<fix>[baremetal_v2_gateway_agent]: improve local deploy and bm2-agent

1. replace smartctl to filter /dev/disk/by-id for getting diskinfo
2. pass back ks err msg to gateway
3. improve bm2-instance-agent build scripts

Resolves: ZSTAC-45956

Change-Id: I6962787467706d767378796e666f636577677963

Author: unrollable

bm-instance-agent/bm_instance_agent/__init__.py

@@ -1 +1 @@
-__version__ = '4.3.18'
+__version__ = "2.0.0"

bm-instance-agent/bm_instance_agent/api/utils.py

@@ -1,7 +1,6 @@
 from concurrent import futures
 import json
 import sys
-import time
 
 from oslo_concurrency import processutils
 from oslo_log import log as logging
@@ -11,6 +10,7 @@
 
 from bm_instance_agent.conf import CONF
 from bm_instance_agent.objects import HeaderObj
+from bm_instance_agent.common.utils import transcantion
 
 
 THREAD_POOL = None
@@ -123,35 +123,3 @@ def format_exception():
     del exception_info
 
     return data
-
-
-class transcantion(object):
-    """ A tool class for retry
-    """
-
-    def __init__(self, retries, sleep_time=0):
-        self.retries = retries
-        self.sleep_time = sleep_time
-
-    def __enter__(self):
-        return self
-
-    def __exit__(self, exc_type, exc_val, exc_tb):
-        if exc_type is not None:
-            return False
-        return True
-
-    def execute(self, func, *args, **kwargs):
-        err = None
-        for i in range(self.retries):
-            try:
-                if i > 0:
-                    msg = 'Attempt rerun {name}: {i}'.format(
-                        name=func.__name__, i=i)
-                    LOG.warning(msg)
-                return func(*args, **kwargs)
-            except Exception as e:
-                LOG.exception(e)
-                err = e
-            time.sleep(self.sleep_time)
-        raise err

bm-instance-agent/bm_instance_agent/common/utils.py

@@ -7,11 +7,46 @@
 from netaddr import IPAddress
 import netifaces
 from oslo_concurrency import processutils
+from oslo_log import log as logging
 import psutil
 import pyroute2
+import time
 
 from bm_instance_agent import exception
 
+LOG = logging.getLogger(__name__)
+
+class transcantion(object):
+    """ A tool class for retry
+    """
+
+    def __init__(self, retries, sleep_time=0):
+        self.retries = retries
+        self.sleep_time = sleep_time
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        if exc_type is not None:
+            return False
+        return True
+
+    def execute(self, func, *args, **kwargs):
+        err = None
+        for i in range(self.retries):
+            try:
+                if i > 0:
+                    msg = 'Attempt rerun {name}: {i}'.format(
+                        name=func.__name__, i=i)
+                    LOG.warning(msg)
+                return func(*args, **kwargs)
+            except Exception as e:
+                LOG.exception(e)
+                err = e
+            time.sleep(self.sleep_time)
+        raise err
+
 
 class IpLink:
     def __init__(self, chunk):

bm-instance-agent/bm_instance_agent/manager.py

@@ -3,6 +3,7 @@
 from oslo_log import log as logging
 from stevedore import driver
 
+from __init__ import __version__
 from bm_instance_agent.common import utils as bm_utils
 from bm_instance_agent import exception
 from bm_instance_agent.objects import BmInstanceObj
@@ -16,7 +17,6 @@
 BM_INSTANCE_UUID = None
 DRIVER = None
 ZWATCH_AGENT_CONF_PATH = "/usr/local/zstack/zwatch-vm-agent/conf.yaml"
-VERSION = '2.0.0'
 
 
 class AgentManager(object):
@@ -73,7 +73,7 @@ def ping(self, bm_instance):
         self.driver.ping(instance_obj)
         self.driver.discovery_target(instance_obj)
         self._check_gateway_ip(instance_obj)
-        return {'version': VERSION, 'ping': {'bmInstanceUuid': BM_INSTANCE_UUID}}
+        return {'version': __version__, 'ping': {'bmInstanceUuid': BM_INSTANCE_UUID}}
 
     def reboot(self, bm_instance):
         instance_obj = BmInstanceObj.from_json(bm_instance)

bm-instance-agent/bm_instance_agent/systems/linux/driver.py

@@ -94,14 +94,26 @@ def attach_volume(self, instance_obj, volume_obj):
         conf_path = '/etc/iscsi/initiatorname.iscsi'
         initiator = ('InitiatorName=iqn.2015-01.io.zstack:initiator.'
                      'instance.{uuid}').format(uuid=instance_obj.uuid)
-        with open(conf_path, 'w') as f:
-            f.write(initiator)
 
