Make retry settings configurable

16 files changed, 235 insertions, 19 deletions

diff --git a/library/cpp/threading/CMakeLists.txt b/library/cpp/threading/CMakeLists.txt
index b42430ff77c..681ef6b24e3 100644
--- a/library/cpp/threading/CMakeLists.txt
+++ b/library/cpp/threading/CMakeLists.txt
@@ -10,6 +10,7 @@ add_subdirectory(atomic)
 add_subdirectory(chunk_queue)
 add_subdirectory(equeue)
 add_subdirectory(future)
+add_subdirectory(hot_swap)
 add_subdirectory(light_rw_lock)
 add_subdirectory(local_executor)
 add_subdirectory(poor_man_openmp)
diff --git a/library/cpp/threading/hot_swap/CMakeLists.darwin-x86_64.txt b/library/cpp/threading/hot_swap/CMakeLists.darwin-x86_64.txt
new file mode 100644
index 00000000000..7ec6e7f8b0d
--- /dev/null
+++ b/library/cpp/threading/hot_swap/CMakeLists.darwin-x86_64.txt
@@ -0,0 +1,18 @@
+
+# This file was generated by the build system used internally in the Yandex monorepo.
+# Only simple modifications are allowed (adding source-files to targets, adding simple properties
+# like target_include_directories). These modifications will be ported to original
+# ya.make files by maintainers. Any complex modifications which can't be ported back to the
+# original buildsystem will not be accepted.
+
+
+
+add_library(cpp-threading-hot_swap)
+target_link_libraries(cpp-threading-hot_swap PUBLIC
+  contrib-libs-cxxsupp
+  yutil
+  cpp-deprecated-atomic
+)
+target_sources(cpp-threading-hot_swap PRIVATE
+  ${CMAKE_SOURCE_DIR}/library/cpp/threading/hot_swap/hot_swap.cpp
+)
diff --git a/library/cpp/threading/hot_swap/CMakeLists.linux-aarch64.txt b/library/cpp/threading/hot_swap/CMakeLists.linux-aarch64.txt
new file mode 100644
index 00000000000..af861b6b248
--- /dev/null
+++ b/library/cpp/threading/hot_swap/CMakeLists.linux-aarch64.txt
@@ -0,0 +1,19 @@
+
+# This file was generated by the build system used internally in the Yandex monorepo.
+# Only simple modifications are allowed (adding source-files to targets, adding simple properties
+# like target_include_directories). These modifications will be ported to original
+# ya.make files by maintainers. Any complex modifications which can't be ported back to the
+# original buildsystem will not be accepted.
+
+
+
+add_library(cpp-threading-hot_swap)
+target_link_libraries(cpp-threading-hot_swap PUBLIC
+  contrib-libs-linux-headers
+  contrib-libs-cxxsupp
+  yutil
+  cpp-deprecated-atomic
+)
+target_sources(cpp-threading-hot_swap PRIVATE
+  ${CMAKE_SOURCE_DIR}/library/cpp/threading/hot_swap/hot_swap.cpp
+)
diff --git a/library/cpp/threading/hot_swap/CMakeLists.linux-x86_64.txt b/library/cpp/threading/hot_swap/CMakeLists.linux-x86_64.txt
new file mode 100644
index 00000000000..af861b6b248
--- /dev/null
+++ b/library/cpp/threading/hot_swap/CMakeLists.linux-x86_64.txt
@@ -0,0 +1,19 @@
+
+# This file was generated by the build system used internally in the Yandex monorepo.
+# Only simple modifications are allowed (adding source-files to targets, adding simple properties
+# like target_include_directories). These modifications will be ported to original
+# ya.make files by maintainers. Any complex modifications which can't be ported back to the
+# original buildsystem will not be accepted.
+
+
+
+add_library(cpp-threading-hot_swap)
+target_link_libraries(cpp-threading-hot_swap PUBLIC
+  contrib-libs-linux-headers
+  contrib-libs-cxxsupp
+  yutil
+  cpp-deprecated-atomic
+)
+target_sources(cpp-threading-hot_swap PRIVATE
+  ${CMAKE_SOURCE_DIR}/library/cpp/threading/hot_swap/hot_swap.cpp
+)
diff --git a/library/cpp/threading/hot_swap/CMakeLists.txt b/library/cpp/threading/hot_swap/CMakeLists.txt
new file mode 100644
index 00000000000..f8b31df0c11
--- /dev/null
+++ b/library/cpp/threading/hot_swap/CMakeLists.txt
@@ -0,0 +1,17 @@
+
+# This file was generated by the build system used internally in the Yandex monorepo.
+# Only simple modifications are allowed (adding source-files to targets, adding simple properties
+# like target_include_directories). These modifications will be ported to original
+# ya.make files by maintainers. Any complex modifications which can't be ported back to the
+# original buildsystem will not be accepted.
+
+
+if (CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" AND NOT HAVE_CUDA)
+  include(CMakeLists.linux-aarch64.txt)
+elseif (CMAKE_SYSTEM_NAME STREQUAL "Darwin" AND CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64")
+  include(CMakeLists.darwin-x86_64.txt)
+elseif (WIN32 AND CMAKE_SYSTEM_PROCESSOR STREQUAL "AMD64" AND NOT HAVE_CUDA)
+  include(CMakeLists.windows-x86_64.txt)
+elseif (CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" AND NOT HAVE_CUDA)
+  include(CMakeLists.linux-x86_64.txt)
+endif()
diff --git a/library/cpp/threading/hot_swap/CMakeLists.windows-x86_64.txt b/library/cpp/threading/hot_swap/CMakeLists.windows-x86_64.txt
new file mode 100644
index 00000000000..7ec6e7f8b0d
--- /dev/null
+++ b/library/cpp/threading/hot_swap/CMakeLists.windows-x86_64.txt
@@ -0,0 +1,18 @@
+
+# This file was generated by the build system used internally in the Yandex monorepo.
+# Only simple modifications are allowed (adding source-files to targets, adding simple properties
+# like target_include_directories). These modifications will be ported to original
+# ya.make files by maintainers. Any complex modifications which can't be ported back to the
+# original buildsystem will not be accepted.
+
+
+
+add_library(cpp-threading-hot_swap)
+target_link_libraries(cpp-threading-hot_swap PUBLIC
+  contrib-libs-cxxsupp
+  yutil
+  cpp-deprecated-atomic
+)
+target_sources(cpp-threading-hot_swap PRIVATE
+  ${CMAKE_SOURCE_DIR}/library/cpp/threading/hot_swap/hot_swap.cpp
+)
diff --git a/ydb/core/kqp/node_service/kqp_node_service.cpp b/ydb/core/kqp/node_service/kqp_node_service.cpp
index 4843b7ab228..6e70cfdfc50 100644
--- a/ydb/core/kqp/node_service/kqp_node_service.cpp
+++ b/ydb/core/kqp/node_service/kqp_node_service.cpp
@@ -11,6 +11,7 @@
 #include <ydb/core/kqp/compute_actor/kqp_compute_actor.h>
 #include <ydb/core/kqp/rm_service/kqp_resource_estimation.h>
 #include <ydb/core/kqp/rm_service/kqp_rm_service.h>
+#include <ydb/core/kqp/runtime/kqp_read_actor.h>
 #include <ydb/core/kqp/common/kqp_resolve.h>
 
