From 8d369e517315d2f1e8a94878f16b2916bc22d131 Mon Sep 17 00:00:00 2001
From: rbpittman <rapittma@cisco.com>
Date: Tue, 15 Apr 2025 04:20:08 -0400
Subject: [PATCH] Disable PFC-WD during PCBB and some wmk test improvements
 (#17889)

* Improve validation so multiple failures can be reported.

* Disable PFC-WD during PCBB tests.

* Wait for PFC-WD to stop.

* Remove packet aging code from pfcwd fixture.
---
 tests/qos/test_tunnel_qos_remap.py | 36 +++++++++++++++++++++++++-----
 tests/qos/tunnel_qos_remap_base.py |  2 ++
 2 files changed, 33 insertions(+), 5 deletions(-)

diff --git a/tests/qos/test_tunnel_qos_remap.py b/tests/qos/test_tunnel_qos_remap.py
index 9aa9d45a757..ae13b45e5d2 100644
--- a/tests/qos/test_tunnel_qos_remap.py
+++ b/tests/qos/test_tunnel_qos_remap.py
@@ -13,6 +13,8 @@
 from tests.common.fixtures.duthost_utils import dut_qos_maps_module       # noqa F401
 from tests.common.fixtures.duthost_utils import separated_dscp_to_tc_map_on_uplink
 from tests.common.helpers.assertions import pytest_require, pytest_assert
+from tests.common.snappi_tests.qos_fixtures import get_pfcwd_config, reapply_pfcwd
+from tests.common.snappi_tests.common_helpers import stop_pfcwd
 
 from tests.common.dualtor.mux_simulator_control import toggle_all_simulator_ports_to_lower_tor,\
     toggle_all_simulator_ports_to_rand_selected_tor, toggle_all_simulator_ports_to_rand_unselected_tor  # noqa F401
@@ -65,6 +67,18 @@ def check_running_condition(tbinfo, duthost):
                    "Only run when tunnel_qos_remap is enabled", True)
 
 
+@pytest.fixture(scope='module', autouse=True)
+def disable_pfcwd(duthosts):
+    pfcwd_value = {}
+    for duthost in duthosts:
+        pfcwd_value[duthost.hostname] = get_pfcwd_config(duthost)
+        stop_pfcwd(duthost)
+    yield
+    for duthost in duthosts:
+        reapply_pfcwd(duthost, pfcwd_value[duthost.hostname])
+    return
+
+
 def _last_port_in_last_lag(lags):
     """
     A helper function to get the last LAG member in the last portchannel
@@ -552,6 +566,7 @@ def test_pfc_watermark_extra_lossless_standby(ptfhost, fanouthosts, rand_selecte
     active_tor_mac = rand_unselected_dut.facts['router_mac']
     mg_facts = rand_selected_dut.get_extended_minigraph_facts(tbinfo)
     ptfadapter.dataplane.flush()
+    failures = []
     for inner_dscp, outer_dscp, prio, queue in TEST_DATA:
         wmk_stat_queue = queue
         if "cisco-8000" in dut_config["asic_type"]:
@@ -612,9 +627,14 @@ def test_pfc_watermark_extra_lossless_standby(ptfhost, fanouthosts, rand_selecte
         logger.info(("Congested queue watermark on {}|{} is {}, increased by {}," +
                      "minimum required increase is {}").format(
             actual_port_name, wmk_stat_queue, queue_wmk, queue_wmk - base_queue_wmk, required_wmk_inc_bytes))
-        assert queue_wmk > required_wmk_bytes, \
-            "Failed to detect congestion due to PFC pause, failed check {} > {}".format(
+        if queue_wmk <= required_wmk_bytes:
+            msg = "For inner_dscp, outer_dscp, prio, queue = ({}, {}, {}, {}):\n".format(
+                inner_dscp, outer_dscp, prio, queue)
+            msg += "  Failed to detect congestion due to PFC pause, failed check {} > {}".format(
                 queue_wmk, required_wmk_bytes)
+            logger.info(msg)
+            failures.append(msg)
+    assert len(failures) == 0, "Watermark failures were found:\n{}".format("\n".join(failures))
 
 
 def test_pfc_watermark_extra_lossless_active(ptfhost, fanouthosts, rand_selected_dut, rand_unselected_dut,
@@ -645,6 +665,7 @@ def test_pfc_watermark_extra_lossless_active(ptfhost, fanouthosts, rand_selected
     active_tor_mac = rand_selected_dut.facts['router_mac']
     mg_facts = rand_unselected_dut.get_extended_minigraph_facts(tbinfo)
     ptfadapter.dataplane.flush()
+    failures = []
     for inner_dscp, outer_dscp, prio, queue in TEST_DATA:
         pkt, tunnel_pkt = build_testing_packet(src_ip=DUMMY_IP,
                                                dst_ip=dualtor_meta['target_server_ip'],
@@ -703,9 +724,14 @@ def test_pfc_watermark_extra_lossless_active(ptfhost, fanouthosts, rand_selected
                      "minimum required increase is {}").format(
             dualtor_meta['selected_port'], queue, queue_wmk,
             queue_wmk - base_queue_wmk, required_wmk_inc_bytes))
-        assert queue_wmk > required_wmk_bytes, \
-            "Failed to detect congestion due to PFC pause, failed check {} > {}".format(
-                queue_wmk, base_queue_wmk)
+        if queue_wmk <= required_wmk_bytes:
+            msg = "For inner_dscp, outer_dscp, prio, queue = ({}, {}, {}, {}):\n".format(
+                inner_dscp, outer_dscp, prio, queue)
+            msg += "  Failed to detect congestion due to PFC pause, failed check {} > {}".format(
+                queue_wmk, required_wmk_bytes)
+            logger.info(msg)
+            failures.append(msg)
+    assert len(failures) == 0, "Watermark failures were found:\n{}".format("\n".join(failures))
 
 
 @pytest.mark.disable_loganalyzer
diff --git a/tests/qos/tunnel_qos_remap_base.py b/tests/qos/tunnel_qos_remap_base.py
index 6d24f3ae960..4025b6e6424 100644
--- a/tests/qos/tunnel_qos_remap_base.py
+++ b/tests/qos/tunnel_qos_remap_base.py
@@ -504,6 +504,8 @@ def stop_pfc_storm(storm_handler):
     Stop sending PFC pause frames from fanout switch
     """
     storm_handler.stop_storm()
+    # Wait for PFC pause to stop
+    time.sleep(2)
 
 
 def run_ptf_test(ptfhost, test_case='', test_params={}):