Add --fast-state-serializer flag and responding to AdnlMessageCreateC…

…hannel with Nop (ton-blockchain#1096)

* Add --fast-state-serializer and tools for jemalloc

* Disable fast state serializer by default unless RAM is >= 90GB
* Print jemalloc stats once a minute
* Dump jemalloc profile on request

* Respond to AdnlMessageCreateChannel with Nop

---------

Co-authored-by: SpyCheese <[email protected]>

adnl/adnl-peer.cpp

@@ -504,6 +504,12 @@ void AdnlPeerPairImpl::create_channel(pubkeys::Ed25519 pub, td::uint32 date) {
 
 void AdnlPeerPairImpl::process_message(const adnlmessage::AdnlMessageCreateChannel &message) {
   create_channel(message.key(), message.date());
+  if (respond_to_channel_create_after_.is_in_past()) {
+    respond_to_channel_create_after_ = td::Timestamp::in(td::Random::fast(1.0, 2.0));
+    std::vector<OutboundAdnlMessage> messages;
+    messages.emplace_back(adnlmessage::AdnlMessageNop{}, 0);
+    send_messages(std::move(messages));
+  }
 }
 
 void AdnlPeerPairImpl::process_message(const adnlmessage::AdnlMessageConfirmChannel &message) {

adnl/adnl-peer.hpp

@@ -214,6 +214,7 @@ class AdnlPeerPairImpl : public AdnlPeerPair {
   pubkeys::Ed25519 channel_pub_;
   td::int32 channel_pk_date_;
   td::actor::ActorOwn<AdnlChannel> channel_;
+  td::Timestamp respond_to_channel_create_after_;
 
   td::uint64 in_seqno_ = 0;
   td::uint64 out_seqno_ = 0;

tdutils/td/utils/port/Stat.cpp

@@ -413,4 +413,52 @@ Result<CpuStat> cpu_stat() {
 #endif
 }
 
+Result<uint64> get_total_ram() {
+#if TD_LINUX
+  TRY_RESULT(fd, FileFd::open("/proc/meminfo", FileFd::Read));
+  SCOPE_EXIT {
+    fd.close();
+  };
+  constexpr int TMEM_SIZE = 10000;
+  char mem[TMEM_SIZE];
+  TRY_RESULT(size, fd.read(MutableSlice(mem, TMEM_SIZE - 1)));
+  if (size >= TMEM_SIZE - 1) {
+    return Status::Error("Failed for read /proc/meminfo");
+  }
+  mem[size] = 0;
+  const char* s = mem;
+  while (*s) {
+    const char *name_begin = s;
+    while (*s != 0 && *s != '\n') {
+      s++;
+    }
+    auto name_end = name_begin;
+    while (is_alpha(*name_end)) {
+      name_end++;
+    }
+    Slice name(name_begin, name_end);
+    if (name == "MemTotal") {
+      Slice value(name_end, s);
+      if (!value.empty() && value[0] == ':') {
+        value.remove_prefix(1);
+      }
+      value = trim(value);
+      value = split(value).first;
+      TRY_RESULT_PREFIX(mem, to_integer_safe<uint64>(value), "Invalid value of MemTotal");
+      if (mem >= 1ULL << (64 - 10)) {
+        return Status::Error("Invalid value of MemTotal");
+      }
+      return mem * 1024;
+    }
+    if (*s == 0) {
+      break;
+    }
+    s++;
+  }
+  return Status::Error("No MemTotal in /proc/meminfo");
+#else
+  return Status::Error("Not supported");
+#endif
+}
+
 }  // namespace td

tdutils/td/utils/port/Stat.h

@@ -64,4 +64,6 @@ Status update_atime(CSlice path) TD_WARN_UNUSED_RESULT;
 
 #endif
 
+Result<uint64> get_total_ram() TD_WARN_UNUSED_RESULT;
+
 }  // namespace td

validator-engine/CMakeLists.txt

@@ -14,5 +14,9 @@ add_executable(validator-engine ${VALIDATOR_ENGINE_SOURCE})
 target_link_libraries(validator-engine overlay tdutils tdactor adnl tl_api dht
   rldp rldp2 catchain validatorsession full-node validator ton_validator validator
   fift-lib memprof git ${JEMALLOC_LIBRARIES})
+if (JEMALLOC_FOUND)
+  target_include_directories(validator-engine PRIVATE ${JEMALLOC_INCLUDE_DIR})
+  target_compile_definitions(validator-engine PRIVATE -DTON_USE_JEMALLOC=1)
+endif()
 
 install(TARGETS validator-engine RUNTIME DESTINATION bin)

validator-engine/validator-engine.cpp

@@ -75,6 +75,10 @@
 #include "block/precompiled-smc/PrecompiledSmartContract.h"
 #include "interfaces/validator-manager.h"
 
