Dynamic catchain delays, state serialization improvements (ton-blockc…

…hain#1140) * Validator improvements * Fix cancelling state serialization * Disable state serializer on all mainnet validators * Flag --catchain-max-block-delay-slow * Set default catchain-max-block-delay to 0.4, delay-slow to 1.0 --------- Co-authored-by: SpyCheese <[email protected]>
ChorusOne · Aug 30, 2024 · e081111 · e081111
1 parent 97c57c3
commit e081111
Show file tree

Hide file tree

Showing 15 changed files with 84 additions and 27 deletions.
diff --git a/catchain/catchain-receiver.cpp b/catchain/catchain-receiver.cpp
@@ -368,6 +368,12 @@ void CatChainReceiverImpl::add_block(td::BufferSlice payload, std::vector<CatCha
   }
 
   int height = prev->height_ + 1;
+  auto max_block_height = get_max_block_height(opts_, sources_.size());
+  if (height > max_block_height) {
+    VLOG(CATCHAIN_WARNING) << this << ": cannot create block: max height exceeded (" << max_block_height << ")";
+    active_send_ = false;
+    return;
+  }
   auto block_data = create_tl_object<ton_api::catchain_block_data>(std::move(prev), std::move(deps_arr));
   auto block = create_tl_object<ton_api::catchain_block>(incarnation_, local_idx_, height, std::move(block_data),
                                                          td::BufferSlice());

diff --git a/validator-engine/validator-engine.cpp b/validator-engine/validator-engine.cpp
@@ -1448,6 +1448,9 @@ td::Status ValidatorEngine::load_global_config() {
   if (catchain_max_block_delay_) {
     validator_options_.write().set_catchain_max_block_delay(catchain_max_block_delay_.value());
   }
+  if (catchain_max_block_delay_slow_) {
+    validator_options_.write().set_catchain_max_block_delay_slow(catchain_max_block_delay_slow_.value());
+  }
 
   std::vector<ton::BlockIdExt> h;
   for (auto &x : conf.validator_->hardforks_) {
@@ -4072,7 +4075,7 @@ int main(int argc, char *argv[]) {
     logger_ = td::TsFileLog::create(fname.str()).move_as_ok();
     td::log_interface = logger_.get();
   });
-  p.add_checked_option('s', "state-ttl", "state will be gc'd after this time (in seconds) default=3600",
+  p.add_checked_option('s', "state-ttl", "state will be gc'd after this time (in seconds) default=86400",
                        [&](td::Slice fname) {
                          auto v = td::to_double(fname);
                          if (v <= 0) {
@@ -4233,7 +4236,7 @@ int main(int argc, char *argv[]) {
       "preload all cells from CellDb on startup (recommended to use with big enough celldb-cache-size and celldb-direct-io)",
       [&]() { acts.push_back([&x]() { td::actor::send_closure(x, &ValidatorEngine::set_celldb_preload_all, true); }); });
   p.add_checked_option(
-      '\0', "catchain-max-block-delay", "delay before creating a new catchain block, in seconds (default: 0.5)",
+      '\0', "catchain-max-block-delay", "delay before creating a new catchain block, in seconds (default: 0.4)",
       [&](td::Slice s) -> td::Status {
         auto v = td::to_double(s);
         if (v < 0) {
@@ -4242,6 +4245,16 @@ int main(int argc, char *argv[]) {
         acts.push_back([&x, v]() { td::actor::send_closure(x, &ValidatorEngine::set_catchain_max_block_delay, v); });
         return td::Status::OK();
       });
+  p.add_checked_option(
+      '\0', "catchain-max-block-delay-slow", "max extended catchain block delay (for too long rounds), (default: 1.0)",
+      [&](td::Slice s) -> td::Status {
+        auto v = td::to_double(s);
+        if (v < 0) {
+          return td::Status::Error("catchain-max-block-delay-slow should be non-negative");
+        }
+        acts.push_back([&x, v]() { td::actor::send_closure(x, &ValidatorEngine::set_catchain_max_block_delay_slow, v); });
+        return td::Status::OK();
+      });
   p.add_option(
       '\0', "fast-state-serializer",
       "faster persistent state serializer, but requires more RAM (enabled automatically on machines with >= 90GB RAM)",

