Add sensors to Node Broker (ydb-platform#8091)

pixcc · Aug 26, 2024 · 796027b · 796027b
1 parent 649471e
commit 796027b
Show file tree

Hide file tree

Showing 11 changed files with 111 additions and 9 deletions.
diff --git a/ydb/core/mind/node_broker.cpp b/ydb/core/mind/node_broker.cpp
@@ -7,7 +7,9 @@
 #include <ydb/core/base/nameservice.h>
 #include <ydb/core/base/path.h>
 #include <ydb/core/cms/console/config_helpers.h>
+#include <ydb/core/protos/counters_node_broker.pb.h>
 #include <ydb/core/protos/node_broker.pb.h>
+#include <ydb/core/tablet/tablet_counters_protobuf.h>
 #include <ydb/core/tablet_flat/tablet_flat_executed.h>
 #include <ydb/core/tx/scheme_cache/scheme_cache.h>
 
@@ -66,6 +68,7 @@ void TNodeBroker::OnActivateExecutor(const TActorContext &ctx)
 
     EnableStableNodeNames = appData->FeatureFlags.GetEnableStableNodeNames();
 
+    Executor()->RegisterExternalTabletCounters(TabletCountersPtr);
     ClearState();
 
     ProcessTx(CreateTxInitScheme(), ctx);
@@ -296,6 +299,7 @@ void TNodeBroker::ProcessListNodesRequest(TEvNodeBroker::TEvListNodes::TPtr &ev)
     if (version != Epoch.Version)
         resp->PreSerializedData = EpochCache;
 
+    TabletCounters->Percentile()[COUNTER_LIST_NODES_BYTES].IncrementFor(resp->GetCachedByteSize());
     LOG_TRACE_S(TActorContext::AsActorContext(), NKikimrServices::NODE_BROKER,
                 "Send TEvNodesInfo for epoch " << Epoch.ToString());
 
@@ -427,6 +431,7 @@ void TNodeBroker::PrepareEpochCache()
         FillNodeInfo(entry.second, *info.AddExpiredNodes());
 
     Y_PROTOBUF_SUPPRESS_NODISCARD info.SerializeToString(&EpochCache);
+    TabletCounters->Simple()[COUNTER_EPOCH_SIZE_BYTES].Set(EpochCache.Size());
 }
 
 void TNodeBroker::AddNodeToEpochCache(const TNodeInfo &node)
@@ -441,6 +446,7 @@ void TNodeBroker::AddNodeToEpochCache(const TNodeInfo &node)
     Y_PROTOBUF_SUPPRESS_NODISCARD info.SerializeToString(&delta);
 
     EpochCache += delta;
+    TabletCounters->Simple()[COUNTER_EPOCH_SIZE_BYTES].Set(EpochCache.Size());
 }
 
 void TNodeBroker::SubscribeForConfigUpdates(const TActorContext &ctx)
@@ -822,6 +828,7 @@ void TNodeBroker::Handle(TEvConsole::TEvReplaceConfigSubscriptionsResponse::TPtr
 void TNodeBroker::Handle(TEvNodeBroker::TEvListNodes::TPtr &ev,
                          const TActorContext &)
 {
+    TabletCounters->Cumulative()[COUNTER_LIST_NODES_REQUESTS].Increment(1);
     auto &rec = ev->Get()->Record;
 
     ui64 epoch = rec.GetMinEpoch();
@@ -836,6 +843,7 @@ void TNodeBroker::Handle(TEvNodeBroker::TEvListNodes::TPtr &ev,
 void TNodeBroker::Handle(TEvNodeBroker::TEvResolveNode::TPtr &ev,
                          const TActorContext &ctx)
 {
+    TabletCounters->Cumulative()[COUNTER_RESOLVE_NODE_REQUESTS].Increment(1);
     ui32 nodeId = ev->Get()->Record.GetNodeId();
     TAutoPtr<TEvNodeBroker::TEvResolvedNode> resp = new TEvNodeBroker::TEvResolvedNode;
 
