Fix memory issue, launch load tablets with different ids on different nodes, KIKIMR-20348

Fix memory issue, launch load tablets with different ids on different nodes

Launch tablets with different ids on different nodes

4 files changed, 57 insertions, 10 deletions

diff --git a/ydb/core/load_test/events.h b/ydb/core/load_test/events.h
index d124d3fc01..86e712916e 100644
--- a/ydb/core/load_test/events.h
+++ b/ydb/core/load_test/events.h
@@ -6,6 +6,8 @@
 #include <library/cpp/monlib/dynamic_counters/percentile/percentile_lg.h>
 #include <library/cpp/json/writer/json_value.h>
 
+#include <google/protobuf/text_format.h>
+
 namespace NKikimr {
 struct TEvLoad {
     enum EEv {
@@ -91,7 +93,13 @@ struct TEvLoad {
 
     struct TEvNodeFinishResponse : public TEventPB<TEvNodeFinishResponse,
         NKikimr::TEvNodeFinishResponse, EvNodeFinishResponse>
-    {};
+    {
+        TString ToString() const {
+            TString str;
+            google::protobuf::TextFormat::PrintToString(Record, &str);
+            return str;
+        }
+    };
 
     struct TEvYqlSingleQueryResponse : public TEventLocal<TEvYqlSingleQueryResponse, TEvLoad::EvYqlSingleQueryResponse> {
         TString Result;  // empty in case if there is an error
diff --git a/ydb/core/load_test/group_write.cpp b/ydb/core/load_test/group_write.cpp
index 8592522e06..59921cef21 100644
--- a/ydb/core/load_test/group_write.cpp
+++ b/ydb/core/load_test/group_write.cpp
@@ -125,9 +125,9 @@ class TLogWriterLoadTestActor : public TActorBootstrapped<TLogWriterLoadTestActo
             , RequestRateAtStart((EpochDuration / TDuration::Seconds(1)) * requestsPerSecondAtStart)
             , RequestRateOnFinish((EpochDuration / TDuration::Seconds(1)) * requestsPerSecondOnFinish)
             , CurrentEpochEnd(now)
-            , PlannedForCurrentEpoch(std::max(1., CalculateRequestRate(now)))
             , LoadStart(now)
-            , LoadDuration(duration) {
+            , LoadDuration(duration)
+            , PlannedForCurrentEpoch(std::max(1., CalculateRequestRate(now))) {
             CalculateDelayForNextRequest(now);
         }
 
@@ -142,12 +142,12 @@ class TLogWriterLoadTestActor : public TActorBootstrapped<TLogWriterLoadTestActo
         TDuration CurrentDelay = TDuration::Seconds(0);
         TMonotonic Now = TMonotonic::Max();
 
-        double PlannedForCurrentEpoch;
-        ui32 ResponsesAwaiting = 0;
-
         const TMonotonic LoadStart;
         const TMaybe<TDuration> LoadDuration;
 
+        double PlannedForCurrentEpoch;
+        ui32 ResponsesAwaiting = 0;
+
         double CalculateRequestRate(TMonotonic now) {
             double ratio = LoadDuration ? (now - LoadStart) / *LoadDuration : 0;
             return ratio * (RequestRateOnFinish - RequestRateAtStart) + RequestRateAtStart;
@@ -887,6 +887,8 @@ class TLogWriterLoadTestActor : public TActorBootstrapped<TLogWriterLoadTestActo
     TMonotonic LastScheduleTime = TMonotonic::Max();
     TMonotonic LastWakeupTime = TMonotonic::Max();
 
+    static constexpr ui64 DefaultTabletId = 5000;
+
 public:
     static constexpr NKikimrServices::TActivity::EType ActorActivityType() {
         return NKikimrServices::TActivity::BS_LOAD_ACTOR;
@@ -902,6 +904,8 @@ public:
         if (cmd.HasDurationSeconds()) {
             TestDuration = TDuration::Seconds(cmd.GetDurationSeconds());
         }
+
+        std::unordered_map<TString, ui64> tabletIds;
         for (const auto& profile : cmd.GetTablets()) {
             if (!profile.TabletsSize()) {
                 ythrow TLoadActorException() << "TPerTabletProfile.Tablets must have at least one item";
@@ -978,10 +982,33 @@ public:
                             });
                 }
 
