Pre resolve self address for discovery

2 files changed, 34 insertions, 2 deletions

diff --git a/library/cpp/actors/interconnect/interconnect_resolve.cpp b/library/cpp/actors/interconnect/interconnect_resolve.cpp
index 9d245f03ac..d638ff830c 100644
--- a/library/cpp/actors/interconnect/interconnect_resolve.cpp
+++ b/library/cpp/actors/interconnect/interconnect_resolve.cpp
@@ -1,6 +1,7 @@
 #include "interconnect.h"
 #include "interconnect_address.h"
 #include "events_local.h"
+#include "logging.h"
 
 #include <library/cpp/actors/core/actor_bootstrapped.h>
 #include <library/cpp/actors/core/hfunc.h>
@@ -10,7 +11,10 @@ namespace NActors {
 
     using namespace NActors::NDnsResolver;
 
-    class TInterconnectResolveActor : public TActorBootstrapped<TInterconnectResolveActor> {
+    class TInterconnectResolveActor
+        : public TActorBootstrapped<TInterconnectResolveActor>
+        , public TInterconnectLoggingBase
+    {
     public:
         TInterconnectResolveActor(
                 const TString& host, ui16 port, ui32 nodeId, const TString& defaultAddress,
@@ -41,6 +45,7 @@ namespace NActors {
         void Bootstrap() {
             TMaybe<TString> errorText;
             if (auto addr = ExtractDefaultAddr(errorText)) {
+                LOG_TRACE_IC("ICR01", "Host: %s, CACHED address: %s", Host.c_str(), DefaultAddress.c_str());
                 if (NodeId) {
                     return SendLocalNodeInfoAndDie({{*addr}});
                 } else {
@@ -58,6 +63,7 @@ namespace NActors {
                 return;
             }
 
+            LOG_DEBUG_IC("ICR02", "Host: %s, RESOLVING address ...", Host.c_str());
             Send(MakeDnsResolverActorId(),
                 NodeId
                     ? static_cast<IEventBase*>(new TEvDns::TEvGetHostByName(Host, AF_UNSPEC))
@@ -111,6 +117,7 @@ namespace NActors {
         }
 
         void SendAddressInfoAndDie(NAddr::IRemoteAddrPtr addr) {
+            LOG_DEBUG_IC("ICR03", "Host: %s, RESOLVED address", Host.c_str());
             auto reply = new TEvAddressInfo;
             reply->Address = std::move(addr);
             TActivationContext::Send(new IEventHandle(ReplyTo, ReplyFrom, reply));
@@ -118,6 +125,7 @@ namespace NActors {
         }
 
         void SendLocalNodeInfoAndDie(std::vector<NInterconnect::TAddress> addresses) {
+            LOG_DEBUG_IC("ICR04", "Host: %s, RESOLVED address", Host.c_str());
             auto reply = std::make_unique<TEvLocalNodeInfo>();
             reply->NodeId = *NodeId;
             reply->Addresses = std::move(addresses);
@@ -126,6 +134,7 @@ namespace NActors {
         }
 
         void SendErrorAndDie(const TString& errorText) {
+            LOG_DEBUG_IC("ICR05", "Host: %s, ERROR resolving: %s", Host.c_str(), errorText.c_str());
             auto *event = new TEvResolveError;
             event->Explain = errorText;
             event->Host = Host;
diff --git a/ydb/core/yq/libs/actors/nodes_manager.cpp b/ydb/core/yq/libs/actors/nodes_manager.cpp
index 746934c5de..9ebdd7f6f2 100644
--- a/ydb/core/yq/libs/actors/nodes_manager.cpp
+++ b/ydb/core/yq/libs/actors/nodes_manager.cpp
@@ -4,6 +4,7 @@
 #include <library/cpp/actors/core/events.h>
 #include <library/cpp/actors/core/hfunc.h>
 #include <library/cpp/actors/core/actor_bootstrapped.h>
+#include <library/cpp/actors/interconnect/events_local.h>
 #include <ydb/library/yql/providers/dq/worker_manager/interface/events.h>
 #include <ydb/library/yql/public/issue/yql_issue_message.h>
 #include <ydb/public/sdk/cpp/client/ydb_driver/driver.h>
@@ -76,7 +77,7 @@ public:
         Become(&TNodesManagerActor::StateFunc);
         ServiceCounters.Counters->GetCounter("EvBootstrap", true)->Inc();
         LOG_I("Bootstrap, InstanceId: " << InstanceId);
-        NodesHealthCheck();
+        ResolveSelfAddress();        
     }
 
 private:
@@ -176,6 +177,7 @@ private:
         hFunc(NDqs::TEvFreeWorkersNotify, Handle)
         hFunc(NActors::TEvents::TEvUndelivered, OnUndelivered)
         hFunc(NFq::TEvInternalService::TEvHealthCheckResponse, HandleResponse)
+        hFunc(NActors::TEvAddressInfo, Handle)
         )
 
     void HandleWakeup(NActors::TEvents::TEvWakeup::TPtr& ev) {
@@ -188,7 +190,26 @@ private:
         }
     }
 
+    void Handle(NActors::TEvAddressInfo::TPtr& ev) {
+        if (ev->Get()->Address) {
+            Address = NAddr::PrintHost(*ev->Get()->Address);
+            NodesHealthCheck();
+        } else {
+            LOG_E("TNodesManagerActor error resolving");
+            ResolveSelfAddress();
+        }
+    }
+
+    void ResolveSelfAddress() {
+        LOG_D("TNodesManagerActor::ResolveSelfAddress");
+        auto resolve = MakeHolder<NActors::TEvResolveAddress>();
+        resolve->Address = HostName();
+        resolve->Port = IcPort;
+        Send(GetNameserviceActorId(), resolve.Release());
+    }
+
     void NodesHealthCheck() {
+        LOG_T("TNodesManagerActor::NodesHealthCheck");
         const TDuration ttl = TDuration::Seconds(5);
         Schedule(ttl, new NActors::TEvents::TEvWakeup(WU_NodesHealthCheck));
 
@@ -200,6 +221,7 @@ private:
         node.set_node_id(SelfId().NodeId());
         node.set_instance_id(InstanceId);
         node.set_hostname(HostName());
+        node.set_node_address(Address);
         node.set_active_workers(AtomicGet(WorkerManagerCounters.ActiveWorkers->GetAtomic()));
         node.set_memory_limit(AtomicGet(WorkerManagerCounters.MkqlMemoryLimit->GetAtomic()));
         node.set_memory_allocated(AtomicGet(WorkerManagerCounters.MkqlMemoryAllocated->GetAtomic()));
@@ -292,6 +314,7 @@ private:
     ui32 NextPeer = 0;
     TString InstanceId;
     TActorId InternalServiceId;
+    TString Address;
 };
 
 TActorId MakeNodesManagerId() {