sbcli-dev 3.8.67__zip → 3.8.70__zip

{sbcli_dev-3.8.67 → sbcli_dev-3.8.70}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sbcli-dev
-Version: 3.8.67
+Version: 3.8.70
 Summary: CLI for managing SimplyBlock cluster
 Home-page: https://www.simplyblock.io/
 Author: Hamdy

{sbcli_dev-3.8.67 → sbcli_dev-3.8.70}/env_var

@@ -1,5 +1,5 @@
 SIMPLY_BLOCK_COMMAND_NAME=sbcli-dev
-SIMPLY_BLOCK_VERSION=3.8.67
+SIMPLY_BLOCK_VERSION=3.8.70
 
 
 SIMPLY_BLOCK_DOCKER_IMAGE=simplyblock/simplyblock:main

{sbcli_dev-3.8.67 → sbcli_dev-3.8.70}/sbcli_dev.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sbcli-dev
-Version: 3.8.67
+Version: 3.8.70
 Summary: CLI for managing SimplyBlock cluster
 Home-page: https://www.simplyblock.io/
 Author: Hamdy

{sbcli_dev-3.8.67 → sbcli_dev-3.8.70}/simplyblock_core/controllers/device_controller.py

@@ -252,19 +252,20 @@ def restart_device(device_id, force=False):
 
     # add to jm raid
     if snode.jm_device and snode.jm_device.raid_bdev:
-        nvme_controller = device_obj.nvme_controller
-        jm_part = None
-        for part in snode.jm_device.jm_nvme_bdev_list:
-            if part.startswith(nvme_controller):
-                jm_part = part
-                break
-
-        if not jm_part:
-            if snode.jm_device.status == NVMeDevice.STATUS_ONLINE:
+        # looking for jm partition
+        rpc_client = RPCClient(snode.mgmt_ip, snode.rpc_port, snode.rpc_username, snode.rpc_password)
+        jm_dev_part = f"{dev.nvme_bdev[:-1]}1"
+        ret = rpc_client.get_bdevs(jm_dev_part)
+        if ret:
+            logger.info(f"JM part found: {jm_dev_part}")
+            if snode.jm_device.status == JMDevice.STATUS_UNAVAILABLE:
+                restart_jm_device(snode.jm_device.get_id(), force=True)
+
+            if snode.jm_device.status == JMDevice.STATUS_ONLINE and \
+                    jm_dev_part not in snode.jm_device.jm_nvme_bdev_list:
                 remove_jm_device(snode.jm_device.get_id(), force=True)
                 time.sleep(3)
-
-            restart_jm_device(snode.jm_device.get_id(), force=True)
+                restart_jm_device(snode.jm_device.get_id(), force=True)
 
     return "Done"
 
@@ -771,8 +772,8 @@ def remove_jm_device(device_id, force=False):
 
     ret = rpc_client.bdev_alceml_delete(snode.jm_device.alceml_bdev)
 
-    if snode.jm_device.testing_bdev:
-        ret = rpc_client.bdev_passtest_delete(snode.jm_device.testing_bdev)
+    # if snode.jm_device.testing_bdev:
+    #     ret = rpc_client.bdev_passtest_delete(snode.jm_device.testing_bdev)
 
     if len(snode.jm_device.jm_nvme_bdev_list) == 2:
         ret = rpc_client.bdev_raid_delete(snode.jm_device.raid_bdev)

{sbcli_dev-3.8.67 → sbcli_dev-3.8.70}/simplyblock_core/controllers/health_controller.py

@@ -66,7 +66,7 @@ def _check_node_docker_api(ip):
     # return False
 
 
-def _check_node_rpc(rpc_ip, rpc_port, rpc_username, rpc_password, timeout=60, retry=3):
+def _check_node_rpc(rpc_ip, rpc_port, rpc_username, rpc_password, timeout=5, retry=2):
     try:
         rpc_client = RPCClient(
             rpc_ip, rpc_port, rpc_username, rpc_password,

{sbcli_dev-3.8.67 → sbcli_dev-3.8.70}/simplyblock_core/services/storage_node_monitor.py

@@ -144,31 +144,29 @@ while True:
             spdk_process = health_controller._check_spdk_process_up(snode.mgmt_ip)
             logger.info(f"Check: spdk process {snode.mgmt_ip}:5000 ... {spdk_process}")
 
