[stable-25-1] Make reconnect to kesus work again (#18496)

Author: Enjection

ydb/core/quoter/kesus_quoter_proxy.cpp

@@ -289,6 +289,7 @@ class TKesusQuoterProxy : public TActorBootstrapped<TKesusQuoterProxy> {
     bool Connected = false;
     TInstant DisconnectTime;
     ui64 OfflineAllocationCookie = 0;
+    ui64 KesusReconnectCount = 0;
 
     TMap<TString, THolder<TResourceState>> Resources; // Map because iterators are needed to remain valid during insertions.
     THashMap<ui64, decltype(Resources)::iterator> ResIndex;
@@ -803,7 +804,11 @@ class TKesusQuoterProxy : public TActorBootstrapped<TKesusQuoterProxy> {
                 SendToService(CreateUpdateEvent(TEvQuota::EUpdateState::Broken));
             } else {
                 KESUS_PROXY_LOG_WARN("Failed to connect to tablet. Status: " << ev->Get()->Status);
-                ConnectToKesus(true);
+                if (!ConnectToKesus(true)) {
+                    KESUS_PROXY_LOG_WARN("Too many reconnect attempts, assuming kesus dead");
+                    SendToService(CreateUpdateEvent(TEvQuota::EUpdateState::Broken));
+                    KesusReconnectCount = 0;
+                }
             }
         }
     }
@@ -1119,7 +1124,6 @@ class TKesusQuoterProxy : public TActorBootstrapped<TKesusQuoterProxy> {
 
     NTabletPipe::TClientConfig GetPipeConnectionOptions(bool reconnection) {
         NTabletPipe::TClientConfig cfg;
-        cfg.CheckAliveness = true;
         cfg.RetryPolicy = {
             .RetryLimitCount = 3u,
             .DoFirstRetryInstantly = !reconnection
@@ -1134,9 +1138,13 @@ class TKesusQuoterProxy : public TActorBootstrapped<TKesusQuoterProxy> {
         CookieToResourcePath.clear(); // we will resend all requests with new cookies
     }
 
-    void ConnectToKesus(bool reconnection) {
+    bool ConnectToKesus(bool reconnection) {
         if (reconnection) {
             KESUS_PROXY_LOG_INFO("Reconnecting to kesus");
+            ++KesusReconnectCount;
+            if (KesusReconnectCount > 5) {
+                return false;
+            }
         } else {
             KESUS_PROXY_LOG_DEBUG("Connecting to kesus");
         }
@@ -1149,6 +1157,7 @@ class TKesusQuoterProxy : public TActorBootstrapped<TKesusQuoterProxy> {
                     GetKesusTabletId(),
                     GetPipeConnectionOptions(reconnection)));
         Connected = false;
+        return true;
     }
 
     void PassAway() override {

ydb/core/quoter/ut_helpers.cpp

@@ -41,6 +41,7 @@ void TKesusQuoterTestSetup::SetupLogging() {
     Server->GetRuntime()->SetLogPriority(NKikimrServices::KESUS_TABLET, NActors::NLog::PRI_TRACE);
     Server->GetRuntime()->SetLogPriority(NKikimrServices::QUOTER_SERVICE, NActors::NLog::PRI_TRACE);
     Server->GetRuntime()->SetLogPriority(NKikimrServices::QUOTER_PROXY, NActors::NLog::PRI_TRACE);
+    Server->GetRuntime()->SetLogPriority(NKikimrServices::PIPE_CLIENT, NActors::NLog::PRI_TRACE);
 }
 
 void TKesusQuoterTestSetup::RegisterQuoterService() {