Merge branch 'bugfix/fix_ot_ci_testcases_v5.4' into 'release/v5.4'

ci: improve BR host related test cases (v5.4)

See merge request espressif/esp-idf!47263

examples/openthread/ot_ci_function.py

@@ -1,7 +1,8 @@
-# SPDX-FileCopyrightText: 2022-2025 Espressif Systems (Shanghai) CO LTD
+# SPDX-FileCopyrightText: 2022-2026 Espressif Systems (Shanghai) CO LTD
 # SPDX-License-Identifier: Unlicense OR CC0-1.0
 # !/usr/bin/env python3
 # this file defines some functions for testing cli and br under pytest framework
+import ipaddress
 import logging
 import os
 import re
@@ -380,6 +381,78 @@ def is_joined_wifi_network(br: IdfDut) -> bool:
     return check_if_host_receive_ra(br)
 
 
+def wait_for_host_ra_route(
+    br: IdfDut,
+    *,
+    retries: int = 12,
+    interval_s: int = 5,
+) -> None:
+    interface_name = get_host_interface_name()
+    log_ipv6_addr_route_by_interface(interface_name, title='Wait RA (initial)')
+
+    for attempt in range(1, retries + 1):
+        if is_joined_wifi_network(br):
+            log_ipv6_addr_route_by_interface(interface_name, title='RA Ready!')
+            return
+
+        logging.info(f'Host route not ready yet, retry {attempt}/{retries}...')
+        log_ipv6_addr_route_by_interface(interface_name, title=f'Wait RA ({attempt}/{retries})')
+
+        time.sleep(interval_s)
+
+    raise AssertionError('Host did not receive RA / OMR route in time')
+
+
+def host_global_address_has_onlink_prefix(interface_name: str, onlinkprefix: str) -> bool:
+    onlinkprefix = onlinkprefix.strip()
+    if not onlinkprefix:
+        return False
+    base = onlinkprefix.rstrip(':')
+    try:
+        network = ipaddress.IPv6Network(f'{base}::/64', strict=False)
+    except ValueError:
+        logging.warning(f'Invalid onlinkprefix for /64 check: {onlinkprefix}')
+        return False
+
+    out = subprocess.getoutput(f'ip -6 addr show dev {interface_name}')
+    for line in out.splitlines():
+        if 'inet6' not in line or 'scope global' not in line:
+            continue
+        m = re.search(r'inet6 ([^\s]+)/\d+', line)
+        if not m:
+            continue
+        addr_s = m.group(1).split('%')[0]
+        try:
+            if ipaddress.IPv6Address(addr_s) in network:
+                return True
+        except ValueError:
+            continue
+    return False
+
+
+def wait_for_host_onlink_global_address(
+    br: IdfDut,
+    *,
+    retries: int = 12,
+    interval_s: int = 5,
+) -> str:
+    interface_name = get_host_interface_name()
+    onlinkprefix = get_onlinkprefix(br)
+    logging.info(f'Wait for host GUA in BR onlink prefix {onlinkprefix!r} on {interface_name}')
+    log_ipv6_addr_route_by_interface(interface_name, title='Wait onlink GUA (initial)')
+
+    for attempt in range(1, retries + 1):
+        if host_global_address_has_onlink_prefix(interface_name, onlinkprefix):
+            log_ipv6_addr_route_by_interface(interface_name, title='Onlink GUA ready!')
+            return onlinkprefix
+
+        logging.info(f'Host onlink GUA not ready yet, retry {attempt}/{retries}...')
+        log_ipv6_addr_route_by_interface(interface_name, title=f'Wait onlink GUA ({attempt}/{retries})')
+        time.sleep(interval_s)
+
+    raise AssertionError(f'Host did not get a global IPv6 address in onlink prefix {onlinkprefix!r} in time')
+
+
 thread_ipv6_group = 'ff04:0:0:0:0:0:0:125'
 
 
@@ -546,35 +619,72 @@ def open_host_interface() -> None:
         assert flag
 
 
+def ensure_avahi_running(restart_if_needed: bool = True) -> bool:
+    out_str = subprocess.getoutput('pgrep -a avahi-daemon 2>/dev/null')
+    if out_str.strip():
+        logging.info(f'avahi process list:\n{out_str}')
+        return True
+
+    logging.warning('avahi-daemon not running')
+    if restart_if_needed:
+        logging.warning('restarting avahi-daemon once...')
+        restart_avahi()
+        out_str = subprocess.getoutput('pgrep -a avahi-daemon 2>/dev/null')
+        if out_str.strip():
+            logging.info(f'avahi process list after restart:\n{out_str}')
+            return True
+
+    logging.error('avahi-daemon is still not running')
+    return False
+
+
 def get_domain() -> str:
     hostname = socket.gethostname()
     logging.info(f'hostname is: {hostname}')
-    command = 'ps -auxww | grep avahi-daemon | grep running'
+    out_str = subprocess.getoutput('pgrep -a avahi-daemon 2>/dev/null')
-    out_str = subprocess.getoutput(command)
+    if not out_str.strip():
-    logging.info(f'avahi status:\n {out_str}')
+        out_str = subprocess.getoutput('ps -C avahi-daemon -o args= --no-headers 2>/dev/null')
-    role = re.findall(r'\[([\w\W]+)\.local\]', str(out_str))[0]
+    logging.info(f'avahi status:\n{out_str}')
-    logging.info(f'active host is: {role}')
+    matches = re.findall(r'\[([\w\W]+?)\.local\]', str(out_str))
-    return str(role)
+    if matches:
+        role = matches[0]
+        logging.info(f'active host is: {role}')
+        return str(role)
+    short = subprocess.getoutput('hostname -s').strip() or hostname.split('.')[0]
+    logging.warning(
+        'Could not parse [.local] from avahi process args; using short hostname %r',
+        short,
+    )
+    return short
 
