Skip to content

Commit

Permalink
Fix estimating block size, repeat collation on error (ton-blockchain#…
Browse files Browse the repository at this point in the history
…1178)

* Fix extimating block size, repeat collation on error

* Cancel collation when it is non needed
  • Loading branch information
SpyCheese authored Oct 1, 2024
1 parent fc5e71f commit 257cd8c
Show file tree
Hide file tree
Showing 8 changed files with 150 additions and 44 deletions.
6 changes: 6 additions & 0 deletions crypto/block/block.h
Original file line number Diff line number Diff line change
Expand Up @@ -239,6 +239,12 @@ struct ParamLimits {
bool deserialize(vm::CellSlice& cs);
int classify(td::uint64 value) const;
bool fits(unsigned cls, td::uint64 value) const;
void multiply_by(double x) {
CHECK(x > 0.0);
for (td::uint32& y : limits_) {
y = (td::uint32)std::min<double>(y * x, 1e9);
}
}

private:
std::array<td::uint32, limits_cnt> limits_;
Expand Down
5 changes: 4 additions & 1 deletion validator/fabric.h
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,8 @@ namespace ton {

namespace validator {

enum CollateMode { skip_store_candidate = 1 };

td::actor::ActorOwn<Db> create_db_actor(td::actor::ActorId<ValidatorManager> manager, std::string db_root_,
td::Ref<ValidatorManagerOptions> opts);
td::actor::ActorOwn<LiteServerCache> create_liteserver_cache_actor(td::actor::ActorId<ValidatorManager> manager,
Expand Down Expand Up @@ -81,7 +83,8 @@ void run_validate_query(ShardIdFull shard, BlockIdExt min_masterchain_block_id,
void run_collate_query(ShardIdFull shard, const BlockIdExt& min_masterchain_block_id, std::vector<BlockIdExt> prev,
Ed25519_PublicKey creator, td::Ref<ValidatorSet> validator_set,
td::Ref<CollatorOptions> collator_opts, td::actor::ActorId<ValidatorManager> manager,
td::Timestamp timeout, td::Promise<BlockCandidate> promise);
td::Timestamp timeout, td::Promise<BlockCandidate> promise,
td::CancellationToken cancellation_token, unsigned mode, int attempt_idx = 0);
void run_collate_hardfork(ShardIdFull shard, const BlockIdExt& min_masterchain_block_id, std::vector<BlockIdExt> prev,
td::actor::ActorId<ValidatorManager> manager, td::Timestamp timeout,
td::Promise<BlockCandidate> promise);
Expand Down
11 changes: 10 additions & 1 deletion validator/impl/collator-impl.h
Original file line number Diff line number Diff line change
Expand Up @@ -76,6 +76,9 @@ class Collator final : public td::actor::Actor {
td::Timestamp timeout;
td::Timestamp queue_cleanup_timeout_, soft_timeout_, medium_timeout_;
td::Promise<BlockCandidate> main_promise;
unsigned mode_ = 0;
int attempt_idx_;
bool allow_repeat_collation_ = false;
ton::BlockSeqno last_block_seqno{0};
ton::BlockSeqno prev_mc_block_seqno{0};
ton::BlockSeqno new_block_seqno{0};
Expand All @@ -90,7 +93,8 @@ class Collator final : public td::actor::Actor {
public:
Collator(ShardIdFull shard, bool is_hardfork, BlockIdExt min_masterchain_block_id, std::vector<BlockIdExt> prev,
Ref<ValidatorSet> validator_set, Ed25519_PublicKey collator_id, Ref<CollatorOptions> collator_opts,
td::actor::ActorId<ValidatorManager> manager, td::Timestamp timeout, td::Promise<BlockCandidate> promise);
td::actor::ActorId<ValidatorManager> manager, td::Timestamp timeout, td::Promise<BlockCandidate> promise,
td::CancellationToken cancellation_token, unsigned mode, int attempt_idx);
~Collator() override = default;
bool is_busy() const {
return busy_;
Expand Down Expand Up @@ -318,6 +322,7 @@ class Collator final : public td::actor::Actor {
bool insert_out_msg(Ref<vm::Cell> out_msg);
bool insert_out_msg(Ref<vm::Cell> out_msg, td::ConstBitPtr msg_hash);
bool register_out_msg_queue_op(bool force = false);
bool register_dispatch_queue_op(bool force = false);
bool update_min_mc_seqno(ton::BlockSeqno some_mc_seqno);
bool combine_account_transactions();
bool update_public_libraries();
Expand Down Expand Up @@ -348,10 +353,14 @@ class Collator final : public td::actor::Actor {
bool create_block();
Ref<vm::Cell> collate_shard_block_descr_set();
bool create_collated_data();

