Merge latest copy of azure master into ignw master

ansible/ansible.cfg

@@ -157,7 +157,7 @@ callback_whitelist = profile_tasks
 # current IP information.
 fact_caching = jsonfile
 fact_caching_connection = ~/.ansible/cache
-fact_caching_timeout = 1200
+fact_caching_timeout = 86400
 
 
 # retry files

ansible/group_vars/sonic/sku-sensors-data.yml

@@ -2279,3 +2279,99 @@ sensors_checks:
       temp: []
 
     psu_skips: {}
+
+  Arista-7260CX3-Q64:
+    alarms:
+      fan:
+      - pmbus-i2c-3-58/fan1/fan1_alarm
+      - pmbus-i2c-4-58/fan1/fan1_alarm
+      - pmbus-i2c-3-58/fan1/fan1_fault
+      - pmbus-i2c-4-58/fan1/fan1_fault
+      - la_cpld-i2c-85-60/fan1/fan1_fault
+      - la_cpld-i2c-85-60/fan2/fan2_fault
+      - la_cpld-i2c-85-60/fan3/fan3_fault
+      - la_cpld-i2c-85-60/fan4/fan4_fault
+      power:
+      - pmbus-i2c-3-58/iin/curr1_max_alarm
+      - pmbus-i2c-3-58/iout1/curr2_max_alarm
+      - pmbus-i2c-3-58/iout1/curr2_crit_alarm
+      - pmbus-i2c-3-58/iout2/curr3_crit_alarm
+      - pmbus-i2c-3-58/vin/in1_alarm
+      - pmbus-i2c-3-58/vout1/in2_lcrit_alarm
+      - pmbus-i2c-3-58/vout1/in2_crit_alarm
+      - pmbus-i2c-4-58/iin/curr1_max_alarm
+      - pmbus-i2c-4-58/iout1/curr2_max_alarm
+      - pmbus-i2c-4-58/iout1/curr2_crit_alarm
+      - pmbus-i2c-4-58/iout2/curr3_crit_alarm
+      - pmbus-i2c-4-58/vin/in1_alarm
+      - pmbus-i2c-4-58/vout1/in2_lcrit_alarm
+      - pmbus-i2c-4-58/vout1/in2_crit_alarm
+      temp:
+      - coretemp-isa-0000/Physical id 0/temp1_crit_alarm
+      - coretemp-isa-0000/Core 0/temp2_crit_alarm
+      - coretemp-isa-0000/Core 1/temp3_crit_alarm
+      - lm73-i2c-88-48/Front panel temp sensor/temp1_min_alarm
+      - lm73-i2c-88-48/Front panel temp sensor/temp1_max_alarm
+      - max6658-i2c-1-4c/Asic temp sensor/temp1_min_alarm
+      - max6658-i2c-1-4c/Asic temp sensor/temp1_max_alarm
+      - max6658-i2c-1-4c/Asic temp sensor/temp1_crit_alarm
+      - max6658-i2c-73-4c/Back panel temp sensor 1/temp1_min_alarm
+      - max6658-i2c-73-4c/Back panel temp sensor 1/temp1_max_alarm
+      - max6658-i2c-73-4c/Back panel temp sensor 1/temp1_crit_alarm
+      - max6658-i2c-73-4c/Back panel temp sensor 2/temp2_min_alarm
+      - max6658-i2c-73-4c/Back panel temp sensor 2/temp2_max_alarm
+      - max6658-i2c-73-4c/Back panel temp sensor 2/temp2_crit_alarm
+      - max6658-i2c-73-4c/Back panel temp sensor 2/temp2_fault
+      - pmbus-i2c-3-58/Power supply 1 exhaust temp sensor/temp3_alarm
+      - pmbus-i2c-3-58/Power supply 1 inlet temp sensor/temp2_alarm
+      - pmbus-i2c-3-58/Power supply 1 hotspot sensor/temp1_alarm
+      - pmbus-i2c-4-58/Power supply 2 exhaust temp sensor/temp3_alarm
+      - pmbus-i2c-4-58/Power supply 2 inlet temp sensor/temp2_alarm
+      - pmbus-i2c-4-58/Power supply 2 hotspot sensor/temp1_alarm
+
+    compares:
+      fan: []
+      power:
+      - - pmbus-i2c-3-58/iin/curr1_input
+        - pmbus-i2c-3-58/iin/curr1_max
+      - - pmbus-i2c-3-58/iout1/curr2_input
+        - pmbus-i2c-3-58/iout1/curr2_max
+      - - pmbus-i2c-4-58/iin/curr1_input
+        - pmbus-i2c-4-58/iin/curr1_max
+      - - pmbus-i2c-4-58/iout1/curr2_input
+        - pmbus-i2c-4-58/iout1/curr2_max
+      temp:
+      - - coretemp-isa-0000/Physical id 0/temp1_input
+        - coretemp-isa-0000/Physical id 0/temp1_max
+      - - coretemp-isa-0000/Core 0/temp2_input
+        - coretemp-isa-0000/Core 0/temp2_max
+      - - coretemp-isa-0000/Core 1/temp3_input
+        - coretemp-isa-0000/Core 1/temp3_max
+      - - lm73-i2c-88-48/Front panel temp sensor/temp1_input
+        - lm73-i2c-88-48/Front panel temp sensor/temp1_max
+      - - max6658-i2c-1-4c/Asic temp sensor/temp1_input
+        - max6658-i2c-1-4c/Asic temp sensor/temp1_max
+      - - max6658-i2c-73-4c/Back panel temp sensor 1/temp1_input
+        - max6658-i2c-73-4c/Back panel temp sensor 1/temp1_max
+      - - max6658-i2c-73-4c/Back panel temp sensor 2/temp2_input
+        - max6658-i2c-73-4c/Back panel temp sensor 2/temp2_max
+
+    non_zero:
+      fan:
+      - pmbus-i2c-3-58/fan1/fan1_input
+      - pmbus-i2c-4-58/fan1/fan1_input
+      - la_cpld-i2c-85-60/fan1/fan1_input
+      - la_cpld-i2c-85-60/fan2/fan2_input
+      - la_cpld-i2c-85-60/fan3/fan3_input
+      - la_cpld-i2c-85-60/fan4/fan4_input
+      power:
+      - pmbus-i2c-4-58/pin/power1_input
+      - pmbus-i2c-4-58/pout1/power2_input
+      - pmbus-i2c-4-58/pout2/power3_input
+      - pmbus-i2c-3-58/pin/power1_input
+      - pmbus-i2c-3-58/pout1/power2_input
+      - pmbus-i2c-3-58/pout2/power3_input
+      temp:
+      - pch_haswell-virtual-0/temp1/temp1_input
+
+    psu_skips: {}