 #include <ydb/core/base/wilson.h>
@@ -75,7 +76,7 @@ struct TMemoryQuotaManager : public NYql::NDq::TGuaranteeQuotaManager {
         , ui64 txId
         , ui64 taskId
         , ui64 limit
-        , bool instantAlloc) 
+        , bool instantAlloc)
     : NYql::NDq::TGuaranteeQuotaManager(limit, limit)
     , ResourceManager(std::move(resourceManager))
     , MemoryPool(memoryPool)
@@ -96,7 +97,7 @@ struct TMemoryQuotaManager : public NYql::NDq::TGuaranteeQuotaManager {
             return false;
         }
 
-        if (!ResourceManager->AllocateResources(TxId, TaskId, 
+        if (!ResourceManager->AllocateResources(TxId, TaskId,
                 NRm::TKqpResourcesRequest{.MemoryPool = MemoryPool, .Memory = extraSize})) {
             LOG_W("Can not allocate memory. TxId: " << TxId << ", taskId: " << TaskId << ", memory: +" << extraSize);
             return false;
@@ -106,7 +107,7 @@ struct TMemoryQuotaManager : public NYql::NDq::TGuaranteeQuotaManager {
     }
 
     void FreeExtraQuota(ui64 extraSize) override {
-        ResourceManager->FreeResources(TxId, TaskId, 
+        ResourceManager->FreeResources(TxId, TaskId,
             NRm::TKqpResourcesRequest{.MemoryPool = MemoryPool, .Memory = extraSize}
         );
     }
@@ -142,6 +143,9 @@ public:
         , AsyncIoFactory(std::move(asyncIoFactory))
     {
         Buckets = std::make_shared<TBucketArray>();
+        if (config.HasIteratorReadsRetrySettings()) {
+            SetIteratorReadsRetrySettings(config.GetIteratorReadsRetrySettings());
+        }
     }
 
     void Bootstrap() {
@@ -578,10 +582,31 @@ private:
             LOG_I("Updated table service config: " << Config.DebugString());
         }
 
