Improve collator node pings and collation manager stats

ton-blockchain · Nov 28, 2024 · 923f1cd · 923f1cd
1 parent 5fae8db
commit 923f1cd
Show file tree

Hide file tree

Showing 8 changed files with 66 additions and 7 deletions.
diff --git a/tl/generate/scheme/ton_api.tl b/tl/generate/scheme/ton_api.tl
@@ -750,7 +750,7 @@ engine.validator.perfTimerStats stats:(vector engine.validator.PerfTimerStatsByN
 engine.validator.shardOutQueueSize size:long = engine.validator.ShardOutQueueSize;
 
 engine.validator.collationManagerStats.shard shard_id:tonNode.shardId self_collate:Bool select_mode:string active:Bool collators:(vector int256) = engine.validator.collationManagerStats.Shard;
-engine.validator.collationManagerStats.collator adnl_id:int256 active:Bool alive:Bool ping_in:double = engine.validator.collationManagerStats.Collator;
+engine.validator.collationManagerStats.collator adnl_id:int256 active:Bool alive:Bool ping_in:double last_ping_ago:double last_ping_status:string = engine.validator.collationManagerStats.Collator;
 engine.validator.collationManagerStats.localId adnl_id:int256 shards:(vector engine.validator.collationManagerStats.shard)
     collators:(vector engine.validator.collationManagerStats.collator) = engine.validator.collationManagerStats.LocalId;
 engine.validator.collationManagerStats local_ids:(vector engine.validator.collationManagerStats.localId) = engine.validator.CollationManagerStats;

diff --git a/tl/generate/scheme/ton_api.tlo b/tl/generate/scheme/ton_api.tlo
diff --git a/validator-engine-console/validator-engine-console-query.cpp b/validator-engine-console/validator-engine-console-query.cpp
@@ -1775,8 +1775,24 @@ td::Status GetCollationManagerStatsQuery::receive(td::BufferSlice data) {
         if (collator == nullptr) {
           return td::Status::Error("collator not found");
         }
-        td::TerminalIO::out() << "      " << id << " alive=" << (int)collator->alive_
-                              << " ping_in=" << collator->ping_in_ << "\n";
+        td::StringBuilder sb;
+        sb << "      " << id << "\n";
+        sb << "        alive=" << (int)collator->alive_;
+        if (collator->active_) {
+          sb << " ping_in=" << td::StringBuilder::FixedDouble(std::max(collator->ping_in_, 0.0), 3);
+        }
+        sb << " last_ping_ago=";
+        if (collator->last_ping_ago_ < 0.0) {
+          sb << "never";
+        } else {
+          std::string status = collator->last_ping_status_;
+          std::erase_if(status, [](char c) { return c < (char)32; });
+          if (status.size() > 128) {
+            status.resize(128);
+          }
+          sb << td::StringBuilder::FixedDouble(collator->last_ping_ago_, 3) << ": " << status;
+        }
+        td::TerminalIO::out() << sb.as_cslice() << "\n";
       }
     }
   }

diff --git a/validator/collation-manager.cpp b/validator/collation-manager.cpp
@@ -261,6 +261,8 @@ void CollationManager::get_stats(
     } else {
       obj->ping_in_ = -1.0;
     }
+    obj->last_ping_ago_ = collator.last_ping_at ? td::Time::now() - collator.last_ping_at.at() : -1.0;
+    obj->last_ping_status_ = collator.last_ping_status.is_ok() ? "OK" : collator.last_ping_status.message().str();
     stats->collators_.push_back(std::move(obj));
   }
   promise.set_value(std::move(stats));
@@ -323,7 +325,7 @@ void CollationManager::alarm() {
         td::actor::send_closure(SelfId, &CollationManager::got_pong, id, std::move(R));
       };
       LOG(DEBUG) << "sending ping to " << id;
-      td::actor::send_closure(rldp_, &rldp::Rldp::send_query, local_id_, id, "collatorping", std::move(P),
+      td::actor::send_closure(rldp_, &rldp::Rldp::send_query, local_id_, id, "ping", std::move(P),
                               td::Timestamp::in(2.0), std::move(query));
     } else {
       alarm_timestamp().relax(collator.ping_at);
@@ -340,20 +342,23 @@ void CollationManager::got_pong(adnl::AdnlNodeIdShort id, td::Result<td::BufferS
   collator.sent_ping = false;
 
   auto r_pong = [&]() -> td::Result<tl_object_ptr<ton_api::collatorNode_pong>> {
-    TRY_RESULT_PREFIX(data, std::move(R), "rldp query error: ");
+    TRY_RESULT(data, std::move(R));
     auto r_error = fetch_tl_object<ton_api::collatorNode_error>(data, true);
     if (r_error.is_ok()) {
       auto error = r_error.move_as_ok();
       return td::Status::Error(error->code_, error->message_);
     }
     return fetch_tl_object<ton_api::collatorNode_pong>(data, true);
   }();
+  collator.last_ping_at = td::Timestamp::now();
   if (r_pong.is_error()) {
-    LOG(DEBUG) << "pong from " << id << " : " << r_pong.move_as_error();
+    LOG(DEBUG) << "pong from " << id << " : " << r_pong.error();
     collator.alive = false;
+    collator.last_ping_status = r_pong.move_as_error();
   } else {
     LOG(DEBUG) << "pong from " << id << " : OK";
     collator.alive = true;
+    collator.last_ping_status = td::Status::OK();
   }
   collator.ping_at = td::Timestamp::in(td::Random::fast(10.0, 20.0));
   if (collator.active_cnt && !collator.sent_ping) {

diff --git a/validator/collation-manager.hpp b/validator/collation-manager.hpp
@@ -65,6 +65,8 @@ class CollationManager : public td::actor::Actor {
     td::Timestamp ping_at = td::Timestamp::now();
     bool sent_ping = false;
     size_t active_cnt = 0;
+    td::Timestamp last_ping_at = td::Timestamp::never();
+    td::Status last_ping_status = td::Status::Error("not pinged");
   };
   std::map<adnl::AdnlNodeIdShort, CollatorInfo> collators_;
 

diff --git a/validator/collator-node.cpp b/validator/collator-node.cpp
@@ -167,6 +167,10 @@ void CollatorNode::new_masterchain_block_notification(td::Ref<MasterchainState>
   }
 }
 