-        cmd = 'systemctl daemon-reload && systemctl restart iscsid'
-        processutils.execute(cmd, shell=True)
+        update_conf = False
+        if os.path.exists(conf_path):
+            with open(conf_path, 'r') as f:
+                if f.read().strip() != initiator:
+                    update_conf = True
+        else:
+            update_conf = True
+
+        if update_conf:
+            with open(conf_path, 'w') as f:
+                f.write(initiator)
+
+            cmd = 'systemctl daemon-reload && systemctl restart iscsid'
+            processutils.execute(cmd, shell=True)
 
         cmd = ['iscsiadm', '-m', 'session', '--rescan']
-        processutils.execute(*cmd)
+        # parameter[delay_on_retry] of func[processutils.execute] will not verify exit_code
+        with bm_utils.transcantion(retries=5, sleep_time=10) as cursor:
+            cursor.execute(processutils.execute, *cmd)
 
     def detach_volume(self, instance_obj, volume_obj):
         """ Detach a given iSCSI lun

bm-instance-agent/tools/package.sh

@@ -1,14 +1,14 @@
 #!/bin/bash
 
+version=`awk -F' = ' '{gsub(/"/,"",$2);print $2}' bm_instance_agent/__init__.py`
 bin_name='bm-instance-agent.tar.gz'
+agent_name=zstack-bm-agent-`uname -m`-${version}.bin
 
 # Get the shell scirpt's dir
 shell_dir="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"
 pushd ${shell_dir}
 
-# Get the git repo's root dir
-project_dir=`git rev-parse --show-toplevel`/bm-instance-agent
-pushd ${project_dir}
+pushd ${shell_dir}/../
 
 temp=`mktemp -d`
 pex . -v \
@@ -28,7 +28,7 @@ popd
 popd
 popd
 
-cat ./tools/install.sh ${temp}/${bin_name} > ./install.bin
-sed -i "s/MD5_SUM/${md5}/g" ./install.bin
-chmod +x ./install.bin
+cat ./tools/install.sh ${temp}/${bin_name} > ${agent_name}
+sed -i "s/MD5_SUM/${md5}/g" ${agent_name}
+chmod +x ${agent_name}
 rm -rf ${temp}

bm-instance-agent/tools/zwatch-vm-agent-aarch64

@@ -72,6 +72,7 @@ class BaremetalV2GatewayAgentPlugin(kvmagent.KvmAgent):
     DEFAULT_PXE_PATH = os.path.join(PXELINUX_CFG_DIR, 'default')
     GRUB_CFG_DIR = os.path.join(TFTPBOOT_DIR, 'EFI/centos/')
     GRUB_CFG_PATH = os.path.join(GRUB_CFG_DIR, 'grub.cfg')
+    KS_ERR_LOG_PATH = os.path.join(TFTPBOOT_DIR, 'ks.errlogs/')
     X86_64_BOOTIMG_DIR = os.path.join(TFTPBOOT_DIR, 'x86_64')
     AARCH64_BOOTIMG_DIR = os.path.join(TFTPBOOT_DIR, 'aarch64')
 
@@ -179,6 +180,9 @@ def _ensure_env(self, network_obj):
         if not os.path.exists(self.BM_IMGS_DIR):
             linux.mkdir(self.BM_IMGS_DIR)
 
+        if not os.path.exists(self.KS_ERR_LOG_PATH):
+            linux.mkdir(self.KS_ERR_LOG_PATH)
+
         # download pxe images from management node
         # static repo url like: http://10.10.0.1:8080/zstack/static/zstack-repo/x86_64/c76
         bmtempdir = tempfile.mkdtemp()
@@ -283,17 +287,16 @@ def _prepare_tftp(self):
             with open(self.MAPFILE_PATH, 'w') as f:
                 f.write(map_file)
 