+#if TON_USE_JEMALLOC
+#include <jemalloc/jemalloc.h>
+#endif
+
 Config::Config() {
   out_port = 3278;
   full_node = ton::PublicKeyHash::zero();
@@ -1179,6 +1183,55 @@ class CheckDhtServerStatusQuery : public td::actor::Actor {
   td::Promise<td::BufferSlice> promise_;
 };
 
+#if TON_USE_JEMALLOC
+class JemallocStatsWriter : public td::actor::Actor {
+ public:
+  void start_up() override {
+    alarm();
+  }
+
+  void alarm() override {
+    alarm_timestamp() = td::Timestamp::in(60.0);
+    auto r_stats = get_stats();
+    if (r_stats.is_error()) {
+      LOG(WARNING) << "Jemalloc stats error : " << r_stats.move_as_error();
+    } else {
+      auto s = r_stats.move_as_ok();
+      LOG(WARNING) << "JEMALLOC_STATS : [ timestamp=" << (ton::UnixTime)td::Clocks::system()
+                   << " allocated=" << s.allocated << " active=" << s.active << " metadata=" << s.metadata
+                   << " resident=" << s.resident << " ]";
+    }
+  }
+
+ private:
+  struct JemallocStats {
+    size_t allocated, active, metadata, resident;
+  };
+
+  static td::Result<JemallocStats> get_stats() {
+    size_t sz = sizeof(size_t);
+    static size_t epoch = 1;
+    if (mallctl("epoch", &epoch, &sz, &epoch, sz)) {
+      return td::Status::Error("Failed to refrash stats");
+    }
+    JemallocStats stats;
+    if (mallctl("stats.allocated", &stats.allocated, &sz, nullptr, 0)) {
+      return td::Status::Error("Cannot get stats.allocated");
+    }
+    if (mallctl("stats.active", &stats.active, &sz, nullptr, 0)) {
+      return td::Status::Error("Cannot get stats.active");
+    }
+    if (mallctl("stats.metadata", &stats.metadata, &sz, nullptr, 0)) {
+      return td::Status::Error("Cannot get stats.metadata");
+    }
+    if (mallctl("stats.resident", &stats.resident, &sz, nullptr, 0)) {
+      return td::Status::Error("Cannot get stats.resident");
+    }
+    return stats;
+  }
+};
+#endif
+
 void ValidatorEngine::set_local_config(std::string str) {
   local_config_ = str;
 }
@@ -1202,6 +1255,9 @@ void ValidatorEngine::schedule_shutdown(double at) {
 }
 void ValidatorEngine::start_up() {
   alarm_timestamp() = td::Timestamp::in(1.0 + td::Random::fast(0, 100) * 0.01);
+#if TON_USE_JEMALLOC
+  td::actor::create_actor<JemallocStatsWriter>("mem-stat").release();
+#endif
 }
 
 void ValidatorEngine::alarm() {
@@ -1412,6 +1468,18 @@ td::Status ValidatorEngine::load_global_config() {
   }
   validator_options_.write().set_hardforks(std::move(h));
 
+  auto r_total_ram = td::get_total_ram();
+  if (r_total_ram.is_error()) {
+    LOG(ERROR) << "Failed to get total RAM size: " << r_total_ram.move_as_error();
+  } else {
+    td::uint64 total_ram = r_total_ram.move_as_ok();
+    LOG(WARNING) << "Total RAM = " << td::format::as_size(total_ram);
+    if (total_ram >= (90ULL << 30)) {
+      fast_state_serializer_enabled_ = true;
+    }
+  }
+  validator_options_.write().set_fast_state_serializer_enabled(fast_state_serializer_enabled_);
+
   return td::Status::OK();
 }
 
@@ -3894,7 +3962,7 @@ void need_scheduler_status(int sig) {
   need_scheduler_status_flag.store(true);
 }
 
-void dump_memory_stats() {
+void dump_memprof_stats() {
   if (!is_memprof_on()) {
     return;
   }
@@ -3919,8 +3987,20 @@ void dump_memory_stats() {
   LOG(WARNING) << td::tag("fast_backtrace_success_rate", get_fast_backtrace_success_rate());
 }
 