+void CollatorNode::update_shard_client_handle(BlockHandle shard_client_handle) {
+  shard_client_handle_ = shard_client_handle;
+}
+
 void CollatorNode::update_validator_group_info(ShardIdFull shard, std::vector<BlockIdExt> prev,
                                                CatchainSeqno cc_seqno) {
   if (!can_collate_shard(shard)) {
@@ -225,7 +229,12 @@ void CollatorNode::update_validator_group_info(ShardIdFull shard, std::vector<Bl
           }
           ++cache_it;
         }
-        generate_block(shard, cc_seqno, info.prev, {}, td::Timestamp::in(10.0), [](td::Result<BlockCandidate>) {});
+        auto S = check_out_of_sync();
+        if (S.is_ok()) {
+          generate_block(shard, cc_seqno, info.prev, {}, td::Timestamp::in(10.0), [](td::Result<BlockCandidate>) {});
+        } else {
+          LOG(DEBUG) << "not generating block automatically: " << S;
+        }
       }
       return;
     }
@@ -535,9 +544,22 @@ void CollatorNode::process_result(std::shared_ptr<CacheEntry> cache_entry, td::R
   cache_entry->promises.clear();
 }
 
+td::Status CollatorNode::check_out_of_sync() {
+  if (last_masterchain_state_.is_null() || !shard_client_handle_) {
+    return td::Status::Error("not inited");
+  }
+  auto now = (UnixTime)td::Clocks::system();
+  if (last_masterchain_state_->get_unix_time() < now - 60 || shard_client_handle_->unix_time() < now - 60) {
+    return td::Status::Error(PSTRING() << "out of sync: mc " << now - last_masterchain_state_->get_unix_time()
+                                       << "s ago, shardclient " << now - shard_client_handle_->unix_time() << "s ago");
+  }
+  return td::Status::OK();
+}
+
 void CollatorNode::process_ping(adnl::AdnlNodeIdShort src, ton_api::collatorNode_ping& ping,
                                 td::Promise<td::BufferSlice> promise) {
   LOG(DEBUG) << "got ping from " << src;
+  TRY_STATUS_PROMISE(promise, check_out_of_sync());
   promise.set_result(create_serialize_tl_object<ton_api::collatorNode_pong>(0));
 }
 

diff --git a/validator/collator-node.hpp b/validator/collator-node.hpp
@@ -36,6 +36,7 @@ class CollatorNode : public td::actor::Actor {
   void del_shard(ShardIdFull shard);
 
   void new_masterchain_block_notification(td::Ref<MasterchainState> state);
+  void update_shard_client_handle(BlockHandle shard_client_handle);
   void update_validator_group_info(ShardIdFull shard, std::vector<BlockIdExt> prev, CatchainSeqno cc_seqno);
 
   void update_options(td::Ref<ValidatorManagerOptions> opts) {
@@ -84,6 +85,7 @@ class CollatorNode : public td::actor::Actor {
   std::map<std::pair<ShardIdFull, CatchainSeqno>, FutureValidatorGroup> future_validator_groups_;
 
   td::Ref<MasterchainState> last_masterchain_state_;
+  BlockHandle shard_client_handle_;
 
   td::Result<FutureValidatorGroup*> get_future_validator_group(ShardIdFull shard, CatchainSeqno cc_seqno);
 
@@ -92,6 +94,8 @@ class CollatorNode : public td::actor::Actor {
                       td::Promise<BlockCandidate> promise);
   void process_result(std::shared_ptr<CacheEntry> cache_entry, td::Result<BlockCandidate> R);
 
+  td::Status check_out_of_sync();
+
  public:
   static tl_object_ptr<ton_api::collatorNode_Candidate> serialize_candidate(const BlockCandidate& block, bool compress);
   static td::Result<BlockCandidate> deserialize_candidate(tl_object_ptr<ton_api::collatorNode_Candidate> f,

diff --git a/validator/manager.cpp b/validator/manager.cpp
@@ -2754,6 +2754,9 @@ void ValidatorManagerImpl::update_shard_client_block_handle(BlockHandle handle,
       last_liteserver_state_ = std::move(state);
     }
   }
+  for (auto &c : collator_nodes_) {
+    td::actor::send_closure(c.second.actor, &CollatorNode::update_shard_client_handle, shard_client_handle_);
+  }
   shard_client_update(seqno);
   promise.set_value(td::Unit());
 }
@@ -3509,6 +3512,13 @@ void ValidatorManagerImpl::add_collator(adnl::AdnlNodeIdShort id, ShardIdFull sh
   if (it == collator_nodes_.end()) {
     it = collator_nodes_.emplace(id, Collator()).first;
     it->second.actor = td::actor::create_actor<CollatorNode>("collatornode", id, opts_, actor_id(this), adnl_, rldp_);
+    if (last_masterchain_state_.not_null()) {
+      td::actor::send_closure(it->second.actor, &CollatorNode::new_masterchain_block_notification,
+                              last_masterchain_state_);
+    }
+    if (shard_client_handle_) {
+      td::actor::send_closure(it->second.actor, &CollatorNode::update_shard_client_handle, shard_client_handle_);
+    }
   }
   if (!it->second.shards.insert(shard).second) {
     return;