bool create_block_candidate();
void return_block_candidate(td::Result<td::Unit> saved);
bool update_last_proc_int_msg(const std::pair<ton::LogicalTime, ton::Bits256>& new_lt_hash);

td::CancellationToken cancellation_token_;
bool check_cancelled();

public:
static td::uint32 get_skip_externals_queue_size();

Expand Down
152 changes: 117 additions & 35 deletions validator/impl/collator.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -45,11 +45,13 @@ using td::Ref;
using namespace std::literals::string_literals;

// Don't increase MERGE_MAX_QUEUE_LIMIT too much: merging requires cleaning the whole queue in out_msg_queue_cleanup
static const td::uint32 FORCE_SPLIT_QUEUE_SIZE = 4096;
static const td::uint32 SPLIT_MAX_QUEUE_SIZE = 100000;
static const td::uint32 MERGE_MAX_QUEUE_SIZE = 2047;
static const td::uint32 SKIP_EXTERNALS_QUEUE_SIZE = 8000;
static const int HIGH_PRIORITY_EXTERNAL = 10; // don't skip high priority externals when queue is big
static constexpr td::uint32 FORCE_SPLIT_QUEUE_SIZE = 4096;
static constexpr td::uint32 SPLIT_MAX_QUEUE_SIZE = 100000;
static constexpr td::uint32 MERGE_MAX_QUEUE_SIZE = 2047;
static constexpr td::uint32 SKIP_EXTERNALS_QUEUE_SIZE = 8000;
static constexpr int HIGH_PRIORITY_EXTERNAL = 10; // don't skip high priority externals when queue is big

static constexpr int MAX_ATTEMPTS = 5;

#define DBG(__n) dbg(__n)&&
#define DSTART int __dcnt = 0;
Expand All @@ -74,11 +76,15 @@ static inline bool dbg(int c) {
* @param manager The ActorId of the ValidatorManager.
* @param timeout The timeout for the collator.
* @param promise The promise to return the result.
* @param cancellation_token Token to cancel collation.
* @param mode +1 - skip storing candidate to disk.
* @param attempt_idx The index of the attempt, starting from 0. On later attempts collator decreases block limits and skips some steps.
*/
Collator::Collator(ShardIdFull shard, bool is_hardfork, BlockIdExt min_masterchain_block_id,
std::vector<BlockIdExt> prev, td::Ref<ValidatorSet> validator_set, Ed25519_PublicKey collator_id,
Ref<CollatorOptions> collator_opts, td::actor::ActorId<ValidatorManager> manager,
td::Timestamp timeout, td::Promise<BlockCandidate> promise)
td::Timestamp timeout, td::Promise<BlockCandidate> promise, td::CancellationToken cancellation_token,
unsigned mode, int attempt_idx)
: shard_(shard)
, is_hardfork_(is_hardfork)
, min_mc_block_id{min_masterchain_block_id}
Expand All @@ -93,9 +99,13 @@ Collator::Collator(ShardIdFull shard, bool is_hardfork, BlockIdExt min_mastercha
, soft_timeout_(td::Timestamp::at(timeout.at() - 3.0))
, medium_timeout_(td::Timestamp::at(timeout.at() - 1.5))
, main_promise(std::move(promise))
, perf_timer_("collate", 0.1, [manager](double duration) {
send_closure(manager, &ValidatorManager::add_perf_timer_stat, "collate", duration);
}) {
, mode_(mode)
, attempt_idx_(attempt_idx)
, perf_timer_("collate", 0.1,
[manager](double duration) {
send_closure(manager, &ValidatorManager::add_perf_timer_stat, "collate", duration);
})
, cancellation_token_(std::move(cancellation_token)) {
}