+        if (event.GetConfig().GetTableServiceConfig().HasIteratorReadsRetrySettings()) {
+            SetIteratorReadsRetrySettings(event.GetConfig().GetTableServiceConfig().GetIteratorReadsRetrySettings());
+        }
+
         auto responseEv = MakeHolder<NConsole::TEvConsole::TEvConfigNotificationResponse>(event);
         Send(ev->Sender, responseEv.Release(), IEventHandle::FlagTrackDelivery, ev->Cookie);
     }
 
+    void SetIteratorReadsRetrySettings(const NKikimrConfig::TTableServiceConfig::TIteratorReadsRetrySettings& settings) {
+        auto ptr = MakeIntrusive<NKikimr::NKqp::TIteratorReadBackoffSettings>();
+        ptr->StartRetryDelay = TDuration::MilliSeconds(settings.GetStartDelayMs());
+        ptr->MaxShardAttempts = settings.GetMaxShardRetries();
+        ptr->MaxShardResolves = settings.GetMaxShardResolves();
+        ptr->UnsertaintyRatio = settings.GetUnsertaintyRatio();
+        ptr->Multiplier = settings.GetMultiplier();
+        if (settings.GetMaxTotalRetries()) {
+            ptr->MaxTotalRetries = settings.GetMaxTotalRetries();
+        }
+        if (settings.GetIteratorResponseTimeoutMs()) {
+            ptr->ReadResponseTimeout = TDuration::MilliSeconds(settings.GetIteratorResponseTimeoutMs());
+        }
+        ptr->MaxRetryDelay = TDuration::MilliSeconds(settings.GetMaxDelayMs());
+        SetReadIteratorBackoffSettings(ptr);
+    }
+
     void HandleWork(TEvents::TEvUndelivered::TPtr& ev) {
         switch (ev->Get()->SourceType) {
             case TEvKqpNode::TEvStartKqpTasksResponse::EventType: {
diff --git a/ydb/core/kqp/runtime/CMakeLists.darwin-x86_64.txt b/ydb/core/kqp/runtime/CMakeLists.darwin-x86_64.txt
index 02e0d962816..78dfb6918e4 100644
--- a/ydb/core/kqp/runtime/CMakeLists.darwin-x86_64.txt
+++ b/ydb/core/kqp/runtime/CMakeLists.darwin-x86_64.txt
@@ -36,6 +36,7 @@ target_link_libraries(core-kqp-runtime PUBLIC
   dq-actors-protos
   yql-dq-common
   yql-dq-runtime
+  cpp-threading-hot_swap
   tools-enum_parser-enum_serialization_runtime
 )
 target_sources(core-kqp-runtime PRIVATE
diff --git a/ydb/core/kqp/runtime/CMakeLists.linux-aarch64.txt b/ydb/core/kqp/runtime/CMakeLists.linux-aarch64.txt
index 9962998778a..94a3bb9722b 100644
--- a/ydb/core/kqp/runtime/CMakeLists.linux-aarch64.txt
+++ b/ydb/core/kqp/runtime/CMakeLists.linux-aarch64.txt
@@ -37,6 +37,7 @@ target_link_libraries(core-kqp-runtime PUBLIC
   dq-actors-protos
   yql-dq-common
   yql-dq-runtime
+  cpp-threading-hot_swap
   tools-enum_parser-enum_serialization_runtime
 )
 target_sources(core-kqp-runtime PRIVATE
diff --git a/ydb/core/kqp/runtime/CMakeLists.linux-x86_64.txt b/ydb/core/kqp/runtime/CMakeLists.linux-x86_64.txt
index 9962998778a..94a3bb9722b 100644
--- a/ydb/core/kqp/runtime/CMakeLists.linux-x86_64.txt
+++ b/ydb/core/kqp/runtime/CMakeLists.linux-x86_64.txt
@@ -37,6 +37,7 @@ target_link_libraries(core-kqp-runtime PUBLIC
   dq-actors-protos
   yql-dq-common
   yql-dq-runtime
+  cpp-threading-hot_swap
   tools-enum_parser-enum_serialization_runtime
 )
 target_sources(core-kqp-runtime PRIVATE
diff --git a/ydb/core/kqp/runtime/CMakeLists.windows-x86_64.txt b/ydb/core/kqp/runtime/CMakeLists.windows-x86_64.txt
index 02e0d962816..78dfb6918e4 100644
--- a/ydb/core/kqp/runtime/CMakeLists.windows-x86_64.txt
+++ b/ydb/core/kqp/runtime/CMakeLists.windows-x86_64.txt
@@ -36,6 +36,7 @@ target_link_libraries(core-kqp-runtime PUBLIC
   dq-actors-protos
   yql-dq-common
   yql-dq-runtime
+  cpp-threading-hot_swap
   tools-enum_parser-enum_serialization_runtime
 )
 target_sources(core-kqp-runtime PRIVATE
diff --git a/ydb/core/kqp/runtime/kqp_read_actor.cpp b/ydb/core/kqp/runtime/kqp_read_actor.cpp
index 4f9a32475c9..3267aa6d45c 100644
--- a/ydb/core/kqp/runtime/kqp_read_actor.cpp
+++ b/ydb/core/kqp/runtime/kqp_read_actor.cpp
@@ -14,6 +14,7 @@
 
