Improve large OutMsgQueue clearance (ton-blockchain#822)

* Improve Collator::opt_msg_queue_cleanup, increase collator timeout * Disable importing ext msgs if queue is too big * Extend timeout in collator if previous block is too old --------- Co-authored-by: SpyCheese <[email protected]>
SpyCheese · Dec 6, 2023 · 7fcf267 · 7fcf267
1 parent 51baec4
commit 7fcf267
Show file tree

Hide file tree

Showing 6 changed files with 105 additions and 25 deletions.
diff --git a/crypto/vm/dict.cpp b/crypto/vm/dict.cpp
@@ -21,6 +21,7 @@
 #include "vm/cellslice.h"
 #include "vm/stack.hpp"
 #include "common/bitstring.h"
+#include "td/utils/Random.h"
 
 #include "td/utils/bits.h"
 
@@ -2007,7 +2008,7 @@ bool DictionaryFixed::combine_with(DictionaryFixed& dict2) {
 
 bool DictionaryFixed::dict_check_for_each(Ref<Cell> dict, td::BitPtr key_buffer, int n, int total_key_len,
                                           const DictionaryFixed::foreach_func_t& foreach_func,
-                                          bool invert_first) const {
+                                          bool invert_first, bool shuffle) const {
   if (dict.is_null()) {
     return true;
   }
@@ -2026,26 +2027,29 @@ bool DictionaryFixed::dict_check_for_each(Ref<Cell> dict, td::BitPtr key_buffer,
   key_buffer += l + 1;
   if (l) {
     invert_first = false;
-  } else if (invert_first) {
+  }
+  bool invert = shuffle ? td::Random::fast(0, 1) == 1: invert_first;
+  if (invert) {
     std::swap(c1, c2);
   }
-  key_buffer[-1] = invert_first;
+  key_buffer[-1] = invert;
   // recursive check_foreach applied to both children
-  if (!dict_check_for_each(std::move(c1), key_buffer, n - l - 1, total_key_len, foreach_func)) {
+  if (!dict_check_for_each(std::move(c1), key_buffer, n - l - 1, total_key_len, foreach_func, false, shuffle)) {
     return false;
   }
-  key_buffer[-1] = !invert_first;
-  return dict_check_for_each(std::move(c2), key_buffer, n - l - 1, total_key_len, foreach_func);
+  key_buffer[-1] = !invert;
+  return dict_check_for_each(std::move(c2), key_buffer, n - l - 1, total_key_len, foreach_func, false, shuffle);
 }
 
-bool DictionaryFixed::check_for_each(const foreach_func_t& foreach_func, bool invert_first) {
+bool DictionaryFixed::check_for_each(const foreach_func_t& foreach_func, bool invert_first, bool shuffle) {
   force_validate();
   if (is_empty()) {
     return true;
   }
   int key_len = get_key_bits();
   unsigned char key_buffer[max_key_bytes];
-  return dict_check_for_each(get_root_cell(), td::BitPtr{key_buffer}, key_len, key_len, foreach_func, invert_first);
+  return dict_check_for_each(get_root_cell(), td::BitPtr{key_buffer}, key_len, key_len, foreach_func, invert_first,
+                             shuffle);
 }
 
 static inline bool set_bit(td::BitPtr ptr, bool value = true) {

diff --git a/crypto/vm/dict.h b/crypto/vm/dict.h
@@ -223,7 +223,7 @@ class DictionaryFixed : public DictionaryBase {
   int get_common_prefix(td::BitPtr buffer, unsigned buffer_len);
   bool cut_prefix_subdict(td::ConstBitPtr prefix, int prefix_len, bool remove_prefix = false);
   Ref<vm::Cell> extract_prefix_subdict_root(td::ConstBitPtr prefix, int prefix_len, bool remove_prefix = false);
-  bool check_for_each(const foreach_func_t& foreach_func, bool invert_first = false);
+  bool check_for_each(const foreach_func_t& foreach_func, bool invert_first = false, bool shuffle = false);
   int filter(filter_func_t check);
   bool combine_with(DictionaryFixed& dict2, const combine_func_t& combine_func, int mode = 0);
   bool combine_with(DictionaryFixed& dict2, const simple_combine_func_t& simple_combine_func, int mode = 0);
@@ -292,7 +292,7 @@ class DictionaryFixed : public DictionaryBase {
   std::pair<Ref<Cell>, bool> extract_prefix_subdict_internal(Ref<Cell> dict, td::ConstBitPtr prefix, int prefix_len,
                                                              bool remove_prefix = false) const;
   bool dict_check_for_each(Ref<Cell> dict, td::BitPtr key_buffer, int n, int total_key_len,
-                           const foreach_func_t& foreach_func, bool invert_first = false) const;
+                           const foreach_func_t& foreach_func, bool invert_first = false, bool shuffle = false) const;
   std::pair<Ref<Cell>, int> dict_filter(Ref<Cell> dict, td::BitPtr key, int n, const filter_func_t& check_leaf,
                                         int& skip_rest) const;
   Ref<Cell> dict_combine_with(Ref<Cell> dict1, Ref<Cell> dict2, td::BitPtr key_buffer, int n, int total_key_len,

