From e46dd294092f9267d9352fdc1e4889d8db2a9e6a Mon Sep 17 00:00:00 2001
From: Volodymyr Samotiy <volodymyrs@nvidia.com>
Date: Fri, 30 Sep 2022 20:52:31 +0300
Subject: [PATCH] [crm] Fix issue with continues EXCEEDED and CLEAR logs for
 ACL group/table counters (#2463)

*Moved exceededLogCounter from CrmResourceEntry to CrmResourceCounter.
---
 orchagent/crmorch.cpp | 23 ++++++++++++++---------
 orchagent/crmorch.h   |  2 +-
 tests/test_crm.py     | 22 ++++++++++++++++------
 3 files changed, 31 insertions(+), 16 deletions(-)

diff --git a/orchagent/crmorch.cpp b/orchagent/crmorch.cpp
index 54e2df78c7e6..f02bbd68f26d 100644
--- a/orchagent/crmorch.cpp
+++ b/orchagent/crmorch.cpp
@@ -340,13 +340,18 @@ void CrmOrch::handleSetCommand(const string& key, const vector<FieldValueTuple>&
             }
             else if (crmThreshTypeResMap.find(field) != crmThreshTypeResMap.end())
             {
-                auto resourceType = crmThreshTypeResMap.at(field);
                 auto thresholdType = crmThreshTypeMap.at(value);
+                auto resourceType = crmThreshTypeResMap.at(field);
+                auto &resource = m_resourcesMap.at(resourceType);
 
-                if (m_resourcesMap.at(resourceType).thresholdType != thresholdType)
+                if (resource.thresholdType != thresholdType)
                 {
-                    m_resourcesMap.at(resourceType).thresholdType = thresholdType;
-                    m_resourcesMap.at(resourceType).exceededLogCounter = 0;
+                    resource.thresholdType = thresholdType;
+
+                    for (auto &cnt : resource.countersMap)
+                    {
+                        cnt.second.exceededLogCounter = 0;
+                    }
                 }
             }
             else if (crmThreshLowResMap.find(field) != crmThreshLowResMap.end())
@@ -723,7 +728,7 @@ void CrmOrch::checkCrmThresholds()
     {
         auto &res = i.second;
 
-        for (const auto &j : i.second.countersMap)
+        for (auto &j : i.second.countersMap)
         {
             auto &cnt = j.second;
             uint64_t utilization = 0;
@@ -762,7 +767,7 @@ void CrmOrch::checkCrmThresholds()
                     throw runtime_error("Unknown threshold type for CRM resource");
             }
 
-            if ((utilization >= res.highThreshold) && (res.exceededLogCounter < CRM_EXCEEDED_MSG_MAX))
+            if ((utilization >= res.highThreshold) && (cnt.exceededLogCounter < CRM_EXCEEDED_MSG_MAX))
             {
                 event_params_t params = {
                     { "percent", to_string(percentageUtil) },
@@ -773,14 +778,14 @@ void CrmOrch::checkCrmThresholds()
                               res.name.c_str(), threshType.c_str(), percentageUtil, cnt.usedCounter, cnt.availableCounter);
 
                 event_publish(g_events_handle, "chk_crm_threshold", &params);
-                res.exceededLogCounter++;
+                cnt.exceededLogCounter++;
             }
-            else if ((utilization <= res.lowThreshold) && (res.exceededLogCounter > 0) && (res.highThreshold != res.lowThreshold))
+            else if ((utilization <= res.lowThreshold) && (cnt.exceededLogCounter > 0) && (res.highThreshold != res.lowThreshold))
             {
                 SWSS_LOG_WARN("%s THRESHOLD_CLEAR for %s %u%% Used count %u free count %u",
                               res.name.c_str(), threshType.c_str(), percentageUtil, cnt.usedCounter, cnt.availableCounter);
 
-                res.exceededLogCounter = 0;
+                cnt.exceededLogCounter = 0;
             }
         } // end of counters loop
     } // end of resources loop
diff --git a/orchagent/crmorch.h b/orchagent/crmorch.h
index a9c32a83bde7..b57d4b2947c9 100644
--- a/orchagent/crmorch.h
+++ b/orchagent/crmorch.h
@@ -74,6 +74,7 @@ class CrmOrch : public Orch
         sai_object_id_t id = 0;
         uint32_t availableCounter = 0;
         uint32_t usedCounter = 0;
+        uint32_t exceededLogCounter = 0;
     };
 
     struct CrmResourceEntry
@@ -88,7 +89,6 @@ class CrmOrch : public Orch
 
         std::map<std::string, CrmResourceCounter> countersMap;
 
-        uint32_t exceededLogCounter = 0;
         CrmResourceStatus resStatus = CrmResourceStatus::CRM_RES_SUPPORTED;
     };
 
diff --git a/tests/test_crm.py b/tests/test_crm.py
index 31d0b57ae569..bee145c34feb 100644
--- a/tests/test_crm.py
+++ b/tests/test_crm.py
@@ -697,12 +697,22 @@ def test_CrmAclGroup(self, dvs, testlog):
         entry_used_counter = getCrmCounterValue(dvs, 'ACL_STATS:INGRESS:PORT', 'crm_stats_acl_group_used')
         assert entry_used_counter == 3
 
-        # remove ACL table
-        #tbl._del("test-aclv6")
-        #time.sleep(2)
-        #atbl = swsscommon.Table(adb, "ASIC_STATE:SAI_OBJECT_TYPE_ACL_TABLE_GROUP")
-        #table_used_counter = getCrmCounterValue(dvs, 'ACL_STATS:INGRESS:PORT', 'crm_stats_acl_group_used')
-        #assert table_used_counter == 0
+        marker = dvs.add_log_marker()
+        crm_update(dvs, "polling_interval", "1")
+        crm_update(dvs, "acl_group_threshold_type", "used")
+        crm_update(dvs, "acl_group_low_threshold", str(0))
+        crm_update(dvs, "acl_group_high_threshold", str(2))
+
+        time.sleep(2)
+        check_syslog(dvs, marker, "ACL_GROUP THRESHOLD_EXCEEDED for TH_USED", 1)
+        check_syslog(dvs, marker, "ACL_GROUP THRESHOLD_CLEAR for TH_USED", 0)
+
+        tbl._del("test-aclv6")
+        time.sleep(2)
+        check_syslog(dvs, marker, "ACL_GROUP THRESHOLD_CLEAR for TH_USED", 1)
+
+        table_used_counter = getCrmCounterValue(dvs, 'ACL_STATS:INGRESS:PORT', 'crm_stats_acl_group_used')
+        assert table_used_counter == 0
 
     def test_CrmSnatEntry(self, dvs, testlog):