 #include <ydb/library/yql/dq/actors/compute/dq_compute_actor_impl.h>
 
+#include <library/cpp/threading/hot_swap/hot_swap.h>
 #include <library/cpp/actors/core/interconnect.h>
 #include <library/cpp/actors/core/actorsystem.h>
 
@@ -21,9 +22,6 @@
 
 namespace {
 
-static constexpr ui64 MAX_SHARD_RETRIES = 5;
-static constexpr ui64 MAX_SHARD_RESOLVES = 3;
-
 bool IsDebugLogEnabled(const NActors::TActorSystem* actorSystem, NActors::NLog::EComponent component) {
     auto* settings = actorSystem->LoggerSettings();
     return settings && settings->Satisfies(NActors::NLog::EPriority::PRI_DEBUG, component);
@@ -64,7 +62,33 @@ THolder<NKikimr::TEvDataShard::TEvReadAck> DefaultAckSettings() {
 NActors::TActorId MainPipeCacheId = NKikimr::MakePipePeNodeCacheID(false);
 NActors::TActorId FollowersPipeCacheId =  NKikimr::MakePipePeNodeCacheID(true);
 
-TDuration StartRetryDelay = TDuration::MilliSeconds(250);
+struct TBackoffStorage {
+    THotSwap<NKikimr::NKqp::TIteratorReadBackoffSettings> SettingsPtr;
+
+    TBackoffStorage() {
+        SettingsPtr.AtomicStore(new NKikimr::NKqp::TIteratorReadBackoffSettings());
+    }
+};
+
+TDuration CalcDelay(size_t attempt, bool allowInstantRetry) {
+    return Singleton<::TBackoffStorage>()->SettingsPtr.AtomicLoad()->CalcShardDelay(attempt, allowInstantRetry);
+}
+
+size_t MaxShardResolves() {
+    return Singleton<::TBackoffStorage>()->SettingsPtr.AtomicLoad()->MaxShardResolves;
+}
+
+size_t MaxShardRetries() {
+    return Singleton<::TBackoffStorage>()->SettingsPtr.AtomicLoad()->MaxShardAttempts;
+}
+
+TMaybe<size_t> MaxTotalRetries() {
+    return Singleton<::TBackoffStorage>()->SettingsPtr.AtomicLoad()->MaxTotalRetries;
+}
+
+TMaybe<TDuration> ShardTimeout() {
+    return Singleton<::TBackoffStorage>()->SettingsPtr.AtomicLoad()->ReadResponseTimeout;
+}
 