/**
Expand All @@ -107,7 +117,11 @@ Collator::Collator(ShardIdFull shard, bool is_hardfork, BlockIdExt min_mastercha
* The results of these queries are handled by corresponding callback functions.
*/
void Collator::start_up() {
LOG(WARNING) << "Collator for shard " << shard_.to_str() << " started";
LOG(WARNING) << "Collator for shard " << shard_.to_str() << " started"
<< (attempt_idx_ ? PSTRING() << " (attempt #" << attempt_idx_ << ")" : "");
if (!check_cancelled()) {
return;
}
LOG(DEBUG) << "Previous block #1 is " << prev_blocks.at(0).to_str();
if (prev_blocks.size() > 1) {
LOG(DEBUG) << "Previous block #2 is " << prev_blocks.at(1).to_str();
Expand Down Expand Up @@ -340,7 +354,15 @@ bool Collator::fatal_error(td::Status error) {
error.ensure_error();
LOG(ERROR) << "cannot generate block candidate for " << show_shard(shard_) << " : " << error.to_string();
if (busy_) {
main_promise(std::move(error));
if (allow_repeat_collation_ && error.code() != ErrorCode::cancelled && attempt_idx_ + 1 < MAX_ATTEMPTS &&
!is_hardfork_ && !timeout.is_in_past()) {
LOG(WARNING) << "Repeating collation (attempt #" << attempt_idx_ + 1 << ")";
run_collate_query(shard_, min_mc_block_id, prev_blocks, created_by_, validator_set_, collator_opts_, manager,
td::Timestamp::in(10.0), std::move(main_promise), std::move(cancellation_token_), mode_,
attempt_idx_ + 1);
} else {
main_promise(std::move(error));
}
busy_ = false;
}
stop();
Expand Down Expand Up @@ -382,6 +404,9 @@ bool Collator::fatal_error(std::string err_msg, int err_code) {
*/
void Collator::check_pending() {
// LOG(DEBUG) << "pending = " << pending;
if (!check_cancelled()) {
return;
}
if (!pending) {
step = 2;
try {
Expand Down Expand Up @@ -712,6 +737,15 @@ bool Collator::unpack_last_mc_state() {
return fatal_error(limits.move_as_error());
}
block_limits_ = limits.move_as_ok();
if (attempt_idx_ == 3) {
LOG(INFO) << "Attempt #3: bytes, gas limits /= 2";
block_limits_->bytes.multiply_by(0.5);
block_limits_->gas.multiply_by(0.5);
} else if (attempt_idx_ == 4) {
LOG(INFO) << "Attempt #4: bytes, gas limits /= 4";
block_limits_->bytes.multiply_by(0.25);
block_limits_->gas.multiply_by(0.25);
}
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() << ", "
Expand Down Expand Up @@ -2093,6 +2127,7 @@ bool Collator::do_collate() {
if (max_lt == start_lt) {
++max_lt;
}
allow_repeat_collation_ = true;
// NB: interchanged 1.2 and 1.1 (is this always correct?)
// 1.1. re-adjust neighbors' out_msg_queues (for oneself)
if (!add_trivial_neighbor()) {
Expand Down Expand Up @@ -2333,6 +2368,9 @@ bool Collator::out_msg_queue_cleanup() {
LOG(WARNING) << "cleaning up outbound queue takes too long, ending";
break;
}
if (!check_cancelled()) {
return false;
}
if (i == queue_parts.size()) {
i = 0;
}
Expand Down Expand Up @@ -3532,6 +3570,9 @@ bool Collator::process_inbound_internal_messages() {
stats_.limits_log += PSTRING() << "INBOUND_INT_MESSAGES: timeout\n";
break;
}
if (!check_cancelled()) {
return false;
}
auto kv = nb_out_msgs_->extract_cur();
CHECK(kv && kv->msg.not_null());
LOG(DEBUG) << "processing inbound message with (lt,hash)=(" << kv->lt << "," << kv->key.to_hex()
Expand Down Expand Up @@ -3565,6 +3606,10 @@ bool Collator::process_inbound_external_messages() {
LOG(INFO) << "skipping processing of inbound external messages";
return true;
}
if (attempt_idx_ >= 2) {
LOG(INFO) << "Attempt #" << attempt_idx_ << ": skip external messages";
return true;
}
if (out_msg_queue_size_ > SKIP_EXTERNALS_QUEUE_SIZE) {
LOG(INFO) << "skipping processing of inbound external messages (except for high-priority) because out_msg_queue is "
"too big ("
Expand All @@ -3586,6 +3631,9 @@ bool Collator::process_inbound_external_messages() {
stats_.limits_log += PSTRING() << "INBOUND_EXT_MESSAGES: timeout\n";
break;
}
if (!check_cancelled()) {
return false;
}
auto ext_msg = ext_msg_struct.cell;
ton::Bits256 hash{ext_msg->get_hash().bits()};
int r = process_external_message(std::move(ext_msg));
Expand Down Expand Up @@ -3692,6 +3740,10 @@ bool Collator::process_dispatch_queue() {
if (max_per_initiator[iter] == 0 || max_total_count[iter] == 0) {
continue;
}
if (iter > 0 && attempt_idx_ >= 1) {
LOG(INFO) << "Attempt #" << attempt_idx_ << ": skip process_dispatch_queue";
break;
}
vm::AugmentedDictionary cur_dispatch_queue{dispatch_queue_->get_root(), 256, block::tlb::aug_DispatchQueue};
std::map<std::tuple<WorkchainId, StdSmcAddress, LogicalTime>, size_t> count_per_initiator;
size_t total_count = 0;
Expand All @@ -3704,13 +3756,13 @@ bool Collator::process_dispatch_queue() {
stats_.limits_log += PSTRING() << "DISPATCH_QUEUE_STAGE_" << iter << ": "
<< block_full_comment(*block_limit_status_, block::ParamLimits::cl_normal)
<< "\n";
return true;
return register_dispatch_queue_op(true);
}
if (soft_timeout_.is_in_past(td::Timestamp::now())) {
block_full_ = true;
LOG(WARNING) << "soft timeout reached, stop processing dispatch queue";
stats_.limits_log += PSTRING() << "DISPATCH_QUEUE_STAGE_" << iter << ": timeout\n";
return true;
return register_dispatch_queue_op(true);
}
StdSmcAddress src_addr;
td::Ref<vm::CellSlice> account_dispatch_queue;
Expand Down Expand Up @@ -3788,6 +3840,7 @@ bool Collator::process_dispatch_queue() {
if (iter == 0) {
have_unprocessed_account_dispatch_queue_ = false;
}
register_dispatch_queue_op(true);
}
return true;
}
Expand All @@ -3811,12 +3864,7 @@ bool Collator::process_deferred_message(Ref<vm::CellSlice> enq_msg, StdSmcAddres
return fatal_error(PSTRING() << "failed to delete message from DispatchQueue: address=" << src_addr.to_hex()
<< ", lt=" << lt);
}
++dispatch_queue_ops_;
if (!(dispatch_queue_ops_ & 63)) {
if (!block_limit_status_->add_proof(dispatch_queue_->get_root_cell())) {
return false;
}
}
register_dispatch_queue_op();
++sender_generated_messages_count_[src_addr];

LogicalTime enqueued_lt = 0;
Expand Down Expand Up @@ -3909,6 +3957,7 @@ bool Collator::process_deferred_message(Ref<vm::CellSlice> enq_msg, StdSmcAddres
++unprocessed_deferred_messages_[src_addr];
LOG(INFO) << "delivering deferred message from account " << src_addr.to_hex() << ", lt=" << lt
<< ", emitted_lt=" << emitted_lt;
block_limit_status_->add_cell(msg_env);
register_new_msg(std::move(new_msg));
msg_metadata = std::move(env.metadata);
return true;
Expand Down Expand Up @@ -4088,11 +4137,7 @@ bool Collator::enqueue_message(block::NewOutMsg msg, td::RefInt256 fwd_fees_rema
}
++dispatch_dict_size;
dispatch_queue_->set(src_addr, block::pack_account_dispatch_queue(dispatch_dict, dispatch_dict_size));
++dispatch_queue_ops_;
if (!(dispatch_queue_ops_ & 63)) {
return block_limit_status_->add_proof(dispatch_queue_->get_root_cell());
}
return true;
return register_dispatch_queue_op();
}

auto next_hop = block::interpolate_addr(src_prefix, dest_prefix, route_info.second);
Expand Down Expand Up @@ -4134,6 +4179,9 @@ bool Collator::process_new_messages(bool enqueue_only) {
stats_.limits_log += PSTRING() << "NEW_MESSAGES: "
<< block_full_comment(*block_limit_status_, block::ParamLimits::cl_normal) << "\n";
}
if (!check_cancelled()) {
return false;
}
LOG(DEBUG) << "have message with lt=" << msg.lt;
int res = process_one_new_message(std::move(msg), enqueue_only);
if (res < 0) {
Expand Down Expand Up @@ -4973,6 +5021,23 @@ bool Collator::register_out_msg_queue_op(bool force) {
}
}

/**
* Registers a dispatch queue message queue operation.
* Adds the proof to the block limit status every 64 operations.
*
* @param force If true, the proof will always be added to the block limit status.
*
* @returns True if the operation was successfully registered, false otherwise.
*/
bool Collator::register_dispatch_queue_op(bool force) {
++dispatch_queue_ops_;
if (force || !(dispatch_queue_ops_ & 63)) {
return block_limit_status_->add_proof(dispatch_queue_->get_root_cell());
} else {
return true;
}
}

/**
* Creates a new shard state and the Merkle update.
*
Expand Down Expand Up @@ -5098,9 +5163,10 @@ bool Collator::compute_out_msg_queue_info(Ref<vm::Cell>& out_msg_queue_info) {
vm::CellSlice maybe_extra = cb.as_cellslice();
cb.reset();

return register_out_msg_queue_op(true) && out_msg_queue_->append_dict_to_bool(cb) // _ out_queue:OutMsgQueue
&& processed_upto_->pack(cb) // proc_info:ProcessedInfo
&& cb.append_cellslice_bool(maybe_extra) // extra:(Maybe OutMsgQueueExtra)
return register_out_msg_queue_op(true) && register_dispatch_queue_op(true) &&
out_msg_queue_->append_dict_to_bool(cb) // _ out_queue:OutMsgQueue
&& processed_upto_->pack(cb) // proc_info:ProcessedInfo
&& cb.append_cellslice_bool(maybe_extra) // extra:(Maybe OutMsgQueueExtra)
&& cb.finalize_to(out_msg_queue_info);
}

Expand Down Expand Up @@ -5492,14 +5558,18 @@ bool Collator::create_block_candidate() {
<< consensus_config.max_collated_data_size << ")");
}
// 4. save block candidate
LOG(INFO) << "saving new BlockCandidate";
td::actor::send_closure_later(
manager, &ValidatorManager::set_block_candidate, block_candidate->id, block_candidate->clone(),
validator_set_->get_catchain_seqno(), validator_set_->get_validator_set_hash(),
[self = get_self()](td::Result<td::Unit> saved) -> void {
LOG(DEBUG) << "got answer to set_block_candidate";
td::actor::send_closure_later(std::move(self), &Collator::return_block_candidate, std::move(saved));
});
if (mode_ & CollateMode::skip_store_candidate) {
td::actor::send_closure_later(actor_id(this), &Collator::return_block_candidate, td::Unit());
} else {
LOG(INFO) << "saving new BlockCandidate";
td::actor::send_closure_later(
manager, &ValidatorManager::set_block_candidate, block_candidate->id, block_candidate->clone(),
validator_set_->get_catchain_seqno(), validator_set_->get_validator_set_hash(),
[self = get_self()](td::Result<td::Unit> saved) -> void {
LOG(DEBUG) << "got answer to set_block_candidate";
td::actor::send_closure_later(std::move(self), &Collator::return_block_candidate, std::move(saved));
});
}
// 5. communicate about bad and delayed external messages
if (!bad_ext_msgs_.empty() || !delay_ext_msgs_.empty()) {
LOG(INFO) << "sending complete_external_messages() to Manager";
Expand Down Expand Up @@ -5643,6 +5713,18 @@ void Collator::after_get_external_messages(td::Result<std::vector<std::pair<Ref<
check_pending();
}

/**
* Checks if collation was cancelled via cancellation token
*
* @returns false if the collation was cancelled, true otherwise
*/
bool Collator::check_cancelled() {
if (cancellation_token_) {
return fatal_error(td::Status::Error(ErrorCode::cancelled, "cancelled"));
}
return true;
}

td::uint32 Collator::get_skip_externals_queue_size() {
return SKIP_EXTERNALS_QUEUE_SIZE;
}
Expand Down
Loading

0 comments on commit 257cd8c

Please sign in to comment.