feat: Choose writer by cluster communication (#2830)

Fixes #1974
This commit is contained in:
Sergey Kuznetsov
2026-01-21 11:41:26 +00:00
committed by GitHub
parent 0ebbaaadef
commit a72e5a180f
44 changed files with 3579 additions and 537 deletions

View File

@@ -19,11 +19,8 @@
#include "cluster/ClusterCommunicationService.hpp"
#include "cluster/ClioNode.hpp"
#include "data/BackendInterface.hpp"
#include "util/Assert.hpp"
#include "util/Spawn.hpp"
#include "util/log/Logger.hpp"
#include "etl/WriterState.hpp"
#include <boost/asio/bind_cancellation_slot.hpp>
#include <boost/asio/cancellation_type.hpp>
@@ -41,76 +38,32 @@
#include <chrono>
#include <ctime>
#include <latch>
#include <memory>
#include <string>
#include <utility>
#include <vector>
namespace {
constexpr auto kTOTAL_WORKERS = 2uz; // 1 reading and 1 writing worker (coroutines)
} // namespace
namespace cluster {
ClusterCommunicationService::ClusterCommunicationService(
std::shared_ptr<data::BackendInterface> backend,
std::unique_ptr<etl::WriterStateInterface> writerState,
std::chrono::steady_clock::duration readInterval,
std::chrono::steady_clock::duration writeInterval
)
: backend_(std::move(backend))
, readInterval_(readInterval)
, writeInterval_(writeInterval)
, finishedCountdown_(kTOTAL_WORKERS)
, selfData_{ClioNode{
.uuid = std::make_shared<boost::uuids::uuid>(boost::uuids::random_generator{}()),
.updateTime = std::chrono::system_clock::time_point{}
}}
: backend_(ctx_, std::move(backend), writerState->clone(), readInterval, writeInterval)
, writerDecider_(ctx_, std::move(writerState))
{
nodesInClusterMetric_.set(1); // The node always sees itself
isHealthy_ = true;
}
void
ClusterCommunicationService::run()
{
ASSERT(not running_ and not stopped_, "Can only be ran once");
running_ = true;
util::spawn(strand_, [this](boost::asio::yield_context yield) {
boost::asio::steady_timer timer(yield.get_executor());
boost::system::error_code ec;
while (running_) {
timer.expires_after(readInterval_);
auto token = cancelSignal_.slot();
timer.async_wait(boost::asio::bind_cancellation_slot(token, yield[ec]));
if (ec == boost::asio::error::operation_aborted or not running_)
break;
doRead(yield);
}
finishedCountdown_.count_down(1);
backend_.subscribeToNewState([this](auto&&... args) {
metrics_.onNewState(std::forward<decltype(args)>(args)...);
});
util::spawn(strand_, [this](boost::asio::yield_context yield) {
boost::asio::steady_timer timer(yield.get_executor());
boost::system::error_code ec;
while (running_) {
doWrite();
timer.expires_after(writeInterval_);
auto token = cancelSignal_.slot();
timer.async_wait(boost::asio::bind_cancellation_slot(token, yield[ec]));
if (ec == boost::asio::error::operation_aborted or not running_)
break;
}
finishedCountdown_.count_down(1);
backend_.subscribeToNewState([this](auto&&... args) {
writerDecider_.onNewState(std::forward<decltype(args)>(args)...);
});
backend_.run();
}
ClusterCommunicationService::~ClusterCommunicationService()
@@ -121,107 +74,7 @@ ClusterCommunicationService::~ClusterCommunicationService()
void
ClusterCommunicationService::stop()
{
if (stopped_)
return;
stopped_ = true;
// for ASAN to see through concurrency correctly we need to exit all coroutines before joining the ctx
running_ = false;
// cancelSignal_ is not thread safe so we execute emit on the same strand
boost::asio::spawn(
strand_, [this](auto&&) { cancelSignal_.emit(boost::asio::cancellation_type::all); }, boost::asio::use_future
)
.wait();
finishedCountdown_.wait();
ctx_.join();
}
std::shared_ptr<boost::uuids::uuid>
ClusterCommunicationService::selfUuid() const
{
// Uuid never changes so it is safe to copy it without using strand_
return selfData_.uuid;
}
ClioNode
ClusterCommunicationService::selfData() const
{
ClioNode result{};
util::spawn(strand_, [this, &result](boost::asio::yield_context) { result = selfData_; });
return result;
}
std::expected<std::vector<ClioNode>, std::string>
ClusterCommunicationService::clusterData() const
{
if (not isHealthy_) {
return std::unexpected{"Service is not healthy"};
}
std::vector<ClioNode> result;
util::spawn(strand_, [this, &result](boost::asio::yield_context) {
result = otherNodesData_;
result.push_back(selfData_);
});
return result;
}
void
ClusterCommunicationService::doRead(boost::asio::yield_context yield)
{
otherNodesData_.clear();
BackendInterface::ClioNodesDataFetchResult expectedResult;
try {
expectedResult = backend_->fetchClioNodesData(yield);
} catch (...) {
expectedResult = std::unexpected{"Failed to fecth Clio nodes data"};
}
if (!expectedResult.has_value()) {
LOG(log_.error()) << "Failed to fetch nodes data";
isHealthy_ = false;
return;
}
// Create a new vector here to not have partially parsed data in otherNodesData_
std::vector<ClioNode> otherNodesData;
for (auto const& [uuid, nodeDataStr] : expectedResult.value()) {
if (uuid == *selfData_.uuid) {
continue;
}
boost::system::error_code errorCode;
auto const json = boost::json::parse(nodeDataStr, errorCode);
if (errorCode.failed()) {
LOG(log_.error()) << "Error parsing json from DB: " << nodeDataStr;
isHealthy_ = false;
return;
}
auto expectedNodeData = boost::json::try_value_to<ClioNode>(json);
if (expectedNodeData.has_error()) {
LOG(log_.error()) << "Error converting json to ClioNode: " << json;
isHealthy_ = false;
return;
}
*expectedNodeData->uuid = uuid;
otherNodesData.push_back(std::move(expectedNodeData).value());
}
otherNodesData_ = std::move(otherNodesData);
nodesInClusterMetric_.set(otherNodesData_.size() + 1);
isHealthy_ = true;
}
void
ClusterCommunicationService::doWrite()
{
selfData_.updateTime = std::chrono::system_clock::now();
boost::json::value jsonValue{};
boost::json::value_from(selfData_, jsonValue);
backend_->writeNodeMessage(*selfData_.uuid, boost::json::serialize(jsonValue.as_object()));
backend_.stop();
}
} // namespace cluster