connection_req
+@click.argument("pool")
+@click.argument("volume")
+@click.argument("name")
+@confirm_opt("Roll back to snapshot {name} for volume {pool}/{volume}")
+def cli_storage_volume_snapshot_rollback(pool, volume, name):
+    """
+    Roll back the Ceph RBD volume VOLUME in pool POOL to the snapshot NAME.
+
+    DANGER: All data written to the volume since the given snapshot will be permanently lost.
+
+    WARNING: A rollback cannot be performed on an RBD volume with active I/O. Doing so will cause
+    undefined behaviour and possible corruption. Ensure that any VM(s) using this RBD volume are
+    stopped or disabled before attempting a snapshot rollback.
+
+    WARNING: RBD snapshots are crash-consistent but not filesystem-aware. If a snapshot was taken
+    of a running VM, restoring that snapshot will be equivalent to having forcibly restarted the
+    VM at the moment of the snapshot.
+    """
+
+    retcode, retmsg = pvc.lib.storage.ceph_snapshot_rollback(
+        CLI_CONFIG, pool, volume, name
+    )
+    finish(retcode, retmsg)
+
+
 ###############################################################################
 # > pvc storage volume snapshot list
 ###############################################################################
@@ -6349,6 +6383,7 @@ cli_storage_volume.add_command(cli_storage_volume_list)
 cli_storage_volume_snapshot.add_command(cli_storage_volume_snapshot_add)
 cli_storage_volume_snapshot.add_command(cli_storage_volume_snapshot_rename)
 cli_storage_volume_snapshot.add_command(cli_storage_volume_snapshot_remove)
+cli_storage_volume_snapshot.add_command(cli_storage_volume_snapshot_rollback)
 cli_storage_volume_snapshot.add_command(cli_storage_volume_snapshot_list)
 cli_storage_volume.add_command(cli_storage_volume_snapshot)
 cli_storage.add_command(cli_storage_volume)

client-cli/pvc/lib/common.py

@@ -108,9 +108,10 @@ class UploadProgressBar(object):
 
 
 class ErrorResponse(requests.Response):
-    def __init__(self, json_data, status_code):
+    def __init__(self, json_data, status_code, headers):
         self.json_data = json_data
         self.status_code = status_code
+        self.headers = headers
 
     def json(self):
         return self.json_data
@@ -206,7 +207,7 @@ def call_api(
     except Exception as e:
         message = "Failed to connect to the API: {}".format(e)
         code = response.status_code if response else 504
-        response = ErrorResponse({"message": message}, code)
+        response = ErrorResponse({"message": message}, code, None)
 
     # Display debug output
     if config["debug"]:

client-cli/pvc/lib/storage.py

@@ -1544,6 +1544,30 @@ def ceph_snapshot_add(config, pool, volume, snapshot):
     return retstatus, response.json().get("message", "")
 
 
+def ceph_snapshot_rollback(config, pool, volume, snapshot):
+    """
+    Roll back Ceph volume to snapshot
+
+    API endpoint: POST /api/v1/storage/ceph/snapshot/{pool}/{volume}/{snapshot}/rollback
+    API arguments:
+    API schema: {"message":"{data}"}
+    """
+    response = call_api(
+        config,
+        "post",
+        "/storage/ceph/snapshot/{pool}/{volume}/{snapshot}/rollback".format(
+            snapshot=snapshot, volume=volume, pool=pool
+        ),
+    )
+
+    if response.status_code == 200:
+        retstatus = True
+    else:
+        retstatus = False
+
+    return retstatus, response.json().get("message", "")
+
+
 def ceph_snapshot_remove(config, pool, volume, snapshot):
     """
     Remove Ceph snapshot

daemon-common/ceph.py

@@ -1082,6 +1082,36 @@ def rename_snapshot(zkhandler, pool, volume, name, new_name):
     )
 