 
-def flush_ipv6_addr_by_interface() -> None:
+def log_ipv6_addr_route_by_interface(interface_name: str, title: str = '') -> Tuple[str, str]:
-    interface_name = get_host_interface_name()
-    logging.info(f'flush ipv6 addr : {interface_name}')
     command_show_addr = f'ip -6 addr show dev {interface_name}'
     command_show_route = f'ip -6 route show dev {interface_name}'
-    addr_before = subprocess.getoutput(command_show_addr)
+    addr = subprocess.getoutput(command_show_addr)
-    route_before = subprocess.getoutput(command_show_route)
+    route = subprocess.getoutput(command_show_route)
-    logging.info(f'Before flush, IPv6 addresses: \n{addr_before}')
+    prefix = f'{title} ' if title else ''
-    logging.info(f'Before flush, IPv6 routes: \n{route_before}')
+    logging.info(f'{prefix}IPv6 addresses on {interface_name}:\n{addr}')
+    logging.info(f'{prefix}IPv6 routes on {interface_name}:\n{route}')
+    return addr, route
+
+
+def flush_ipv6_addr_route_by_interface(interface_name: str, down_up_wait_s: int = 5) -> None:
+    logging.info(f'flush ipv6 addr/route: {interface_name}')
+    log_ipv6_addr_route_by_interface(interface_name, title='Before flush')
+
     subprocess.run(['ip', 'link', 'set', interface_name, 'down'])
     subprocess.run(['ip', '-6', 'addr', 'flush', 'dev', interface_name])
     subprocess.run(['ip', '-6', 'route', 'flush', 'dev', interface_name])
     subprocess.run(['ip', 'link', 'set', interface_name, 'up'])
-    time.sleep(5)
+
-    addr_after = subprocess.getoutput(command_show_addr)
+    time.sleep(down_up_wait_s)
-    route_after = subprocess.getoutput(command_show_route)
+    log_ipv6_addr_route_by_interface(interface_name, title='After flush')
-    logging.info(f'After flush, IPv6 addresses: \n{addr_after}')
+
-    logging.info(f'After flush, IPv6 routes: \n{route_after}')
+
+def flush_ipv6_addr_by_interface() -> None:
+    interface_name = get_host_interface_name()
+    flush_ipv6_addr_route_by_interface(interface_name)
 
 
 class tcp_parameter:

examples/openthread/pytest_otbr.py

@@ -10,6 +10,7 @@ import secrets
 import subprocess
 import threading
 import time
+from typing import Optional
 from typing import Tuple
 
 import ot_ci_function as ocf
@@ -195,19 +196,22 @@ def test_Bidirectional_IPv6_connectivity(Init_interface: bool, dut: Tuple[IdfDut
 
     formBasicWiFiThreadNetwork(br, cli)
     try:
-        assert ocf.is_joined_wifi_network(br)
+        ocf.wait_for_host_ra_route(br)
+        onlinkprefix = ocf.wait_for_host_onlink_global_address(br)
+        logging.info(f'br onlinkprefix: {onlinkprefix}')
         cli_global_unicast_addr = ocf.get_global_unicast_addr(cli, br)
         logging.info(f'cli_global_unicast_addr {cli_global_unicast_addr}')
+        interface_name = ocf.get_host_interface_name()
+        ocf.log_ipv6_addr_route_by_interface(interface_name, title='Before ping test')
         command = 'ping ' + str(cli_global_unicast_addr) + ' -c 10'
         out_str = subprocess.getoutput(command)
+        ocf.log_ipv6_addr_route_by_interface(interface_name, title='After ping test')
         logging.info(f'ping result:\n{out_str}')
         role = re.findall(r' (\d+)%', str(out_str))[0]
         assert role != '100'
-        interface_name = ocf.get_host_interface_name()
         command = 'ifconfig ' + interface_name + ' | grep inet6 | grep global'
         out_bytes = subprocess.check_output(command, shell=True, timeout=5)
         out_str = out_bytes.decode('utf-8')
-        onlinkprefix = ocf.get_onlinkprefix(br)
         pattern = rf'\W+({onlinkprefix}(?:\w+:){{3}}\w+)\W+'
         host_global_unicast_addr = re.findall(pattern, out_str)
         logging.info(f'host_global_unicast_addr: {host_global_unicast_addr}')
@@ -252,7 +256,7 @@ def test_multicast_forwarding_A(Init_interface: bool, dut: Tuple[IdfDut, IdfDut,
 