 }
 
@@ -446,6 +470,7 @@ public:
                     state.AddRange(TSerializedTableRange(range));
                 }
             } else {
+                YQL_ENSURE(Settings.GetRanges().KeyPointsSize() > 0);
                 for (const auto& point : Settings.GetRanges().GetKeyPoints()) {
                     state.AddPoint(TSerializedCellVec(point));
                 }
@@ -491,7 +516,7 @@ public:
     }
 
     void ResolveShard(TShardState* state) {
-        if (state->ResolveAttempt >= MAX_SHARD_RESOLVES) {
+        if (state->ResolveAttempt >= ::MaxShardResolves()) {
             RuntimeError(TStringBuilder() << "Table '" << Settings.GetTable().GetTablePath() << "' resolve limit exceeded",
                 NDqProto::StatusIds::UNAVAILABLE);
             return;
@@ -730,12 +755,23 @@ public:
         }
 
         auto state = Reads[id].Shard;
-        if (state->RetryAttempt == 0 && allowInstantRetry) { // instant retry
-            return DoRetryRead(id);
+
+        TotalRetries += 1;
+        auto limit = ::MaxTotalRetries();
+        if (limit && TotalRetries > *limit) {
+            return RuntimeError(TStringBuilder() << "Table '" << Settings.GetTable().GetTablePath() << "' retry limit exceeded",
+                NDqProto::StatusIds::UNAVAILABLE);
+        }
+
+        state->RetryAttempt += 1;
+        if (state->RetryAttempt > ::MaxShardRetries()) {
+            ResetRead(id);
+            return ResolveShard(state);
         }
-        auto delay = ::StartRetryDelay;
-        for (size_t i = 0; i < state->RetryAttempt; ++i) {
-            delay *= 2;
+
+        auto delay = ::CalcDelay(state->RetryAttempt, allowInstantRetry);
+        if (delay == TDuration::Zero()) {
+            return DoRetryRead(id);
         }
 
         CA_LOG_D("schedule retry #" << id << " after " << delay);
@@ -748,12 +784,6 @@ public:
         }
 
         auto state = Reads[id].Shard;
-
-        state->RetryAttempt += 1;
-        if (state->RetryAttempt >= MAX_SHARD_RETRIES) {
-            ResetRead(id);
-            return ResolveShard(state);
-        }
         CA_LOG_D("Retrying read #" << id);
 
         ResetRead(id);
@@ -843,6 +873,10 @@ public:
         ReadIdByTabletId[state->TabletId].push_back(id);
         Send(PipeCacheId, new TEvPipeCache::TEvForward(ev.Release(), state->TabletId, true),
             IEventHandle::FlagTrackDelivery);
+
+        if (auto delay = ShardTimeout()) {
+            TlsActivationContext->Schedule(*delay, new IEventHandle(SelfId(), SelfId(), new TEvRetryShard(id, Reads[id].LastSeqNo)));
+        }
     }
 
     void NotifyCA() {
@@ -1176,6 +1210,10 @@ public:
                         CA_LOG_D("sending ack for read #" << id << " limit " << limit << " seqno = " << record.GetSeqNo());
                         Send(PipeCacheId, new TEvPipeCache::TEvForward(request.Release(), Reads[id].Shard->TabletId, true),
                             IEventHandle::FlagTrackDelivery);
+
+                        if (auto delay = ShardTimeout()) {
+                            TlsActivationContext->Schedule(*delay, new IEventHandle(SelfId(), SelfId(), new TEvRetryShard(id, Reads[id].LastSeqNo)));
+                        }
                     } else {
                         Reads[id].Finished = true;
                     }
@@ -1361,6 +1399,8 @@ private:
     TIntrusivePtr<TKqpCounters> Counters;
     bool UseFollowers;
     NActors::TActorId PipeCacheId;
+
+    size_t TotalRetries = 0;
 };
 
 
@@ -1385,5 +1425,9 @@ void InterceptReadActorPipeCache(NActors::TActorId id) {
     ::MainPipeCacheId = id;
 }
 
+void SetReadIteratorBackoffSettings(TIntrusivePtr<TIteratorReadBackoffSettings> ptr) {
+    Singleton<::TBackoffStorage>()->SettingsPtr.AtomicStore(ptr);
+}
+
 } // namespace NKqp
 } // namespace NKikimr
diff --git a/ydb/core/kqp/runtime/kqp_read_actor.h b/ydb/core/kqp/runtime/kqp_read_actor.h
index 22c4e05d5c2..12da2c844c6 100644
--- a/ydb/core/kqp/runtime/kqp_read_actor.h
+++ b/ydb/core/kqp/runtime/kqp_read_actor.h
@@ -12,6 +12,36 @@ class TEvReadAck;
 namespace NKikimr {
 namespace NKqp {
 