diff --git a/tdutils/td/utils/Time.h b/tdutils/td/utils/Time.h
@@ -110,6 +110,7 @@ class Timestamp {
   }
 
   friend bool operator==(Timestamp a, Timestamp b);
+  friend Timestamp &operator+=(Timestamp &a, double b);
 
  private:
   double at_{0};
@@ -122,6 +123,11 @@ inline bool operator<(const Timestamp &a, const Timestamp &b) {
   return a.at() < b.at();
 }
 
+inline Timestamp &operator+=(Timestamp &a, double b) {
+  a.at_ += b;
+  return a;
+}
+
 template <class StorerT>
 void store(const Timestamp &timestamp, StorerT &storer) {
   storer.store_binary(timestamp.at() - Time::now() + Clocks::system());

diff --git a/validator/impl/collator.cpp b/validator/impl/collator.cpp
@@ -703,6 +703,11 @@ bool Collator::unpack_last_mc_state() {
     return fatal_error(limits.move_as_error());
   }
   block_limits_ = limits.move_as_ok();
+  if (!is_masterchain()) {
+    // block_limits_->bytes = {131072 / 3, 524288 / 3, 1048576 / 3};
+    // block_limits_->gas = {2000000 / 3, 10000000 / 3, 20000000 / 3};
+    block_limits_->lt_delta = {20, 180, 200};
+  }
   LOG(DEBUG) << "block limits: bytes [" << block_limits_->bytes.underload() << ", " << block_limits_->bytes.soft()
              << ", " << block_limits_->bytes.hard() << "]";
   LOG(DEBUG) << "block limits: gas [" << block_limits_->gas.underload() << ", " << block_limits_->gas.soft() << ", "
@@ -1818,6 +1823,17 @@ bool Collator::init_utime() {
   CHECK(config_);
   // consider unixtime and lt from previous block(s) of the same shardchain
   prev_now_ = prev_state_utime_;
+  // Extend collator timeout if previous block is too old
+  td::Timestamp new_timeout = td::Timestamp::in(std::min(30.0, (td::Clocks::system() - (double)prev_now_) / 2));
+  if (timeout < new_timeout) {
+    double add = new_timeout.at() - timeout.at();
+    timeout = new_timeout;
+    queue_cleanup_timeout_ += add;
+    soft_timeout_ += add;
+    medium_timeout_ += add;
+    alarm_timestamp() = timeout;
+  }
+
   auto prev = std::max<td::uint32>(config_->utime, prev_now_);
   now_ = std::max<td::uint32>(prev + 1, (unsigned)std::time(nullptr));
   if (now_ > now_upper_limit_) {
@@ -2170,18 +2186,30 @@ bool Collator::out_msg_queue_cleanup() {
     }
   }
 