     formBasicWiFiThreadNetwork(br, cli)
     try:
-        assert ocf.is_joined_wifi_network(br)
+        ocf.wait_for_host_ra_route(br)
         ocf.execute_command(br, 'bbr')
         br.expect('server16', timeout=5)
         assert ocf.thread_is_joined_group(cli)
@@ -305,7 +309,7 @@ def test_multicast_forwarding_B(Init_interface: bool, dut: Tuple[IdfDut, IdfDut,
 
     formBasicWiFiThreadNetwork(br, cli)
     try:
-        assert ocf.is_joined_wifi_network(br)
+        ocf.wait_for_host_ra_route(br)
         ocf.execute_command(br, 'bbr')
         br.expect('server16', timeout=5)
         ocf.execute_command(cli, 'udp open')
@@ -362,7 +366,7 @@ def test_service_discovery_of_Thread_device(
 
     formBasicWiFiThreadNetwork(br, cli)
     try:
-        assert ocf.is_joined_wifi_network(br)
+        ocf.wait_for_host_ra_route(br)
         command = 'avahi-browse -rt _testyyy._udp'
         out_str = subprocess.getoutput(command)
         logging.info(f'avahi-browse:\n{out_str}')
@@ -395,7 +399,7 @@ def test_service_discovery_of_Thread_device(
 
 # Case 6: discover dervice published by Wi-Fi device
 @pytest.mark.openthread_br
-@pytest.mark.flaky(reruns=1, reruns_delay=5)
+@pytest.mark.flaky(reruns=3, reruns_delay=5)
 @pytest.mark.parametrize(
     'config, count, app_path, target, port',
     [
@@ -422,13 +426,15 @@ def test_service_discovery_of_WiFi_device(
     dut[0].serial.stop_redirect_thread()
 
     formBasicWiFiThreadNetwork(br, cli)
+    sp: Optional[subprocess.Popen] = None
     try:
-        assert ocf.is_joined_wifi_network(br)
+        ocf.wait_for_host_ra_route(br)
         br_global_unicast_addr = ocf.get_global_unicast_addr(br, br)
         command = 'dns config ' + br_global_unicast_addr
         ocf.execute_command(cli, command)
         cli.expect('Done', timeout=5)
         ocf.wait(cli, 1)
+        assert ocf.ensure_avahi_running(restart_if_needed=True), 'avahi-daemon is not running on this runner'
         domain_name = ocf.get_domain()
         logging.info(f'domain name is: {domain_name}')
         command = 'dns resolve ' + domain_name + '.default.service.arpa.'
@@ -457,7 +463,8 @@ def test_service_discovery_of_WiFi_device(
         assert 'Port:12347' in str(tmp)
     finally:
         ocf.host_close_service()
-        sp.terminate()
+        if sp is not None:
+            sp.terminate()
         ocf.stop_thread(cli)
         ocf.stop_thread(br)
         time.sleep(3)
@@ -490,7 +497,7 @@ def test_ICMP_NAT64(Init_interface: bool, dut: Tuple[IdfDut, IdfDut, IdfDut]) ->
 
     formBasicWiFiThreadNetwork(br, cli)
     try:
-        assert ocf.is_joined_wifi_network(br)
+        ocf.wait_for_host_ra_route(br)
         host_ipv4_address = ocf.get_host_ipv4_address()
         logging.info(f'host_ipv4_address: {host_ipv4_address}')
         rx_nums = ocf.ot_ping(cli, str(host_ipv4_address), count=5)[1]
@@ -528,7 +535,7 @@ def test_UDP_NAT64(Init_interface: bool, dut: Tuple[IdfDut, IdfDut, IdfDut]) ->
 
     formBasicWiFiThreadNetwork(br, cli)
     try:
-        assert ocf.is_joined_wifi_network(br)
+        ocf.wait_for_host_ra_route(br)
         ocf.execute_command(br, 'bbr')
         br.expect('server16', timeout=5)
         ocf.execute_command(cli, 'udp open')
@@ -584,7 +591,7 @@ def test_TCP_NAT64(Init_interface: bool, dut: Tuple[IdfDut, IdfDut, IdfDut]) ->
 
     formBasicWiFiThreadNetwork(br, cli)
     try:
-        assert ocf.is_joined_wifi_network(br)
+        ocf.wait_for_host_ra_route(br)
         ocf.execute_command(br, 'bbr')
         br.expect('server16', timeout=5)
         ocf.execute_command(cli, 'tcpsockclient open')
@@ -796,7 +803,7 @@ def test_br_meshcop(Init_interface: bool, Init_avahi: bool, dut: Tuple[IdfDut, I
         br_thread_para.setnetworkname(networkname)
         ocf.joinThreadNetwork(br, br_thread_para)
         ocf.wait(br, 10)
-        assert ocf.is_joined_wifi_network(br)
+        ocf.wait_for_host_ra_route(br)
         command = 'timeout 3 avahi-browse -r _meshcop._udp'
         try:
             result = subprocess.run(command, capture_output=True, check=True, shell=True)