diff --git a/validator-engine/validator-engine.hpp b/validator-engine/validator-engine.hpp
@@ -214,7 +214,7 @@ class ValidatorEngine : public td::actor::Actor {
   td::optional<td::uint64> celldb_cache_size_ = 1LL << 30;
   bool celldb_direct_io_ = false;
   bool celldb_preload_all_ = false;
-  td::optional<double> catchain_max_block_delay_;
+  td::optional<double> catchain_max_block_delay_, catchain_max_block_delay_slow_;
   bool read_config_ = false;
   bool started_keyring_ = false;
   bool started_ = false;
@@ -300,6 +300,9 @@ class ValidatorEngine : public td::actor::Actor {
   void set_catchain_max_block_delay(double value) {
     catchain_max_block_delay_ = value;
   }
+  void set_catchain_max_block_delay_slow(double value) {
+    catchain_max_block_delay_slow_ = value;
+  }
   void set_fast_state_serializer_enabled(bool value) {
     fast_state_serializer_enabled_ = value;
   }

diff --git a/validator-session/validator-session-state.h b/validator-session/validator-session-state.h
@@ -478,6 +478,14 @@ class ValidatorSessionState : public ValidatorSessionDescription::RootObject {
   auto get_ts(td::uint32 src_idx) const {
     return att_->at(src_idx);
   }
+  td::uint32 cur_attempt_in_round(const ValidatorSessionDescription& desc) const {
+    td::uint32 first_attempt = cur_round_->get_first_attempt(desc.get_self_idx());
+    td::uint32 cur_attempt = desc.get_attempt_seqno(desc.get_ts());
+    if (cur_attempt < first_attempt || first_attempt == 0) {
+      return 0;
+    }
+    return cur_attempt - first_attempt;
+  }
 
   const SentBlock* choose_block_to_sign(ValidatorSessionDescription& desc, td::uint32 src_idx, bool& found) const;
   const SentBlock* get_committed_block(ValidatorSessionDescription& desc, td::uint32 seqno) const;

diff --git a/validator-session/validator-session.cpp b/validator-session/validator-session.cpp
@@ -813,13 +813,25 @@ void ValidatorSessionImpl::request_new_block(bool now) {
   } else {
     double lambda = 10.0 / description().get_total_nodes();
     double x = -1 / lambda * log(td::Random::fast(1, 999) * 0.001);
-    if (x > catchain_max_block_delay_) {  // default = 0.5
-      x = catchain_max_block_delay_;
-    }
+    x = std::min(x, get_current_max_block_delay());  // default = 0.4
     td::actor::send_closure(catchain_, &catchain::CatChain::need_new_block, td::Timestamp::in(x));
   }
 }
 
+double ValidatorSessionImpl::get_current_max_block_delay() const {
+  td::uint32 att = real_state_->cur_attempt_in_round(*description_);
+  td::uint32 att1 = description_->opts().max_round_attempts;
+  if (att <= att1) {
+    return catchain_max_block_delay_;
+  }
+  td::uint32 att2 = att1 + 4;
+  if (att >= att2) {
+    return catchain_max_block_delay_slow_;
+  }
+  return catchain_max_block_delay_ +
+         (catchain_max_block_delay_slow_ - catchain_max_block_delay_) * (double)(att - att1) / (double)(att2 - att1);
+}
+
 void ValidatorSessionImpl::on_new_round(td::uint32 round) {
   if (round != 0) {
     CHECK(cur_round_ < round);

diff --git a/validator-session/validator-session.h b/validator-session/validator-session.h
@@ -109,7 +109,7 @@ class ValidatorSession : public td::actor::Actor {
   virtual void get_validator_group_info_for_litequery(
       td::uint32 cur_round,
       td::Promise<std::vector<tl_object_ptr<lite_api::liteServer_nonfinal_candidateInfo>>> promise) = 0;
-  virtual void set_catchain_max_block_delay(double value) = 0;
+  virtual void set_catchain_max_block_delay(double delay, double delay_slow) = 0;
 
   static td::actor::ActorOwn<ValidatorSession> create(
       catchain::CatChainSessionId session_id, ValidatorSessionOptions opts, PublicKeyHash local_id,

diff --git a/validator-session/validator-session.hpp b/validator-session/validator-session.hpp
@@ -91,6 +91,7 @@ class ValidatorSessionImpl : public ValidatorSession {
   std::unique_ptr<ValidatorSessionDescription> description_;
 
   double catchain_max_block_delay_ = 0.4;
+  double catchain_max_block_delay_slow_ = 1.0;
 
   void on_new_round(td::uint32 round);
   void on_catchain_started();
@@ -150,6 +151,7 @@ class ValidatorSessionImpl : public ValidatorSession {
   }
 