-                if (!tablet.HasTabletId() || !tablet.HasChannel() || !tablet.HasGroupId()) {
-                    ythrow TLoadActorException() << "TTabletInfo.{TabletId,Channel,GroupId} fields are mandatory";
+                if (!tablet.HasChannel() || !tablet.HasGroupId()) {
+                    ythrow TLoadActorException() << "TTabletInfo.{Channel,GroupId} fields are mandatory";
                 }
-                TabletWriters.emplace_back(std::make_unique<TTabletWriter>(Tag, counters, WakeupQueue, QueryDispatcher, tablet.GetTabletId(),
+                if (!tablet.HasTabletName() && !tablet.HasTabletId()) {
+                    ythrow TLoadActorException() << "One of TTabletInfo.{TabletName,TabletId} fields must be specified";
+                }
+
+                ui64 tabletId;
+                if (tablet.HasTabletId()) {
+                    tabletId = tablet.GetTabletId();
+                } else if (tablet.HasTabletName()) {
+                    TString name = tablet.GetTabletName();
+                    auto it = tabletIds.find(name);
+                    if (it != tabletIds.end()) {
+                        tabletId = it->second;
+                    } else {
+                        tabletId = THash<TString>{}(name) & ((1 << 20) - 1);
+                        tabletId = tabletId << 10 + tag;
+                        tabletId = tabletId << 10 + Parent.NodeId();
+                        tabletId = MakeTabletID(0, 0, tabletId);
+                        tabletIds[name] = tabletId;
+                    }
+                } else {
+                    Y_FAIL();
+                }
+
+                TabletWriters.emplace_back(std::make_unique<TTabletWriter>(Tag, counters, WakeupQueue, QueryDispatcher, tabletId,
                     tablet.GetChannel(), tablet.HasGeneration() ?  TMaybe<ui32>(tablet.GetGeneration()) : TMaybe<ui32>(),
                     tablet.GetGroupId(), putHandleClass, writeSettings,
                     getHandleClass, readSettings,
diff --git a/ydb/core/load_test/service_actor.cpp b/ydb/core/load_test/service_actor.cpp
index bd937d5037..07fa8c1e5f 100644
--- a/ydb/core/load_test/service_actor.cpp
+++ b/ydb/core/load_test/service_actor.cpp
@@ -461,6 +461,14 @@ public:
         Send(ev->Sender, response.release());
     }
 
+    void Handle(TEvLoad::TEvLoadTestResponse::TPtr& ev) {
+        if (ev->Get()->Record.GetStatus() != NMsgBusProxy::MSTATUS_OK) {
+            LOG_E("Receieved non-OK LoadTestResponse from another node, Record# " << ev->ToString());
+        } else {
+            LOG_N("Receieved OK LoadTestResponse from another node# " << ev->ToString());
+        }
+    }
+
     ui64 ProcessCmd(const NKikimr::TEvLoadTestRequest& record) {
         ui64 tag = 0;
         if (record.Command_case() != TEvLoadTestRequest::CommandCase::kStop) {
@@ -1317,6 +1325,7 @@ public:
 
     STRICT_STFUNC(StateFunc,
         hFunc(TEvLoad::TEvLoadTestRequest, Handle)
+        hFunc(TEvLoad::TEvLoadTestResponse, Handle)
         hFunc(TEvLoad::TEvLoadTestFinished, Handle)
         hFunc(TEvLoad::TEvNodeFinishResponse, Handle)
         hFunc(NMon::TEvHttpInfo, Handle)
diff --git a/ydb/core/protos/load_test.proto b/ydb/core/protos/load_test.proto
index 6dad8c06ed..3bc3bb84b7 100644
--- a/ydb/core/protos/load_test.proto
+++ b/ydb/core/protos/load_test.proto
@@ -39,7 +39,10 @@ message TEvLoadTestRequest {
             optional NKikimrBlobStorage.EPutHandleClass PutHandleClass = 4;
         }
         message TTabletInfo {
-            optional uint64 TabletId = 1;
+            oneof Id {
+                uint64 TabletId = 8;
+                string TabletName = 7;  // only used to distinguish different tablets, will be reassigned
+            }
             optional uint32 Channel = 2;
             optional uint32 GroupId = 3;
             optional uint32 Generation = 4;