-  auto res = out_msg_queue_->filter([&](vm::CellSlice& cs, td::ConstBitPtr key, int n) -> int {
+  auto queue_root = out_msg_queue_->get_root_cell();
+  if (queue_root.is_null()) {
+    LOG(DEBUG) << "out_msg_queue is empty";
+    return true;
+  }
+  auto old_out_msg_queue = std::make_unique<vm::AugmentedDictionary>(queue_root, 352, block::tlb::aug_OutMsgQueue);
+
+  int deleted = 0;
+  int total = 0;
+  bool fail = false;
+  old_out_msg_queue->check_for_each([&](Ref<vm::CellSlice> value, td::ConstBitPtr key, int n) -> bool {
+    ++total;
     assert(n == 352);
+    vm::CellSlice& cs = value.write();
     // LOG(DEBUG) << "key is " << key.to_hex(n);
     if (queue_cleanup_timeout_.is_in_past(td::Timestamp::now())) {
       LOG(WARNING) << "cleaning up outbound queue takes too long, ending";
       outq_cleanup_partial_ = true;
-      return (1 << 30) + 1;  // retain all remaining outbound queue entries including this one without processing
+      return false;  // retain all remaining outbound queue entries including this one without processing
     }
     if (block_full_) {
       LOG(WARNING) << "BLOCK FULL while cleaning up outbound queue, cleanup completed only partially";
       outq_cleanup_partial_ = true;
-      return (1 << 30) + 1;  // retain all remaining outbound queue entries including this one without processing
+      return false;  // retain all remaining outbound queue entries including this one without processing
     }
     block::EnqueuedMsgDescr enq_msg_descr;
     unsigned long long created_lt;
@@ -2190,7 +2218,8 @@ bool Collator::out_msg_queue_cleanup() {
           && enq_msg_descr.check_key(key)      // check key
           && enq_msg_descr.lt_ == created_lt)) {
       LOG(ERROR) << "cannot unpack EnqueuedMsg with key " << key.to_hex(n);
-      return -1;
+      fail = true;
+      return false;
     }
     LOG(DEBUG) << "scanning outbound message with (lt,hash)=(" << enq_msg_descr.lt_ << ","
                << enq_msg_descr.hash_.to_hex() << ") enqueued_lt=" << enq_msg_descr.enqueued_lt_;
@@ -2208,22 +2237,29 @@ bool Collator::out_msg_queue_cleanup() {
     if (delivered) {
       LOG(DEBUG) << "outbound message with (lt,hash)=(" << enq_msg_descr.lt_ << "," << enq_msg_descr.hash_.to_hex()
                  << ") enqueued_lt=" << enq_msg_descr.enqueued_lt_ << " has been already delivered, dequeueing";
+      ++deleted;
+      out_msg_queue_->lookup_delete_with_extra(key, n);
       if (!dequeue_message(std::move(enq_msg_descr.msg_env_), deliver_lt)) {
         fatal_error(PSTRING() << "cannot dequeue outbound message with (lt,hash)=(" << enq_msg_descr.lt_ << ","
                               << enq_msg_descr.hash_.to_hex() << ") by inserting a msg_export_deq record");
-        return -1;
+        fail = true;
+        return false;
       }
       register_out_msg_queue_op();
       if (!block_limit_status_->fits(block::ParamLimits::cl_normal)) {
         block_full_ = true;
       }
     }
-    return !delivered;
-  });
-  LOG(DEBUG) << "deleted " << res << " messages from out_msg_queue";
-  if (res < 0) {
+    return true;
+  }, false, true /* random order */);
+  LOG(INFO) << "deleted " << deleted << " messages from out_msg_queue, processed " << total << " messages in total";
+  if (fail) {
     return fatal_error("error scanning/updating OutMsgQueue");
   }
+  if (outq_cleanup_partial_ || total > 8000) {
+    LOG(INFO) << "out_msg_queue too big, skipping importing external messages";
+    skip_extmsg_ = true;
+  }
   auto rt = out_msg_queue_->get_root();
   if (verbosity >= 2) {
     std::cerr << "new out_msg_queue is ";
@@ -3277,6 +3313,10 @@ bool Collator::process_inbound_external_messages() {
     LOG(INFO) << "skipping processing of inbound external messages";
     return true;
   }
+  if (out_msg_queue_->get_root_cell().not_null() && out_msg_queue_->get_root_cell()->get_depth() > 12) {
+    LOG(INFO) << "skipping processing of inbound external messages: out msg queue is too big";
+    return true;
+  }
   bool full = !block_limit_status_->fits(block::ParamLimits::cl_soft);
   for (auto& ext_msg_pair : ext_msg_list_) {
     if (full) {