   void request_new_block(bool now);
+  double get_current_max_block_delay() const;
   void get_broadcast_p2p(PublicKeyHash node, ValidatorSessionFileHash file_hash,
                          ValidatorSessionCollatedDataFileHash collated_data_file_hash, PublicKeyHash src,
                          td::uint32 round, ValidatorSessionRootHash root_hash, td::Promise<td::BufferSlice> promise,
@@ -191,8 +193,10 @@ class ValidatorSessionImpl : public ValidatorSession {
   void get_validator_group_info_for_litequery(
       td::uint32 cur_round,
       td::Promise<std::vector<tl_object_ptr<lite_api::liteServer_nonfinal_candidateInfo>>> promise) override;
-  void set_catchain_max_block_delay(double value) override {
-    catchain_max_block_delay_ = value;
+
+  void set_catchain_max_block_delay(double delay, double delay_slow) override {
+    catchain_max_block_delay_ = delay;
+    catchain_max_block_delay_slow_ = delay_slow;
   }
 
   void process_blocks(std::vector<catchain::CatChainBlock *> blocks);

diff --git a/validator/impl/shard.cpp b/validator/impl/shard.cpp
@@ -44,6 +44,7 @@ ShardStateQ::ShardStateQ(const ShardStateQ& other)
     , root(other.root)
     , lt(other.lt)
     , utime(other.utime)
+    , global_id_(other.global_id_)
     , before_split_(other.before_split_)
     , fake_split_(other.fake_split_)
     , fake_merge_(other.fake_merge_) {
@@ -121,6 +122,7 @@ td::Status ShardStateQ::init() {
   }
   lt = info.gen_lt;
   utime = info.gen_utime;
+  global_id_ = info.global_id;
   before_split_ = info.before_split;
   block::ShardId id{info.shard_id};
   ton::BlockId hdr_id{ton::ShardIdFull(id), info.seq_no};

diff --git a/validator/impl/shard.hpp b/validator/impl/shard.hpp
@@ -38,6 +38,7 @@ class ShardStateQ : virtual public ShardState {
   Ref<vm::Cell> root;
   LogicalTime lt{0};
   UnixTime utime{0};
+  td::int32 global_id_{0};
   bool before_split_{false};
   bool fake_split_{false};
   bool fake_merge_{false};
@@ -81,6 +82,9 @@ class ShardStateQ : virtual public ShardState {
   LogicalTime get_logical_time() const override {
     return lt;
   }
+  td::int32 get_global_id() const override {
+    return global_id_;
+  }
   td::optional<BlockIdExt> get_master_ref() const override {
     return master_ref;
   }

diff --git a/validator/interfaces/shard.h b/validator/interfaces/shard.h
@@ -39,6 +39,7 @@ class ShardState : public td::CntObject {
 
   virtual UnixTime get_unix_time() const = 0;
   virtual LogicalTime get_logical_time() const = 0;
+  virtual td::int32 get_global_id() const = 0;
   virtual ShardIdFull get_shard() const = 0;
   virtual BlockSeqno get_seqno() const = 0;
   virtual BlockIdExt get_block_id() const = 0;

diff --git a/validator/manager.cpp b/validator/manager.cpp
@@ -2057,7 +2057,6 @@ void ValidatorManagerImpl::update_shards() {
     }
   }
 
-  bool validating_masterchain = false;
   if (allow_validate_) {
     for (auto &desc : new_shards) {
       auto shard = desc.first;
@@ -2074,9 +2073,6 @@ void ValidatorManagerImpl::update_shards() {
       auto validator_id = get_validator(shard, val_set);
 
       if (!validator_id.is_zero()) {
-        if (shard.is_masterchain()) {
-          validating_masterchain = true;
-        }
         auto val_group_id = get_validator_set_id(shard, val_set, opts_hash, key_seqno, opts);
 
         if (force_recover) {
@@ -2171,16 +2167,14 @@ void ValidatorManagerImpl::update_shards() {
       td::actor::send_closure(SelfId, &ValidatorManagerImpl::written_destroyed_validator_sessions, std::move(gc));
     });
     td::actor::send_closure(db_, &Db::update_destroyed_validator_sessions, gc_list_, std::move(P));
+  }
 
-    if (!serializer_.empty()) {
-      td::actor::send_closure(
-          serializer_, &AsyncStateSerializer::auto_disable_serializer,
-          validating_masterchain &&
-              last_masterchain_state_->get_validator_set(ShardIdFull{masterchainId})->export_vector().size() * 2 <=
-                  last_masterchain_state_->get_total_validator_set(0)->export_vector().size());
-    }
+  if (!serializer_.empty()) {
+    td::actor::send_closure(
+        serializer_, &AsyncStateSerializer::auto_disable_serializer,
+        !validator_groups_.empty() && last_masterchain_state_->get_global_id() == -239);  // mainnet only
   }
-}  // namespace validator
+}
 
 void ValidatorManagerImpl::written_destroyed_validator_sessions(std::vector<td::actor::ActorId<ValidatorGroup>> list) {
   for (auto &v : list) {