 
+def rollback_snapshot(zkhandler, pool, volume, name):
+    if not verifyVolume(zkhandler, pool, volume):
+        return False, 'ERROR: No volume with name "{}" is present in pool "{}".'.format(
+            volume, pool
+        )
+    if not verifySnapshot(zkhandler, pool, volume, name):
+        return (
+            False,
+            'ERROR: No snapshot with name "{}" is present for volume "{}" in pool "{}".'.format(
+                name, volume, pool
+            ),
+        )
+
+        # 1. Roll back the snapshot
+        retcode, stdout, stderr = common.run_os_command(
+            "rbd snap rollback {}/{}@{}".format(pool, volume, name)
+        )
+        if retcode:
+            return (
+                False,
+                'ERROR: Failed to roll back RBD volume "{}" in pool "{}" to snapshot "{}": {}'.format(
+                    volume, pool, name, stderr
+                ),
+            )
+
+    return True, 'Rolled back RBD volume "{}" in pool "{}" to snapshot "{}".'.format(
+        volume, pool, name
+    )
+
+
 def remove_snapshot(zkhandler, pool, volume, name):
     if not verifyVolume(zkhandler, pool, volume):
         return False, 'ERROR: No volume with name "{}" is present in pool "{}".'.format(

node-daemon/pvcnoded/util/fencing.py

@@ -253,12 +253,16 @@ def reboot_via_ipmi(node_name, ipmi_hostname, ipmi_user, ipmi_password, logger):
         state="i",
         prefix=f"fencing {node_name}",
     )
-    ipmi_status_retcode, ipmi_status_stdout, ipmi_status_stderr = common.run_os_command(
+    (
+        ipmi_intermediate_status_retcode,
+        ipmi_intermediate_status_stdout,
+        ipmi_intermediate_status_stderr,
+    ) = common.run_os_command(
         f"/usr/bin/ipmitool -I lanplus -H {ipmi_hostname} -U {ipmi_user} -P {ipmi_password} chassis power status"
     )
-    if ipmi_status_retcode == 0:
+    if ipmi_intermediate_status_retcode == 0:
         logger.out(
-            f"Current chassis power state is: {ipmi_status_stdout.strip()}",
+            f"Current chassis power state is: {ipmi_intermediate_status_stdout.strip()}",
             state="i",
             prefix=f"fencing {node_name}",
         )
@@ -299,12 +303,14 @@ def reboot_via_ipmi(node_name, ipmi_hostname, ipmi_user, ipmi_password, logger):
         state="i",
         prefix=f"fencing {node_name}",
     )
-    ipmi_status_retcode, ipmi_status_stdout, ipmi_status_stderr = common.run_os_command(
+    ipmi_final_status_retcode, ipmi_final_status_stdout, ipmi_final_status_stderr = (
+        common.run_os_command(
             f"/usr/bin/ipmitool -I lanplus -H {ipmi_hostname} -U {ipmi_user} -P {ipmi_password} chassis power status"
         )
+    )
 
-    if ipmi_stop_retcode == 0:
+    if ipmi_intermediate_status_stdout.strip() == "Chassis power is off":
-        if ipmi_status_stdout.strip() == "Chassis Power is on":
+        if ipmi_final_status_stdout.strip() == "Chassis Power is on":
             # We successfully rebooted the node and it is powered on; this is a succeessful fence
             logger.out(
                 "Successfully rebooted dead node; proceeding with fence recovery action",
@@ -312,7 +318,7 @@ def reboot_via_ipmi(node_name, ipmi_hostname, ipmi_user, ipmi_password, logger):
                 prefix=f"fencing {node_name}",
             )
             return True
-        elif ipmi_status_stdout.strip() == "Chassis Power is off":
+        elif ipmi_final_status_stdout.strip() == "Chassis Power is off":
             # We successfully rebooted the node but it is powered off; this might be expected or not, but the node is confirmed off so we can call it a successful fence
             logger.out(
                 "Chassis power is in confirmed off state after successfuly IPMI reboot; proceeding with fence recovery action",
@@ -323,13 +329,13 @@ def reboot_via_ipmi(node_name, ipmi_hostname, ipmi_user, ipmi_password, logger):
         else:
             # We successfully rebooted the node but it is in some unknown power state; since this might indicate a silent failure, we must call it a failed fence
             logger.out(
-                f"Chassis power is in an unknown state ({ipmi_status_stdout.strip()}) after successful IPMI reboot; NOT proceeding fence recovery action",
+                f"Chassis power is in an unknown state ({ipmi_final_status_stdout.strip()}) after successful IPMI reboot; NOT proceeding fence recovery action",
                 state="e",
                 prefix=f"fencing {node_name}",
             )
             return False
     else:
-        if ipmi_status_stdout.strip() == "Chassis Power is off":
+        if ipmi_final_status_stdout.strip() == "Chassis Power is off":
             # We failed to reboot the node but it is powered off; it has probably suffered a serious hardware failure, but the node is confirmed off so we can call it a successful fence
             logger.out(
                 "Chassis power is in confirmed off state after failed IPMI reboot; proceeding with fence recovery action",