+struct TIteratorReadBackoffSettings : TAtomicRefCount<TIteratorReadBackoffSettings> {
+    TDuration StartRetryDelay = TDuration::MilliSeconds(5);
+    size_t MaxShardAttempts = 10;
+    size_t MaxShardResolves = 3;
+    double UnsertaintyRatio = 0.5;
+    double Multiplier = 2.0;
+    TDuration MaxRetryDelay = TDuration::Seconds(1);
+
+    TMaybe<size_t> MaxTotalRetries;
+    TMaybe<TDuration> ReadResponseTimeout;
+
+    TDuration CalcShardDelay(size_t attempt, bool allowInstantRetry) {
+        if (allowInstantRetry && attempt == 1) {
+            return TDuration::Zero();
+        }
+
+        auto delay = StartRetryDelay;
+        for (size_t i = 0; i < attempt; ++i) {
+            delay *= Multiplier;
+            delay = Min(delay, MaxRetryDelay);
+        }
+
+        delay *= (1 - UnsertaintyRatio * RandomNumber<double>());
+
+        return delay;
+    }
+};
+
+void SetReadIteratorBackoffSettings(TIntrusivePtr<TIteratorReadBackoffSettings>);
+
 void RegisterKqpReadActor(NYql::NDq::TDqAsyncIoFactory&, TIntrusivePtr<TKqpCounters>);
 
 void InjectRangeEvReadSettings(const NKikimrTxDataShard::TEvRead&);
diff --git a/ydb/core/kqp/runtime/ya.make b/ydb/core/kqp/runtime/ya.make
index 71c0fde7911..5922f0ef0d4 100644
--- a/ydb/core/kqp/runtime/ya.make
+++ b/ydb/core/kqp/runtime/ya.make
@@ -36,6 +36,7 @@ PEERDIR(
     ydb/library/yql/dq/actors/protos
     ydb/library/yql/dq/common
     ydb/library/yql/dq/runtime
+    library/cpp/threading/hot_swap
 )
 
 YQL_LAST_ABI_VERSION()
diff --git a/ydb/core/kqp/ut/scan/kqp_split_ut.cpp b/ydb/core/kqp/ut/scan/kqp_split_ut.cpp
index 8368a9dd499..a04b3aaf2b9 100644
--- a/ydb/core/kqp/ut/scan/kqp_split_ut.cpp
+++ b/ydb/core/kqp/ut/scan/kqp_split_ut.cpp
@@ -362,6 +362,13 @@ Y_UNIT_TEST_SUITE(KqpSplit) {
             Runtime = Server->GetRuntime();
             KqpProxy = MakeKqpProxyID(Runtime->GetNodeId(0));
 
+            {
+                auto settings = MakeIntrusive<TIteratorReadBackoffSettings>();
+                settings->StartRetryDelay = TDuration::MilliSeconds(250);
+                settings->MaxShardAttempts = 4;
+                SetReadIteratorBackoffSettings(settings);
+            }
+
             Sender = Runtime->AllocateEdgeActor();
 
             CollectKeysTo(&CollectedKeys, Runtime, Sender);
diff --git a/ydb/core/protos/config.proto b/ydb/core/protos/config.proto
index dcd6d3d1c2e..9689aed58dc 100644
--- a/ydb/core/protos/config.proto
+++ b/ydb/core/protos/config.proto
@@ -1256,6 +1256,18 @@ message TTableServiceConfig {
         optional uint32 MaxRetryNumber = 3 [default = 10];
     }
 
+    message TIteratorReadsRetrySettings {
+        optional uint32 StartDelayMs = 1;
+        optional uint32 MaxDelayMs = 8;
+
+        optional uint32 MaxShardRetries = 2;
+        optional uint32 MaxShardResolves = 3;
+        optional double UnsertaintyRatio = 4;
+        optional double Multiplier = 5;
+        optional uint32 IteratorResponseTimeoutMs = 6;
+        optional uint32 MaxTotalRetries = 7;
+    }
+
     optional uint32 QueryLimitBytes = 1;
     optional uint32 ParametersLimitBytes = 2;
     optional uint32 SessionsLimitPerNode = 3;
@@ -1305,6 +1317,7 @@ message TTableServiceConfig {
     }
 
     optional EBindingsMode BindingsMode = 40 [default = BM_ENABLED];
+    optional TIteratorReadsRetrySettings IteratorReadsRetrySettings = 41;
 };
 
 // Config describes immediate controls and allows