This repository has been archived on 2020-05-25. You can view files and clone it, but cannot push or open issues or pull requests.
tdlib-fork/td/telegram/SequenceDispatcher.cpp
levlam 93afb805f1 Add SequenceDispatcher debug.
GitOrigin-RevId: 05f3333ff1caf102da58606af778f4e56a7ec751
2018-06-21 21:54:06 +03:00

268 lines
8.4 KiB
C++

//
// Copyright Aliaksei Levin (levlam@telegram.org), Arseny Smirnov (arseny30@gmail.com) 2014-2018
//
// Distributed under the Boost Software License, Version 1.0. (See accompanying
// file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
//
#include "td/telegram/SequenceDispatcher.h"
#include "td/telegram/Global.h"
#include "td/telegram/net/NetQueryDispatcher.h"
#include "td/utils/format.h"
#include "td/utils/logging.h"
#include "td/utils/misc.h"
#include "td/utils/Status.h"
#include <limits>
namespace td {
/*** Sequence Dispatcher ***/
// Sends queries with invokeAfter.
//
// Each query has three states Start/Wait/Finish
//
// finish_i points to the first not Finish query.
// next_i points to the next query to be sent.
//
// Each query has generation of InvokeAfter chain.
//
// When query is send, its generation is set to current chain generation.
//
// When query is failed and its generation is equals to current generation we must start new chain:
// increment the generation and set next_i to finish_i.
//
// last_sent_i points to the last sent query in current chain.
//
void SequenceDispatcher::send_with_callback(NetQueryPtr query, ActorShared<NetQueryCallback> callback) {
cancel_timeout();
query->debug("Waiting at SequenceDispatcher");
auto query_weak_ref = query.get_weak();
data_.push_back(Data{State::Start, std::move(query_weak_ref), std::move(query), std::move(callback), 0, 0.0, 0.0});
loop();
}
void SequenceDispatcher::check_timeout(Data &data) {
if (data.state_ != State::Start) {
return;
}
data.query_->total_timeout += data.total_timeout_;
data.total_timeout_ = 0;
if (data.query_->total_timeout > data.query_->total_timeout_limit) {
LOG(WARNING) << "Fail " << data.query_ << " to " << data.query_->source_ << " because total_timeout "
<< data.query_->total_timeout << " is greater than total_timeout_limit "
<< data.query_->total_timeout_limit;
data.query_->set_error(Status::Error(
429, PSLICE() << "Too Many Requests: retry after " << static_cast<int32>(data.last_timeout_ + 0.999)));
data.state_ = State::Dummy;
try_resend_query(data, std::move(data.query_));
}
}
void SequenceDispatcher::try_resend_query(Data &data, NetQueryPtr query) {
size_t pos = &data - &data_[0];
CHECK(pos < data_.size());
CHECK(data.state_ == State::Dummy);
data.state_ = State::Wait;
wait_cnt_++;
auto token = pos + id_offset_;
// TODO: is query is ok, use NetQueryCallback::on_result
auto promise = PromiseCreator::lambda([&, self = actor_shared(this, token)](NetQueryPtr query) mutable {
if (!query.empty()) {
send_closure(std::move(self), &SequenceDispatcher::on_resend_ok, std::move(query));
} else {
send_closure(std::move(self), &SequenceDispatcher::on_resend_error);
}
});
send_closure(data.callback_, &NetQueryCallback::on_result_resendable, std::move(query), std::move(promise));
}
SequenceDispatcher::Data &SequenceDispatcher::data_from_token() {
auto token = narrow_cast<size_t>(get_link_token());
auto pos = token - id_offset_;
CHECK(pos < data_.size());
auto &data = data_[pos];
CHECK(data.state_ == State::Wait);
CHECK(wait_cnt_ > 0);
wait_cnt_--;
data.state_ = State::Dummy;
return data;
}
void SequenceDispatcher::on_resend_ok(NetQueryPtr query) {
auto &data = data_from_token();
data.query_ = std::move(query);
do_resend(data);
loop();
}
void SequenceDispatcher::on_resend_error() {
auto &data = data_from_token();
do_finish(data);
loop();
}
void SequenceDispatcher::do_resend(Data &data) {
CHECK(data.state_ == State::Dummy);
data.state_ = State::Start;
if (data.generation_ == generation_) {
next_i_ = finish_i_;
generation_++;
last_sent_i_ = std::numeric_limits<size_t>::max();
}
check_timeout(data);
}
void SequenceDispatcher::do_finish(Data &data) {
CHECK(data.state_ == State::Dummy);
data.state_ = State::Finish;
if (!parent_.empty()) {
send_closure(parent_, &Parent::on_result);
}
}
void SequenceDispatcher::on_result(NetQueryPtr query) {
auto &data = data_from_token();
size_t pos = &data - &data_[0];
CHECK(pos < data_.size());
if (query->last_timeout != 0) {
for (auto i = pos + 1; i < data_.size(); i++) {
data_[i].total_timeout_ += query->last_timeout;
data_[i].last_timeout_ = query->last_timeout;
check_timeout(data_[i]);
}
}
if (query->is_error() && (query->error().code() == NetQuery::ResendInvokeAfter ||
(query->error().code() == 400 && query->error().message() == "MSG_WAIT_FAILED"))) {
VLOG(net_query) << "Resend " << query;
query->resend();
query->debug("Waiting at SequenceDispatcher");
data.query_ = std::move(query);
do_resend(data);
} else {
try_resend_query(data, std::move(query));
}
loop();
}
void SequenceDispatcher::loop() {
for (; finish_i_ < data_.size() && data_[finish_i_].state_ == State::Finish; finish_i_++) {
}
if (next_i_ < finish_i_) {
next_i_ = finish_i_;
}
for (; next_i_ < data_.size() && data_[next_i_].state_ != State::Wait && wait_cnt_ < MAX_SIMULTANEOUS_WAIT;
next_i_++) {
if (data_[next_i_].state_ == State::Finish) {
continue;
}
NetQueryRef invoke_after;
if (last_sent_i_ != std::numeric_limits<size_t>::max() && data_[last_sent_i_].state_ == State::Wait) {
invoke_after = data_[last_sent_i_].net_query_ref_;
}
data_[next_i_].query_->set_invoke_after(invoke_after);
data_[next_i_].query_->last_timeout = 0;
VLOG(net_query) << "Send " << data_[next_i_].query_;
data_[next_i_].query_->debug("send to Td::send_with_callback");
data_[next_i_].query_->set_session_rand(session_rand_);
G()->net_query_dispatcher().dispatch_with_callback(std::move(data_[next_i_].query_),
actor_shared(this, next_i_ + id_offset_));
data_[next_i_].state_ = State::Wait;
wait_cnt_++;
data_[next_i_].generation_ = generation_;
last_sent_i_ = next_i_;
}
try_shrink();
if (finish_i_ == data_.size() && !parent_.empty()) {
set_timeout_in(5);
}
}
void SequenceDispatcher::try_shrink() {
if (finish_i_ * 2 > data_.size() && data_.size() > 5) {
CHECK(finish_i_ <= next_i_);
data_.erase(data_.begin(), data_.begin() + finish_i_);
next_i_ -= finish_i_;
if (last_sent_i_ != std::numeric_limits<size_t>::max()) {
if (last_sent_i_ >= finish_i_) {
last_sent_i_ -= finish_i_;
} else {
last_sent_i_ = std::numeric_limits<size_t>::max();
}
}
id_offset_ += finish_i_;
finish_i_ = 0;
}
}
void SequenceDispatcher::timeout_expired() {
if (finish_i_ != data_.size()) {
return;
}
CHECK(!parent_.empty());
set_timeout_in(1);
LOG(DEBUG) << "SequenceDispatcher ready to close";
send_closure(parent_, &Parent::ready_to_close);
}
void SequenceDispatcher::hangup() {
stop();
}
void SequenceDispatcher::tear_down() {
for (auto &data : data_) {
if (data.query_.empty()) {
continue;
}
data.state_ = State::Dummy;
data.query_->set_error(Status::Error(500, "Internal Server Error: closing"));
do_finish(data);
}
}
void SequenceDispatcher::close_silent() {
for (auto &data : data_) {
if (!data.query_.empty()) {
data.query_->clear();
}
}
stop();
}
/*** MultiSequenceDispatcher ***/
void MultiSequenceDispatcher::send_with_callback(NetQueryPtr query, ActorShared<NetQueryCallback> callback,
uint64 sequence_id) {
CHECK(sequence_id != 0);
auto it_ok = dispatchers_.emplace(sequence_id, Data{0, ActorOwn<SequenceDispatcher>()});
auto &data = it_ok.first->second;
if (it_ok.second) {
LOG(DEBUG) << "Create SequenceDispatcher" << sequence_id;
data.dispatcher_ = create_actor<SequenceDispatcher>("sequence dispatcher", actor_shared(this, sequence_id));
}
data.cnt_++;
query->debug(PSTRING() << "send to SequenceDispatcher " << tag("sequence_id", sequence_id));
send_closure(data.dispatcher_, &SequenceDispatcher::send_with_callback, std::move(query), std::move(callback));
}
void MultiSequenceDispatcher::on_result() {
auto it = dispatchers_.find(get_link_token());
CHECK(it != dispatchers_.end());
it->second.cnt_--;
}
void MultiSequenceDispatcher::ready_to_close() {
auto it = dispatchers_.find(get_link_token());
CHECK(it != dispatchers_.end());
if (it->second.cnt_ == 0) {
LOG(DEBUG) << "Close SequenceDispatcher " << get_link_token();
dispatchers_.erase(it);
}
}
} // namespace td