ansible/library/extract_log.py

@@ -80,8 +80,6 @@
 from datetime import datetime
 from ansible.module_utils.basic import *
 
-from pprint import pprint
-
 
 def extract_lines(directory, filename, target_string):
     path = os.path.join(directory, filename)
@@ -100,6 +98,7 @@ def extract_lines(directory, filename, target_string):
 
     return result
 
+
 def extract_number(s):
     """Extracts number from string, if not number found returns 0"""
     ns = re.findall(r'\d+', s)
@@ -112,6 +111,10 @@ def extract_number(s):
 def convert_date(s):
     dt = None
     re_result = re.findall(r'^\S{3}\s{1,2}\d{1,2} \d{2}:\d{2}:\d{2}\.?\d*', s)
+    # Workaround for pytest-ansible
+    loc = locale.getlocale()
+    locale.setlocale(locale.LC_ALL, (None, None))
+
     if len(re_result) > 0:
         str_date = re_result[0]
         try:
@@ -122,6 +125,7 @@ def convert_date(s):
         re_result = re.findall(r'^\d{4}-\d{2}-\d{2}\.\d{2}:\d{2}:\d{2}\.\d{6}', s)
         str_date = re_result[0]
         dt = datetime.strptime(str_date, '%Y-%m-%d.%X.%f')
+    locale.setlocale(locale.LC_ALL, loc)
 
     return dt
 

ansible/roles/eos/handlers/common_handlers/update_state.yml

@@ -1,5 +1,8 @@
 - name: Reboot the VM
   command: /sbin/shutdown -r now "Ansible updates triggered"
+  async: 300
+  poll: 0
+  ignore_errors: true
 