-            # node_rpc_check = True
-            # 3- check node RPC
-            node_rpc_check = health_controller._check_node_rpc(
-                snode.mgmt_ip, snode.rpc_port, snode.rpc_username, snode.rpc_password, timeout=5, retry=1)
-            logger.info(f"Check: node RPC {snode.mgmt_ip}:{snode.rpc_port} ... {node_rpc_check}")
-
-            # check JM device
-            if snode.jm_device:
-                if snode.jm_device.status in [JMDevice.STATUS_ONLINE, JMDevice.STATUS_UNAVAILABLE]:
-                    ret = health_controller.check_jm_device(snode.jm_device.get_id())
-                    if ret:
-                        logger.info(f"JM bdev is online: {snode.jm_device.get_id()}")
-                        if snode.jm_device.status != JMDevice.STATUS_ONLINE:
-                            device_controller.set_jm_device_state(snode.jm_device.get_id(), JMDevice.STATUS_ONLINE)
-                    else:
-                        logger.error(f"JM bdev is offline: {snode.jm_device.get_id()}")
-                        if snode.jm_device.status != JMDevice.STATUS_UNAVAILABLE:
-                            device_controller.set_jm_device_state(snode.jm_device.get_id(),
-                                                                  JMDevice.STATUS_UNAVAILABLE)
-
-            is_node_online = ping_check and node_api_check and node_rpc_check and spdk_process
+            is_node_online = ping_check and node_api_check and spdk_process
             if is_node_online:
                 set_node_online(snode)
+
+                # check JM device
+                if snode.jm_device:
+                    if snode.jm_device.status in [JMDevice.STATUS_ONLINE, JMDevice.STATUS_UNAVAILABLE]:
+                        ret = health_controller.check_jm_device(snode.jm_device.get_id())
+                        if ret:
+                            logger.info(f"JM bdev is online: {snode.jm_device.get_id()}")
+                            if snode.jm_device.status != JMDevice.STATUS_ONLINE:
+                                device_controller.set_jm_device_state(snode.jm_device.get_id(), JMDevice.STATUS_ONLINE)
+                        else:
+                            logger.error(f"JM bdev is offline: {snode.jm_device.get_id()}")
+                            if snode.jm_device.status != JMDevice.STATUS_UNAVAILABLE:
+                                device_controller.set_jm_device_state(snode.jm_device.get_id(),
+                                                                      JMDevice.STATUS_UNAVAILABLE)
+
             else:
                 set_node_offline(snode)
+                if snode.jm_device.status != JMDevice.STATUS_UNAVAILABLE:
+                    device_controller.set_jm_device_state(snode.jm_device.get_id(), JMDevice.STATUS_UNAVAILABLE)
+
                 if not ping_check and not node_api_check and not spdk_process:
                     # restart on new node
                     storage_node_ops.set_node_status(snode.get_id(), StorageNode.STATUS_SCHEDULABLE)

{sbcli_dev-3.8.67 → sbcli_dev-3.8.70}/simplyblock_core/storage_node_ops.py

@@ -256,14 +256,14 @@ def _create_jm_stack_on_raid(rpc_client, jm_nvme_bdevs, snode, after_restart):
     alceml_name = f"alceml_jm_{snode.get_id()}"
 
     nvme_bdev = raid_bdev
-    test_name = ""
-    if snode.enable_test_device:
-        test_name = f"{raid_bdev}_test"
-        ret = rpc_client.bdev_passtest_create(test_name, raid_bdev)
-        if not ret:
-            logger.error(f"Failed to create passtest bdev {test_name}")
-            return False
-        nvme_bdev = test_name
+    # test_name = ""
+    # if snode.enable_test_device:
+    #     test_name = f"{raid_bdev}_test"
+    #     ret = rpc_client.bdev_passtest_create(test_name, raid_bdev)
+    #     if not ret:
+    #         logger.error(f"Failed to create passtest bdev {test_name}")
+    #         return False
+    #     nvme_bdev = test_name
     pba_init_mode = 3
     if after_restart:
         pba_init_mode = 1

{sbcli_dev-3.8.67 → sbcli_dev-3.8.70}/simplyblock_web/blueprints/snode_ops_k8s.py

@@ -28,7 +28,7 @@ node_name = os.environ.get("HOSTNAME")
 deployment_name = f"snode-spdk-deployment-{node_name}"
 default_namespace = 'default'
 namespace_id_file = '/etc/simplyblock/namespace'
-pod_name = 'snode-spdk-deployment'
+pod_name = deployment_name[:50]
 
 
 config.load_incluster_config()