@@ -859,6 +867,7 @@ void TNodeBroker::Handle(TEvNodeBroker::TEvRegistrationRequest::TPtr &ev,
 {
     LOG_TRACE_S(ctx, NKikimrServices::NODE_BROKER, "Handle TEvNodeBroker::TEvRegistrationRequest"
         << ": request# " << ev->Get()->Record.ShortDebugString());
+    TabletCounters->Cumulative()[COUNTER_REGISTRATION_REQUESTS].Increment(1);
 
     class TResolveTenantActor : public TActorBootstrapped<TResolveTenantActor> {
         TEvNodeBroker::TEvRegistrationRequest::TPtr Ev;
@@ -945,6 +954,7 @@ void TNodeBroker::Handle(TEvNodeBroker::TEvRegistrationRequest::TPtr &ev,
 void TNodeBroker::Handle(TEvNodeBroker::TEvExtendLeaseRequest::TPtr &ev,
                          const TActorContext &ctx)
 {
+    TabletCounters->Cumulative()[COUNTER_EXTEND_LEASE_REQUESTS].Increment(1);
     ui32 nodeId = ev->Get()->Record.GetNodeId();
     ProcessTx(nodeId, CreateTxExtendLease(ev), ctx);
 }
@@ -995,6 +1005,23 @@ void TNodeBroker::Handle(TEvPrivate::TEvResolvedRegistrationRequest::TPtr &ev,
     ProcessTx(CreateTxRegisterNode(ev), ctx);
 }
 
+TNodeBroker::TNodeBroker(const TActorId &tablet, TTabletStorageInfo *info)
+        : TActor(&TThis::StateInit)
+        , TTabletExecutedFlat(info, tablet, new NMiniKQL::TMiniKQLFactory)
+        , EpochDuration(TDuration::Hours(1))
+        , ConfigSubscriptionId(0)
+        , StableNodeNamePrefix("slot-")
+        , TxProcessor(new TTxProcessor(*this, "root", NKikimrServices::NODE_BROKER))
+{
+    TabletCountersPtr.Reset(new TProtobufTabletCounters<
+        ESimpleCounters_descriptor,
+        ECumulativeCounters_descriptor,
+        EPercentileCounters_descriptor,
+        ETxTypes_descriptor
+    >());
+    TabletCounters = TabletCountersPtr.Get();
+}
+
 IActor *CreateNodeBroker(const TActorId &tablet,
                          TTabletStorageInfo *info)
 {

diff --git a/ydb/core/mind/node_broker__extend_lease.cpp b/ydb/core/mind/node_broker__extend_lease.cpp
@@ -2,6 +2,7 @@
 #include "node_broker__scheme.h"
 
 #include <ydb/core/base/appdata.h>
+#include <ydb/core/protos/counters_node_broker.pb.h>
 
 namespace NKikimr {
 namespace NNodeBroker {
@@ -17,6 +18,8 @@ class TNodeBroker::TTxExtendLease : public TTransactionBase<TNodeBroker> {
     {
     }
 
+    TTxType GetTxType() const override { return TXTYPE_EXTEND_LEASE; }
+
     bool Error(TStatus::ECode code,
                const TString &reason,
                const TActorContext &ctx)

diff --git a/ydb/core/mind/node_broker__init_scheme.cpp b/ydb/core/mind/node_broker__init_scheme.cpp
@@ -1,6 +1,8 @@
 #include "node_broker_impl.h"
 #include "node_broker__scheme.h"
 
+#include <ydb/core/protos/counters_node_broker.pb.h>
+
 namespace NKikimr {
 namespace NNodeBroker {
 
@@ -11,6 +13,8 @@ class TNodeBroker::TTxInitScheme : public TTransactionBase<TNodeBroker> {
     {
     }
 
+    TTxType GetTxType() const override { return TXTYPE_INIT_SCHEME; }
+
     bool Execute(TTransactionContext &txc, const TActorContext &ctx) override
     {
         LOG_DEBUG(ctx, NKikimrServices::NODE_BROKER, "TTxInitScheme Execute");

diff --git a/ydb/core/mind/node_broker__load_state.cpp b/ydb/core/mind/node_broker__load_state.cpp
@@ -2,6 +2,7 @@
 #include "node_broker__scheme.h"
 
 #include <ydb/core/base/appdata.h>
+#include <ydb/core/protos/counters_node_broker.pb.h>
 
 namespace NKikimr {
 namespace NNodeBroker {
@@ -13,6 +14,8 @@ class TNodeBroker::TTxLoadState : public TTransactionBase<TNodeBroker> {
     {
     }
 