 - name: Wait for VM to shutdown
   wait_for:

ansible/roles/test/files/acstests/everflow_policer_test.py

@@ -100,7 +100,16 @@ def checkMirroredFlow(self):
         """
         @summary: Send traffic & check how many mirrored packets are received
         @return: count: number of mirrored packets received
+
+        Note:
+        Mellanox crafts the GRE packets with extra information:
+        That is: 22 bytes extra information after the GRE header
         """
+        payload = self.base_pkt
+        if self.asic_type in ["mellanox"]:
+            import binascii
+            payload = binascii.unhexlify("0"*44) + str(payload) # Add the padding
+
         exp_pkt = testutils.simple_gre_packet(
                 eth_src = self.router_mac,
                 ip_src = self.session_src_ip,
@@ -109,15 +118,21 @@ def checkMirroredFlow(self):
                 ip_id = 0,
                 #ip_flags = 0x10, # need to upgrade ptf version to support it
                 ip_ttl = self.session_ttl,
-                inner_frame = self.base_pkt)
+                inner_frame = payload)
 
-        exp_pkt['GRE'].proto = 0x88be
+        if self.asic_type in ["mellanox"]:
+            exp_pkt['GRE'].proto = 0x8949 # Mellanox specific
+        else:
+            exp_pkt['GRE'].proto = 0x88be
 
         masked_exp_pkt = Mask(exp_pkt)
         masked_exp_pkt.set_do_not_care_scapy(scapy.Ether, "dst")
         masked_exp_pkt.set_do_not_care_scapy(scapy.IP, "flags")
         masked_exp_pkt.set_do_not_care_scapy(scapy.IP, "chksum")
 
+        if self.asic_type in ["mellanox"]:
+            masked_exp_pkt.set_do_not_care(304, 176) # Mask the Mellanox specific inner header
+
         self.dataplane.flush()
 
         count = 0

ansible/roles/test/files/ptftests/advanced-reboot.py

@@ -319,16 +319,16 @@ def get_portchannel_info(self):
             for member in content[key]['members']:
                 for vm_key in self.vm_dut_map.keys():
                     if member in self.vm_dut_map[vm_key]['dut_ports']:
-                        self.vm_dut_map[vm_key]['dut_portchannel'] = key
+                        self.vm_dut_map[vm_key]['dut_portchannel'] = str(key)
                         self.vm_dut_map[vm_key]['neigh_portchannel'] = 'Port-Channel1'
                         break
 
     def get_neigh_port_info(self):
         content = self.read_json('neigh_port_info')
         for key in content.keys():
             if content[key]['name'] in self.vm_dut_map.keys():
-                self.vm_dut_map[content[key]['name']]['dut_ports'].append(key)
-                self.vm_dut_map[content[key]['name']]['neigh_ports'].append(content[key]['port'])
+                self.vm_dut_map[content[key]['name']]['dut_ports'].append(str(key))
+                self.vm_dut_map[content[key]['name']]['neigh_ports'].append(str(content[key]['port']))
                 self.vm_dut_map[content[key]['name']]['ptf_ports'].append(self.port_indices[key])
 
     def build_peer_mapping(self):
@@ -355,6 +355,30 @@ def populate_fail_info(self, fails):
                 self.fails[key] = set()
             self.fails[key] |= fails[key]
 