diff --git a/validator/validator-group.cpp b/validator/validator-group.cpp
@@ -78,6 +78,17 @@ void ValidatorGroup::validate_block_candidate(td::uint32 round_id, BlockCandidat
     promise.set_error(td::Status::Error(ErrorCode::notready, "too old"));
     return;
   }
+
+  auto next_block_id = create_next_block_id(block.id.root_hash, block.id.file_hash);
+  block.id = next_block_id;
+
+  CacheKey cache_key = block_to_cache_key(block);
+  auto it = approved_candidates_cache_.find(cache_key);
+  if (it != approved_candidates_cache_.end()) {
+    promise.set_result(it->second);
+    return;
+  }
+
   auto P = td::PromiseCreator::lambda([SelfId = actor_id(this), round_id, block = block.clone(),
                                        promise = std::move(promise)](td::Result<ValidateCandidateResult> R) mutable {
     if (R.is_error()) {
@@ -93,24 +104,32 @@ void ValidatorGroup::validate_block_candidate(td::uint32 round_id, BlockCandidat
           td::Timestamp::in(0.1));
     } else {
       auto v = R.move_as_ok();
-      v.visit(td::overloaded([&](UnixTime ts) { promise.set_result(ts); },
-                             [&](CandidateReject reject) {
-                               promise.set_error(td::Status::Error(ErrorCode::protoviolation,
-                                                                   PSTRING() << "bad candidate: " << reject.reason));
-                             }));
+      v.visit(td::overloaded(
+          [&](UnixTime ts) {
+            td::actor::send_closure(SelfId, &ValidatorGroup::update_approve_cache, block_to_cache_key(block),
+                                    ts);
+            promise.set_result(ts);
+          },
+          [&](CandidateReject reject) {
+            promise.set_error(
+                td::Status::Error(ErrorCode::protoviolation, PSTRING() << "bad candidate: " << reject.reason));
+          }));
     }
   });
   if (!started_) {
     P.set_error(td::Status::Error(ErrorCode::notready, "validator group not started"));
     return;
   }
-  auto next_block_id = create_next_block_id(block.id.root_hash, block.id.file_hash);
   VLOG(VALIDATOR_DEBUG) << "validating block candidate " << next_block_id;
   block.id = next_block_id;
   run_validate_query(shard_, min_ts_, min_masterchain_block_id_, prev_block_ids_, std::move(block), validator_set_,
                      manager_, td::Timestamp::in(15.0), std::move(P));
 }
 
+void ValidatorGroup::update_approve_cache(CacheKey key, UnixTime value) {
+  approved_candidates_cache_[key] = value;
+}
+
 void ValidatorGroup::accept_block_candidate(td::uint32 round_id, PublicKeyHash src, td::BufferSlice block_data,
                                             RootHash root_hash, FileHash file_hash,
                                             std::vector<BlockSignature> signatures,
@@ -155,6 +174,7 @@ void ValidatorGroup::accept_block_candidate(td::uint32 round_id, PublicKeyHash s
                          std::move(approve_sig_set), src == local_id_, manager_, std::move(P));
   prev_block_ids_ = std::vector<BlockIdExt>{next_block_id};
   cached_collated_block_ = nullptr;
+  approved_candidates_cache_.clear();
 }
 
 void ValidatorGroup::retry_accept_block_query(BlockIdExt block_id, td::Ref<BlockData> block,
@@ -310,6 +330,7 @@ void ValidatorGroup::start(std::vector<BlockIdExt> prev, BlockIdExt min_masterch
   min_masterchain_block_id_ = min_masterchain_block_id;
   min_ts_ = min_ts;
   cached_collated_block_ = nullptr;
+  approved_candidates_cache_.clear();
   started_ = true;
 
   if (init_) {

diff --git a/validator/validator-group.hpp b/validator/validator-group.hpp
@@ -126,6 +126,15 @@ class ValidatorGroup : public td::actor::Actor {
   std::shared_ptr<CachedCollatedBlock> cached_collated_block_;
 
   void generated_block_candidate(std::shared_ptr<CachedCollatedBlock> cache, td::Result<BlockCandidate> R);
+
+  typedef std::tuple<td::Bits256, BlockIdExt, FileHash, FileHash> CacheKey;
+  std::map<CacheKey, UnixTime> approved_candidates_cache_;
+
+  void update_approve_cache(CacheKey key, UnixTime value);
+
+  static CacheKey block_to_cache_key(const BlockCandidate& block) {
+    return std::make_tuple(block.pubkey.as_bits256(), block.id, sha256_bits256(block.data), block.collated_file_hash);
+  }
 };
 
 }  // namespace validator