diff --git a/validator/state-serializer.cpp b/validator/state-serializer.cpp
@@ -378,6 +378,7 @@ void AsyncStateSerializer::got_shard_handle(BlockHandle handle) {
 
 void AsyncStateSerializer::got_shard_state(BlockHandle handle, td::Ref<ShardState> state,
                                            std::shared_ptr<vm::CellDbReader> cell_db_reader) {
+  next_idx_++;
   if (!opts_->get_state_serializer_enabled() || auto_disabled_) {
     success_handler();
     return;
@@ -406,7 +407,6 @@ void AsyncStateSerializer::got_shard_state(BlockHandle handle, td::Ref<ShardStat
   });
   td::actor::send_closure(manager_, &ValidatorManager::store_persistent_state_file_gen, handle->id(),
                           masterchain_handle_->id(), write_data, std::move(P));
-  next_idx_++;
 }
 
 void AsyncStateSerializer::fail_handler(td::Status reason) {

diff --git a/validator/validator-group.cpp b/validator/validator-group.cpp
@@ -348,10 +348,12 @@ void ValidatorGroup::create_session() {
                   << ".",
         allow_unsafe_self_blocks_resync_);
   }
-  if (opts_->get_catchain_max_block_delay()) {
-    td::actor::send_closure(session_, &validatorsession::ValidatorSession::set_catchain_max_block_delay,
-                            opts_->get_catchain_max_block_delay().value());
-  }
+  double catchain_delay = opts_->get_catchain_max_block_delay() ? opts_->get_catchain_max_block_delay().value() : 0.4;
+  double catchain_delay_slow =
+      std::max(catchain_delay,
+               opts_->get_catchain_max_block_delay_slow() ? opts_->get_catchain_max_block_delay_slow().value() : 1.0);
+  td::actor::send_closure(session_, &validatorsession::ValidatorSession::set_catchain_max_block_delay, catchain_delay,
+                          catchain_delay_slow);
   if (started_) {
     td::actor::send_closure(session_, &validatorsession::ValidatorSession::start);
   }

diff --git a/validator/validator-options.hpp b/validator/validator-options.hpp
@@ -141,6 +141,9 @@ struct ValidatorManagerOptionsImpl : public ValidatorManagerOptions {
   td::optional<double> get_catchain_max_block_delay() const override {
     return catchain_max_block_delay_;
   }
+  td::optional<double> get_catchain_max_block_delay_slow() const override {
+    return catchain_max_block_delay_slow_;
+  }
   bool get_state_serializer_enabled() const override {
     return state_serializer_enabled_;
   }
@@ -230,6 +233,9 @@ struct ValidatorManagerOptionsImpl : public ValidatorManagerOptions {
   void set_catchain_max_block_delay(double value) override {
     catchain_max_block_delay_ = value;
   }
+  void set_catchain_max_block_delay_slow(double value) override {
+    catchain_max_block_delay_slow_ = value;
+  }
   void set_state_serializer_enabled(bool value) override {
     state_serializer_enabled_ = value;
   }
@@ -289,7 +295,7 @@ struct ValidatorManagerOptionsImpl : public ValidatorManagerOptions {
   td::optional<td::uint64> celldb_cache_size_;
   bool celldb_direct_io_ = false;
   bool celldb_preload_all_ = false;
-  td::optional<double> catchain_max_block_delay_;
+  td::optional<double> catchain_max_block_delay_, catchain_max_block_delay_slow_;
   bool state_serializer_enabled_ = true;
   td::Ref<CollatorOptions> collator_options_{true};
   bool fast_state_serializer_enabled_ = false;

diff --git a/validator/validator.h b/validator/validator.h
@@ -105,6 +105,7 @@ struct ValidatorManagerOptions : public td::CntObject {
   virtual bool get_celldb_direct_io() const = 0;
   virtual bool get_celldb_preload_all() const = 0;
   virtual td::optional<double> get_catchain_max_block_delay() const = 0;
+  virtual td::optional<double> get_catchain_max_block_delay_slow() const = 0;
   virtual bool get_state_serializer_enabled() const = 0;
   virtual td::Ref<CollatorOptions> get_collator_options() const = 0;
   virtual bool get_fast_state_serializer_enabled() const = 0;
@@ -136,6 +137,7 @@ struct ValidatorManagerOptions : public td::CntObject {
   virtual void set_celldb_direct_io(bool value) = 0;
   virtual void set_celldb_preload_all(bool value) = 0;
   virtual void set_catchain_max_block_delay(double value) = 0;
+  virtual void set_catchain_max_block_delay_slow(double value) = 0;
   virtual void set_state_serializer_enabled(bool value) = 0;
   virtual void set_collator_options(td::Ref<CollatorOptions> value) = 0;
   virtual void set_fast_state_serializer_enabled(bool value) = 0;