+    def get_preboot_info(self):
+        '''
+        Prepares the msg string to log when a preboot_oper is defined.
+        preboot_oper can be represented in the following ways
+           eg. 'preboot_oper' - a single VM will be selected and preboot_oper will be applied to it
+               'neigh_bgp_down:2' - 2 VMs will be selected and preboot_oper will be applied to the selected 2 VMs
+               'neigh_lag_member_down:3:1' - this case is used for lag member down operation only. This indicates that
+                                             3 VMs will be selected and 1 of the lag members in the porchannel will be brought down
+        '''
+        msg = ''
+        if self.preboot_oper:
+            msg = 'Preboot oper: %s ' % self.preboot_oper
+            if ':' in self.preboot_oper:
+                oper_list = self.preboot_oper.split(':')
+                msg = 'Preboot oper: %s ' % oper_list[0] # extract the preboot oper_type
+                if len(oper_list) > 2:
+                    # extract the number of VMs and the number of LAG members. preboot_oper will be of the form oper:no of VMS:no of lag members
+                    msg += 'Number of sad path VMs: %s Lag member down in a portchannel: %s' % (oper_list[-2], oper_list[-1])
+                else:
+                    # extract the number of VMs. preboot_oper will be of the form oper:no of VMS
+                    msg += 'Number of sad path VMs: %s' % oper_list[-1]
+
+        return msg
+
     def setUp(self):
         self.fails['dut'] = set()
         self.port_indices = self.read_port_indices()
@@ -427,13 +451,7 @@ def setUp(self):
         self.generate_arp_ping_packet()
 
         if self.reboot_type == 'warm-reboot':
-            # get the number of members down for sad path
-            if self.preboot_oper:
-                if ':' in self.preboot_oper:
-                    oper_type, cnt = self.preboot_oper.split(':')
-                else:
-                    oper_type, cnt = self.preboot_oper, 1
-                self.log("Preboot Oper: %s Number down: %s" % (oper_type, cnt))
+            self.log(self.get_preboot_info())
 
             # Pre-generate list of packets to be sent in send_in_background method.
             generate_start = datetime.datetime.now()
@@ -777,15 +795,19 @@ def wait_for_ssh_threads():
             if self.reboot_type == 'fast-reboot' and no_cp_replies < 0.95 * self.nr_vl_pkts:
                 self.fails['dut'].add("Dataplane didn't route to all servers, when control-plane was down: %d vs %d" % (no_cp_replies, self.nr_vl_pkts))
 
-            if self.reboot_type == 'warm-reboot' and self.preboot_oper is not None:
-                if self.pre_handle is not None:
+            if self.reboot_type == 'warm-reboot':
+                if self.preboot_oper is not None and self.pre_handle is not None:
                     self.log("Postboot checks:")
                     log_info, fails = self.pre_handle.verify(pre_check=False)
                     self.populate_fail_info(fails)
                     for log in log_info:
                         self.log(log)
                     self.log(" ")
 
+                else:
+                    # verify there are no interface flaps after warm boot
+                    self.neigh_lag_status_check()
+
         except Exception as e:
             self.fails['dut'].add(e)
         finally:
@@ -864,6 +886,21 @@ def wait_for_ssh_threads():
 
             self.assertTrue(is_good, errors)
 
+    def neigh_lag_status_check(self):
+        """
+        Ensure there are no interface flaps after warm-boot
+        """
+        for neigh in self.ssh_targets:
+            self.neigh_handle = Arista(neigh, None, self.test_params)
+            self.neigh_handle.connect()
+            fails, flap_cnt = self.neigh_handle.verify_neigh_lag_no_flap()
+            self.neigh_handle.disconnect()
+            self.fails[neigh] |= fails
+            if not flap_cnt:
+                self.log("No LAG flaps seen on %s after warm boot" % neigh)
+            else:
+                self.fails[neigh].add("LAG flapped %s times on %s after warm boot" % (flap_cnt, neigh))
+
     def extract_no_cpu_replies(self, arr):
       """
       This function tries to extract number of replies from dataplane, when control plane is non working