+void dump_jemalloc_prof() {
+#if TON_USE_JEMALLOC
+  const char *filename = "/tmp/validator-jemalloc.dump";
+  if (mallctl("prof.dump", nullptr, nullptr, &filename, sizeof(const char *)) == 0) {
+    LOG(ERROR) << "Written jemalloc dump to " << filename;
+  } else {
+    LOG(ERROR) << "Failed to write jemalloc dump to " << filename;
+  }
+#endif
+}
+
 void dump_stats() {
-  dump_memory_stats();
+  dump_memprof_stats();
+  dump_jemalloc_prof();
   LOG(WARNING) << td::NamedThreadSafeCounter::get_default();
 }
 
@@ -4158,6 +4238,13 @@ 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_option(
+      '\0', "fast-state-serializer",
+      "faster persistent state serializer, but requires more RAM (enabled automatically on machines with >= 90GB RAM)",
+      [&]() {
+        acts.push_back(
+            [&x]() { td::actor::send_closure(x, &ValidatorEngine::set_fast_state_serializer_enabled, true); });
+      });
   auto S = p.run(argc, argv);
   if (S.is_error()) {
     LOG(ERROR) << "failed to parse options: " << S.move_as_error();

validator-engine/validator-engine.hpp

@@ -220,6 +220,7 @@ class ValidatorEngine : public td::actor::Actor {
   bool started_ = false;
   ton::BlockSeqno truncate_seqno_{0};
   std::string session_logs_file_;
+  bool fast_state_serializer_enabled_ = false;
 
   std::set<ton::CatchainSeqno> unsafe_catchains_;
   std::map<ton::BlockSeqno, std::pair<ton::CatchainSeqno, td::uint32>> unsafe_catchain_rotations_;
@@ -299,6 +300,9 @@ class ValidatorEngine : public td::actor::Actor {
   void set_catchain_max_block_delay(double value) {
     catchain_max_block_delay_ = value;
   }
+  void set_fast_state_serializer_enabled(bool value) {
+    fast_state_serializer_enabled_ = value;
+  }
   void start_up() override;
   ValidatorEngine() {
   }

validator/state-serializer.cpp

@@ -253,6 +253,9 @@ class CachedCellDbReader : public vm::CellDbReader {
 };
 
 void AsyncStateSerializer::prepare_previous_state_cache(ShardIdFull shard) {
+  if (!opts_->get_fast_state_serializer_enabled()) {
+    return;
+  }
   std::vector<ShardIdFull> prev_shards;
   for (const auto& [_, prev_shard] : previous_state_files_) {
     if (shard_intersects(shard, prev_shard)) {

validator/validator-options.hpp

@@ -147,6 +147,9 @@ struct ValidatorManagerOptionsImpl : public ValidatorManagerOptions {
   td::Ref<CollatorOptions> get_collator_options() const override {
     return collator_options_;
   }
+  bool get_fast_state_serializer_enabled() const override {
+    return fast_state_serializer_enabled_;
+  }
 
   void set_zero_block_id(BlockIdExt block_id) override {
     zero_block_id_ = block_id;
@@ -233,6 +236,9 @@ struct ValidatorManagerOptionsImpl : public ValidatorManagerOptions {
   void set_collator_options(td::Ref<CollatorOptions> value) override {
     collator_options_ = std::move(value);
   }
+  void set_fast_state_serializer_enabled(bool value) override {
+    fast_state_serializer_enabled_ = value;
+  }
 
   ValidatorManagerOptionsImpl *make_copy() const override {
     return new ValidatorManagerOptionsImpl(*this);
@@ -286,6 +292,7 @@ struct ValidatorManagerOptionsImpl : public ValidatorManagerOptions {
   td::optional<double> catchain_max_block_delay_;
   bool state_serializer_enabled_ = true;
   td::Ref<CollatorOptions> collator_options_{true};
+  bool fast_state_serializer_enabled_ = false;
 };
 
 }  // namespace validator

validator/validator.h

@@ -107,6 +107,7 @@ struct ValidatorManagerOptions : public td::CntObject {
   virtual td::optional<double> get_catchain_max_block_delay() 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;
 
   virtual void set_zero_block_id(BlockIdExt block_id) = 0;
   virtual void set_init_block_id(BlockIdExt block_id) = 0;
@@ -137,6 +138,7 @@ struct ValidatorManagerOptions : public td::CntObject {
   virtual void set_catchain_max_block_delay(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;
 
   static td::Ref<ValidatorManagerOptions> create(
       BlockIdExt zero_block_id, BlockIdExt init_block_id,