-        if not os.path.exists(self.TFTPD_SYSTEMD_SERVICE_PATH):
-            template_systemd_service = self._load_template(
-                'tftp_systemd_service')
-            systemd_service = template_systemd_service.render(
-                bm_gateway_tftpd_mapfile=self.MAPFILE_PATH,
-                bm_gateway_tftpboot_dir=self.TFTPBOOT_DIR)
-            with open(self.TFTPD_SYSTEMD_SERVICE_PATH, 'w') as f:
-                f.write(systemd_service)
+        template_systemd_service = self._load_template(
+            'tftp_systemd_service')
+        systemd_service = template_systemd_service.render(
+            bm_gateway_tftpd_mapfile=self.MAPFILE_PATH,
+            bm_gateway_tftpboot_dir=self.TFTPBOOT_DIR)
+        with open(self.TFTPD_SYSTEMD_SERVICE_PATH, 'w') as f:
+            f.write(systemd_service)
 
         cmd = ('systemctl daemon-reload && '
-               'systemctl start zstack-baremetal-tftpd')
+               'systemctl restart zstack-baremetal-tftpd')
         shell.call(cmd)
 
     def _destroy_tftp(self):
@@ -916,7 +919,8 @@ def _create_import_data_config(self, volume_driver, cmd):
             chassis_address=cmd.chassisInfo.address,
             chassis_port=cmd.chassisInfo.port,
             api_id=cmd.threadContext.api,
-            task_name=cmd.threadContext["task-name"]
+            task_name=cmd.threadContext["task-name"],
+            provision_mac=instance_obj.provision_mac
         )
 
         with open(ks_config_path, 'w') as f:

kvmagent/kvmagent/plugins/baremetal_v2_gateway_agent.py

@@ -19,6 +19,9 @@ import multiprocessing
 import os
 import sys
 import time
+import traceback
+
+provision_mac = '{{provision_mac}}'
 
 def shell_cmd(cmd, exception=True, workdir=None):
     process = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE, stdin=subprocess.PIPE,
@@ -63,7 +66,7 @@ def get_src_dev():
     time.sleep(1)
     cmd = "iscsiadm --mode node --targetname {{iqn_name}} -p {{gateway_ip}}:3260 --login"
     shell_cmd(cmd)
-    time.sleep(1)
+    time.sleep(3)
     cmd = "systemctl daemon-reload && systemctl restart iscsid"
     shell_cmd(cmd)
     time.sleep(1)
@@ -145,9 +148,13 @@ def main():
 
 try:
     main()
-except Exception as e:
-    shell_cmd("touch /root/ks_error_log")
-    shell_cmd("echo %s >> /root/ks_error_log" % str(e))
+except:
+    # pass back the ks err to tftp server on gateway
+    err_file_name = '/root/ks.{}'.format(provision_mac.replace(':', '-'))
+    traceback.print_exc(file=open(err_file_name,'w+'))
+    shell_cmd("curl -T {} tftp://{}/ks.errlogs/".format(err_file_name, "{{gateway_ip}}"))
+
+    # report fail and update ipmi info
     convert_info = {}
     convert_info['status'] = 'ConvertFailed'
     convert_info['progress'] = 0

kvmagent/kvmagent/plugins/bmv2_gateway_agent/templates/import-data.ks.j2

@@ -19,8 +19,10 @@ import multiprocessing
 import os
 import sys
 import platform
+import traceback
 
 PROVISION_NET = "{{ provision_net }}"
+pxe_iface_mac = ''
 
 units_mapping = {
     'kb': 1024,
@@ -123,7 +125,6 @@ def get_nic_info():
     # Get the pxe interface from /proc/cmdline
     # NOTE: Need to point that the mac addr should start with '01', because
     # the arp type of ethernet is 1.
-    pxe_iface_mac = ''
     with open('/proc/cmdline', 'r') as f:
         for param in f.read().strip().split():
             if 'BOOTIF' in param:
@@ -247,7 +248,13 @@ def main():
     shell_cmd("poweroff")
 
 
-main()
+try:
+    main()
+except:
+    # pass back the ks err to tftp server on gateway
+    err_file_name = '/root/ks.{}'.format(pxe_iface_mac.replace(':','-'))
+    traceback.print_exc(file=open(err_file_name,'w+'))
+    shell_cmd("curl -T {} tftp://{}/ks.errlogs/".format(err_file_name, PROVISION_NET))
 
 
 %end