ansible/roles/test/files/ptftests/arista.py

@@ -137,6 +137,9 @@ def run(self):
             sample["po_changetime"] = json.loads(portchannel_output, strict=False)['interfaces']['Port-Channel1']['lastStatusChangeTimestamp']
 
             if not run_once:
+                # clear Portchannel counters
+                self.do_cmd("clear counters Port-Channel 1")
+
                 self.ipv4_gr_enabled, self.ipv6_gr_enabled, self.gr_timeout = self.parse_bgp_neighbor_once(bgp_neig_output)
                 if self.gr_timeout is not None:
                     log_first_line = "session_begins_%f" % cur_time
@@ -393,18 +396,23 @@ def verify_bgp_neigh_state(self, dut=None, state="Active"):
                     self.fails.add('Verify BGP %s neighbor: Object missing in output' % ver)
         return self.fails, bgp_state
 
-    def change_neigh_lag_state(self, lag, is_up=True):
+    def change_neigh_lag_state(self, intf, is_up=True):
         state = ['shut', 'no shut']
         self.do_cmd('configure')
-        is_match = re.match('(Port-Channel|Ethernet)\d+', lag)
+        is_match = re.match('(Port-Channel|Ethernet)\d+', intf)
         if is_match:
-            output = self.do_cmd('interface %s' % lag)
+            output = self.do_cmd('interface %s' % intf)
             if 'Invalid' not in output:
                 self.do_cmd(state[is_up])
                 self.do_cmd('exit')
-            self.do_cmd('exit')
+        self.do_cmd('exit')
+
+    def change_neigh_intfs_state(self, intfs, is_up=True):
+        for intf in intfs:
+            self.change_neigh_lag_state(intf, is_up=is_up)
 
     def verify_neigh_lag_state(self, lag, state="connected", pre_check=True):
+        states = state.split(',')
         lag_state = False
         msg_prefix = ['Postboot', 'Preboot']
         is_match = re.match('(Port-Channel|Ethernet)\d+', lag)
@@ -415,14 +423,31 @@ def verify_neigh_lag_state(self, lag, state="connected", pre_check=True):
                 obj = json.loads(data)
 
                 if 'interfaces' in obj and lag in obj['interfaces']:
-                    lag_state = (obj['interfaces'][lag]['interfaceStatus'] == state)
+                    lag_state = (obj['interfaces'][lag]['interfaceStatus'] in states)
                 else:
                     self.fails.add('%s: Verify LAG %s: Object missing in output' % (msg_prefix[pre_check], lag))
                 return self.fails, lag_state
 
         self.fails.add('%s: Invalid interface name' % msg_prefix[pre_check])
         return self.fails, lag_state
 
+    def verify_neigh_lag_no_flap(self):
+        flap_cnt = sys.maxint
+        output = self.do_cmd('show interfaces Po1 | json')
+        if 'Invalid' not in output:
+            data = '\n'.join(output.split('\r\n')[1:-1])
+            obj = json.loads(data)
+
+            if 'interfaces' in obj and 'Port-Channel1' in obj['interfaces']:
+                intf_cnt_info = obj['interfaces']['Port-Channel1']['interfaceCounters']
+                flap_cnt = intf_cnt_info['linkStatusChanges']
+            else:
+                self.fails.add('Object missing in output for Port-Channel1')
+            return self.fails, flap_cnt
+
+        self.fails.add('Invalid interface name - Po1')
+        return self.fails, flap_cnt
+
     def check_gr_peer_status(self, output):
         # [0] True 'ipv4_gr_enabled', [1] doesn't matter 'ipv6_enabled', [2] should be >= 120
         if not self.ipv4_gr_enabled:

ansible/roles/test/files/ptftests/fib.py

@@ -17,7 +17,6 @@
 ]
 
 EXCLUDE_IPV6_PREFIXES = [
-        '::/0',                 # Currently no IPv6 default route
         '::/128',               # Unspecified           RFC 4291
         '::1/128',              # Loopback              RFC 4291
         'ff00::/8'              # Multicast             RFC 4291