+    TTxType GetTxType() const override { return TXTYPE_LOAD_STATE; }
+
     bool Execute(TTransactionContext &txc, const TActorContext &ctx) override
     {
         LOG_DEBUG(ctx, NKikimrServices::NODE_BROKER, "TTxLoadState Execute");

diff --git a/ydb/core/mind/node_broker__register_node.cpp b/ydb/core/mind/node_broker__register_node.cpp
@@ -2,6 +2,7 @@
 #include "node_broker__scheme.h"
 
 #include <ydb/core/base/appdata.h>
+#include <ydb/core/protos/counters_node_broker.pb.h>
 
 namespace NKikimr {
 namespace NNodeBroker {
@@ -21,6 +22,8 @@ class TNodeBroker::TTxRegisterNode : public TTransactionBase<TNodeBroker> {
     {
     }
 
+    TTxType GetTxType() const override { return TXTYPE_REGISTER_NODE; }
+
     bool Error(TStatus::ECode code,
                const TString &reason,
                const TActorContext &ctx)

diff --git a/ydb/core/mind/node_broker__update_config.cpp b/ydb/core/mind/node_broker__update_config.cpp
@@ -1,6 +1,8 @@
 #include "node_broker_impl.h"
 #include "node_broker__scheme.h"
 
+#include <ydb/core/protos/counters_node_broker.pb.h>
+
 namespace NKikimr {
 namespace NNodeBroker {
 
@@ -24,6 +26,8 @@ class TNodeBroker::TTxUpdateConfig : public TTransactionBase<TNodeBroker> {
     {
     }
 
+    TTxType GetTxType() const override { return TXTYPE_UPDATE_CONFIG; }
+
     bool ProcessNotification(const TActorContext &ctx)
     {
         auto &rec = Notification->Get()->Record;

diff --git a/ydb/core/mind/node_broker__update_config_subscription.cpp b/ydb/core/mind/node_broker__update_config_subscription.cpp
@@ -1,6 +1,8 @@
 #include "node_broker_impl.h"
 #include "node_broker__scheme.h"
 
+#include <ydb/core/protos/counters_node_broker.pb.h>
+
 namespace NKikimr {
 namespace NNodeBroker {
 
@@ -14,6 +16,8 @@ class TNodeBroker::TTxUpdateConfigSubscription : public TTransactionBase<TNodeBr
     {
     }
 
+    TTxType GetTxType() const override { return TXTYPE_UPDATE_CONFIG_SUBSCRIPTION; }
+
     bool Execute(TTransactionContext &txc,
                  const TActorContext &ctx) override
     {

diff --git a/ydb/core/mind/node_broker__update_epoch.cpp b/ydb/core/mind/node_broker__update_epoch.cpp
@@ -1,6 +1,8 @@
 #include "node_broker_impl.h"
 #include "node_broker__scheme.h"
 
+#include <ydb/core/protos/counters_node_broker.pb.h>
+
 namespace NKikimr {
 namespace NNodeBroker {
 
@@ -11,6 +13,8 @@ class TNodeBroker::TTxUpdateEpoch : public TTransactionBase<TNodeBroker> {
     {
     }
 
+    TTxType GetTxType() const override { return TXTYPE_UPDATE_EPOCH; }
+
     bool Execute(TTransactionContext &txc, const TActorContext &ctx) override
     {
         LOG_DEBUG_S(ctx, NKikimrServices::NODE_BROKER, "TTxUpdateEpoch Execute");

diff --git a/ydb/core/mind/node_broker_impl.h b/ydb/core/mind/node_broker_impl.h
@@ -341,16 +341,11 @@ class TNodeBroker : public TActor<TNodeBroker>
     TSchedulerCookieHolder EpochTimerCookieHolder;
     TString EpochCache;
 
