Write all crit events to cerr (#696)

* Write all crit events to cerr * Make writing almost atomic * Use global logger
ydb-platform · Mar 28, 2024 · d3cd2bf · d3cd2bf
1 parent 8a3124b
commit d3cd2bf
Show file tree

Hide file tree

Showing 4 changed files with 20 additions and 3 deletions.
diff --git a/cloud/blockstore/libs/daemon/common/bootstrap.cpp b/cloud/blockstore/libs/daemon/common/bootstrap.cpp
@@ -199,6 +199,7 @@ void TBootstrapBase::Init()
 
     BootstrapLogging = CreateLoggingService("console", logSettings);
     Log = BootstrapLogging->CreateLog("BLOCKSTORE_SERVER");
+    SetCriticalEventsLog(Log);
     Configs->Log = Log;
     STORAGE_INFO("NBS server version: " << GetFullVersionString());
 

diff --git a/cloud/filestore/libs/daemon/common/bootstrap.cpp b/cloud/filestore/libs/daemon/common/bootstrap.cpp
@@ -11,13 +11,13 @@
 #include <cloud/filestore/libs/server/server.h>
 #include <cloud/filestore/libs/storage/core/config.h>
 #include <cloud/filestore/libs/storage/init/actorsystem.h>
-
 #include <cloud/storage/core/libs/common/scheduler.h>
 #include <cloud/storage/core/libs/common/task_queue.h>
 #include <cloud/storage/core/libs/common/thread_pool.h>
 #include <cloud/storage/core/libs/common/timer.h>
 #include <cloud/storage/core/libs/daemon/mlock.h>
 #include <cloud/storage/core/libs/diagnostics/cgroup_stats_fetcher.h>
+#include <cloud/storage/core/libs/diagnostics/critical_events.h>
 #include <cloud/storage/core/libs/diagnostics/logging.h>
 #include <cloud/storage/core/libs/diagnostics/monitoring.h>
 #include <cloud/storage/core/libs/diagnostics/stats_updater.h>
@@ -99,6 +99,7 @@ TBootstrapCommon::TBootstrapCommon(
     BootstrapLogging->Start();
 
     Log = BootstrapLogging->CreateLog(logComponent);
+    SetCriticalEventsLog(Log);
 }
 
 TBootstrapCommon::~TBootstrapCommon()

diff --git a/cloud/storage/core/libs/diagnostics/critical_events.cpp b/cloud/storage/core/libs/diagnostics/critical_events.cpp
@@ -1,7 +1,7 @@
 #include "critical_events.h"
-
 #include "public.h"
 
+#include <library/cpp/logger/log.h>
 #include <library/cpp/monlib/dynamic_counters/counters.h>
 
 #include <util/string/builder.h>
@@ -13,11 +13,17 @@ using namespace NMonitoring;
 namespace {
 
 NMonitoring::TDynamicCountersPtr CriticalEvents;
+TLog Log;
 
 }  // namespace
 
 ////////////////////////////////////////////////////////////////////////////////
 
+void SetCriticalEventsLog(TLog log)
+{
+    Log = std::move(log);
+}
+
 void InitCriticalEventsCounter(NMonitoring::TDynamicCountersPtr counters)
 {
     CriticalEvents = std::move(counters);
@@ -50,7 +56,15 @@ TString ReportCriticalEvent(
     fullMessage << "CRITICAL_EVENT:" << sensorName;
     if (message) {
         fullMessage << ":" << message;
-        Cerr << fullMessage << Endl;
+    }
+
+    if (Log.IsNotNullLog()) {
+        Log.AddLog("%s", fullMessage.c_str());
+    } else {
+        // Write message and \n in one call. This will reduce the chance of
+        // shuffling with writings of other threads.
+        Cerr << fullMessage + '\n';
+        Cerr.Flush();
     }
 
     return fullMessage;

diff --git a/cloud/storage/core/libs/diagnostics/critical_events.h b/cloud/storage/core/libs/diagnostics/critical_events.h
@@ -15,6 +15,7 @@ namespace NCloud {
 
 ////////////////////////////////////////////////////////////////////////////////
 
+void SetCriticalEventsLog(TLog log);
 void InitCriticalEventsCounter(NMonitoring::TDynamicCountersPtr counters);
 
 TString ReportCriticalEvent(