From 86eaf594e60c43c1c39b5996dd65bca410d5a6ea Mon Sep 17 00:00:00 2001
From: xuweibj <xwbj@cn.ibm.com>
Date: Thu, 30 Aug 2018 02:45:40 -0400
Subject: [PATCH] retry in rpower reset when get 'Connection aborted'

---
 .../agent/hwctl/executor/openbmc_power.py     | 27 +++++++++++++------
 .../lib/python/agent/hwctl/openbmc_client.py  |  2 +-
 2 files changed, 20 insertions(+), 9 deletions(-)

diff --git a/xCAT-openbmc-py/lib/python/agent/hwctl/executor/openbmc_power.py b/xCAT-openbmc-py/lib/python/agent/hwctl/executor/openbmc_power.py
index 819fbc887..b6bc0b4e4 100644
--- a/xCAT-openbmc-py/lib/python/agent/hwctl/executor/openbmc_power.py
+++ b/xCAT-openbmc-py/lib/python/agent/hwctl/executor/openbmc_power.py
@@ -109,12 +109,15 @@ class OpenBMCPowerTask(ParallelNodesCommand):
                     off_flag = False
                     start_timeStamp = int(time.time())
                     for i in range (0, 30):
-                        states = obmc.list_power_states()
-                        status = obmc.get_host_state(states)
-                        if openbmc.RPOWER_STATES.get(status) == 'off':
-                            off_flag = True
-                            break
                         gevent.sleep( 2 )
+                        try:
+                            states = obmc.list_power_states()
+                            status = obmc.get_host_state(states)
+                            if openbmc.RPOWER_STATES.get(status) == 'off':
+                                off_flag = True
+                                break
+                        except SelfServerException as e:
+                            continue
 
                     end_timeStamp = int(time.time())
 
@@ -123,10 +126,18 @@ class OpenBMCPowerTask(ParallelNodesCommand):
                                  'to off after waiting %s seconds. (State= %s).' % (end_timeStamp - start_timeStamp, status)
                         raise SelfServerException(error)
 
-                ret = obmc.set_power_state('on')
-                self.callback.update_node_attributes('status', node, POWER_STATE_DB['on'])
+                for i in range(0, 2):
+                    try:
+                        ret = obmc.set_power_state('on')
+                        self.callback.update_node_attributes('status', node, POWER_STATE_DB['on'])
+                        self.callback.info('%s: %s'  % (node, 'reset'))
+                        return
+                    except SelfServerException as e:
+                        self.callback.syslog('%s: %s' % (node, e.message))
+                        gevent.sleep( 1 )
+                        continue
 
-                self.callback.info('%s: %s'  % (node, 'reset'))
+                self.callback.error(e.message, node)
 
         except (SelfServerException, SelfClientException) as e:
             self.callback.error(e.message, node)
diff --git a/xCAT-openbmc-py/lib/python/agent/hwctl/openbmc_client.py b/xCAT-openbmc-py/lib/python/agent/hwctl/openbmc_client.py
index 1670a51da..32f081d06 100644
--- a/xCAT-openbmc-py/lib/python/agent/hwctl/openbmc_client.py
+++ b/xCAT-openbmc-py/lib/python/agent/hwctl/openbmc_client.py
@@ -345,7 +345,7 @@ class OpenBMCRest(object):
                 e.message = "Login to BMC failed: Can't connect to {0} {1}.".format(e.host_and_port, e.detail_msg)
             else:
                 e.message = 'BMC did not respond. ' \
-                            'Validate BMC configuration and retry the command.'
+                            'Validate BMC configuration and retry the command. ' + e.detail_msg
             self._print_error_log(e.message, cmd)
             raise
         except ValueError: