Roll errors after reaching Cap size

Fixes openbmc/openbmc#2687

-cap on error exists to deal with small flashes, right now the cap
is 100 for high severity errors and it is 10 for info(and below)
severity) errors.
-previous policy was to stop logging new errors once cap is reached
for high severity errors.
-based on feedback from field engineers, it's better to log new
errors, external tools will typically periodically read and store
errors, so we won't lose them.so the new policy now is to roll errors.

Change-Id: I18b509d81c076fcb80667300cbfda050027bb422
Signed-off-by: Nagaraju Goruganti <ngorugan@in.ibm.com>

2 files changed, 27 insertions, 34 deletions

diff --git a/log_manager.cpp b/log_manager.cpp
index 0d8e753..cc5e187 100644
--- a/log_manager.cpp
+++ b/log_manager.cpp
@@ -28,7 +28,6 @@ namespace internal
 void Manager::commit(uint64_t transactionId, std::string errMsg)
 {
     auto reqLevel = level::ERR; // Default to ERR
-    size_t realErrCnt = entries.size() - infoErrors.size();
     auto levelmap = g_errLevelMap.find(errMsg);
 
     if (levelmap != g_errLevelMap.end())
@@ -38,17 +37,9 @@ void Manager::commit(uint64_t transactionId, std::string errMsg)
 
     if (static_cast<Entry::Level>(reqLevel) < Entry::sevLowerLimit)
     {
-        if (capped)
+        if (realErrors.size() >= ERROR_CAP)
         {
-            return;
-        }
-        if (realErrCnt >= ERROR_CAP)
-        {
-            log<level::ERR>("Reached error cap, Ignoring error",
-                            entry("SIZE=%d", realErrCnt),
-                            entry("ERROR_CAP=%d", ERROR_CAP));
-            capped = true;
-            return;
+            erase(realErrors.front());
         }
     }
     else
@@ -171,6 +162,10 @@ void Manager::commit(uint64_t transactionId, std::string errMsg)
     {
         infoErrors.push_back(entryId);
     }
+    else
+    {
+        realErrors.push_back(entryId);
+    }
     auto ms = std::chrono::duration_cast<std::chrono::milliseconds>(
                 std::chrono::system_clock::now().time_since_epoch()).count();
     auto objPath =  std::string(OBJ_ENTRY) + '/' +
@@ -223,13 +218,22 @@ void Manager::erase(uint32_t entryId)
         fs::path errorPath(ERRLOG_PERSIST_PATH);
         errorPath /= std::to_string(entryId);
         fs::remove(errorPath);
-        if (entry->second->severity() >= Entry::sevLowerLimit)
+
+        auto removeId = [](std::list<uint32_t>& ids , uint32_t id)
         {
-            auto it = std::find(infoErrors.begin(), infoErrors.end(), entryId);
-            if (it != infoErrors.end())
+            auto it = std::find(ids.begin(), ids.end(), id);
+            if (it != ids.end())
             {
-                infoErrors.erase(it);
+                ids.erase(it);
             }
+        };
+        if (entry->second->severity() >= Entry::sevLowerLimit)
+        {
+            removeId(infoErrors, entryId);
+        }
+        else
+        {
+            removeId(realErrors, entryId);
         }
         entries.erase(entry);
     }
@@ -238,13 +242,6 @@ void Manager::erase(uint32_t entryId)
         logging::log<level::ERR>("Invalid entry ID to delete",
                 logging::entry("ID=%d", entryId));
     }
-
-    size_t realErrCnt = entries.size() - infoErrors.size();
-
-    if (realErrCnt <  ERROR_CAP)
-    {
-        capped = false;
-    }
 }
 
 void Manager::restore()
@@ -280,6 +277,10 @@ void Manager::restore()
                 {
                     infoErrors.push_back(idNum);
                 }
+                else
+                {
+                    realErrors.push_back(idNum);
+                }
 
                 entries.insert(std::make_pair(idNum, std::move(e)));
                 errorIds.push_back(idNum);
diff --git a/log_manager.hpp b/log_manager.hpp
index 399a243..b165ff3 100644
--- a/log_manager.hpp
+++ b/log_manager.hpp
@@ -54,8 +54,7 @@ class Manager : public details::ServerObject<details::ManagerIface>
         Manager(sdbusplus::bus::bus& bus, const char* objPath) :
                 details::ServerObject<details::ManagerIface>(bus, objPath),
                 busLog(bus),
-                entryId(0),
-                capped(false) {};
+                entryId(0){};
 
         /*
          * @fn commit()
@@ -112,21 +111,14 @@ class Manager : public details::ServerObject<details::ManagerIface>
         /** @brief Persistent map of Entry dbus objects and their ID */
         std::map<uint32_t, std::unique_ptr<Entry>> entries;
 
+        /** @brief List of error ids for high severity errors */
+        std::list<uint32_t> realErrors;
+
         /** @brief List of error ids for Info(and below) severity */
         std::list<uint32_t> infoErrors;
 
         /** @brief Id of last error log entry */
         uint32_t entryId;
-
-        /**
-         * @brief Flag to log error for the first time when error cap is
-         *      reached.
-         * @details Flag used to log error message for the first time when the
-         *      error cap value is reached. It is reset when user delete's error
-         *      entries and total entries existing is less than the error cap
-         *      value.
-         */
-        bool capped;
 };
 
 } //namespace internal