+    TTabletCountersBase* TabletCounters;
+    TAutoPtr<TTabletCountersBase> TabletCountersPtr;
+
 public:
-    TNodeBroker(const TActorId &tablet, TTabletStorageInfo *info)
-        : TActor(&TThis::StateInit)
-        , TTabletExecutedFlat(info, tablet, new NMiniKQL::TMiniKQLFactory)
-        , EpochDuration(TDuration::Hours(1))
-        , ConfigSubscriptionId(0)
-        , StableNodeNamePrefix("slot-")
-        , TxProcessor(new TTxProcessor(*this, "root", NKikimrServices::NODE_BROKER))
-    {
-    }
+    TNodeBroker(const TActorId &tablet, TTabletStorageInfo *info);
 
     static constexpr NKikimrServices::TActivity::EType ActorActivityType()
     {

diff --git a/ydb/core/protos/counters_node_broker.proto b/ydb/core/protos/counters_node_broker.proto
@@ -0,0 +1,54 @@
+import "ydb/core/protos/counters.proto";
+
+package NKikimr.NNodeBroker;
+
+option java_package = "ru.yandex.kikimr.proto";
+
+option (NKikimr.TabletTypeName) = "NodeBroker"; // Used as prefix for all counters
+
+enum ESimpleCounters {
+    COUNTER_EPOCH_SIZE_BYTES = 0           [(CounterOpts) = {Name: "EpochSizeBytes"}];
+}
+
+enum ECumulativeCounters {
+    COUNTER_LIST_NODES_REQUESTS = 0        [(CounterOpts) = {Name: "ListNodesRequests"}];
+    COUNTER_RESOLVE_NODE_REQUESTS = 1      [(CounterOpts) = {Name: "ResolveNodeRequests"}];
+    COUNTER_REGISTRATION_REQUESTS = 2      [(CounterOpts) = {Name: "RegistrationRequests"}];
+    COUNTER_EXTEND_LEASE_REQUESTS = 3      [(CounterOpts) = {Name: "ExtendLeaseRequests"}];
+}
+
+enum EPercentileCounters {
+    COUNTER_LIST_NODES_BYTES = 0 [(CounterOpts) = {
+        Name: "ListNodesBytes",
+        Ranges: { Value: 0            Name: "0"           }
+        Ranges: { Value: 128          Name: "128"         }
+        Ranges: { Value: 1024         Name: "1024"        }
+        Ranges: { Value: 2048         Name: "2048"        }
+        Ranges: { Value: 4096         Name: "4096"        }
+        Ranges: { Value: 8192         Name: "8192"        }
+        Ranges: { Value: 16384        Name: "16384"       }
+        Ranges: { Value: 32768        Name: "32768"       }
+        Ranges: { Value: 65536        Name: "65536"       }
+        Ranges: { Value: 131072       Name: "131072"      }
+        Ranges: { Value: 262144       Name: "262144"      }
+        Ranges: { Value: 524288       Name: "524288"      }
+        Ranges: { Value: 1048576      Name: "1048576"     }
+        Ranges: { Value: 2097152      Name: "2097152"     }
+        Ranges: { Value: 4194304      Name: "4194304"     }
+        Ranges: { Value: 8388608      Name: "8388608"     }
+        Ranges: { Value: 16777216     Name: "16777216"    }
+        Ranges: { Value: 33554432     Name: "33554432"    }
+        Ranges: { Value: 67108864     Name: "67108864"    }
+        Ranges: { Value: 134217728    Name: "134217728"   }
+    }];
+}
+
+enum ETxTypes {
+    TXTYPE_EXTEND_LEASE = 0                 [(TxTypeOpts) = {Name: "TTxExtendLease"}];
+    TXTYPE_INIT_SCHEME = 1                  [(TxTypeOpts) = {Name: "TTxInitScheme"}];
+    TXTYPE_LOAD_STATE = 2                   [(TxTypeOpts) = {Name: "TTxLoadState"}];
+    TXTYPE_REGISTER_NODE = 3                [(TxTypeOpts) = {Name: "TTxRegisterNode"}];
+    TXTYPE_UPDATE_CONFIG = 4                [(TxTypeOpts) = {Name: "TTxUpdateConfig"}];
+    TXTYPE_UPDATE_CONFIG_SUBSCRIPTION = 5   [(TxTypeOpts) = {Name: "TTxUpdateConfigSubscription"}];
+    TXTYPE_UPDATE_EPOCH = 6                 [(TxTypeOpts) = {Name: "TTxUpdateEpoch"}];
+}
diff --git a/ydb/core/protos/ya.make b/ydb/core/protos/ya.make
@@ -51,6 +51,7 @@ SRCS(
     counters_kesus.proto
     counters_keyvalue.proto
     counters_mediator.proto
+    counters_node_broker.proto
     counters_pq.proto
     counters_replication.proto
     counters_schemeshard.proto