extensions/kafka/PublishKafka.cpp

/**
 * Licensed to the Apache Software Foundation (ASF) under one or more
 * contributor license agreements.  See the NOTICE file distributed with
 * this work for additional information regarding copyright ownership.
 * The ASF licenses this file to You under the Apache License, Version 2.0
 * (the "License"); you may not use this file except in compliance with
 * the License.  You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
#include "PublishKafka.h"

#include <map>
#include <set>
#include <string>
#include <vector>

#include "core/ProcessContext.h"
#include "core/ProcessSession.h"
#include "core/Resource.h"
#include "range/v3/algorithm/all_of.hpp"
#include "rdkafka_utils.h"
#include "utils/ProcessorConfigUtils.h"
#include "utils/StringUtils.h"
#include "utils/gsl.h"

namespace org::apache::nifi::minifi::processors {

namespace {
// Message
enum class MessageStatus : uint8_t { InFlight, Error, Success };

const char* to_string(const MessageStatus s) {
  switch (s) {
    case MessageStatus::InFlight: return "InFlight";
    case MessageStatus::Error: return "Error";
    case MessageStatus::Success: return "Success";
  }
  throw std::runtime_error{"PublishKafka to_string(MessageStatus): unreachable code"};
}

struct MessageResult {
  MessageStatus status = MessageStatus::InFlight;
  rd_kafka_resp_err_t err_code = RD_KAFKA_RESP_ERR_NO_ERROR;
};

struct FlowFileResult {
  bool flow_file_error = false;
  std::vector<MessageResult> messages;
};
}  // namespace

class PublishKafka::Messages {
  std::mutex mutex_;
  std::condition_variable cv_;
  std::vector<FlowFileResult> flow_files_;
  bool interrupted_ = false;
  const std::shared_ptr<core::logging::Logger> logger_;

  [[nodiscard]] std::string logStatus(const std::unique_lock<std::mutex>& lock) const {
    gsl_Expects(lock.owns_lock());
    const auto messageresult_ok = [](const MessageResult r) {
      return r.status == MessageStatus::Success && r.err_code == RD_KAFKA_RESP_ERR_NO_ERROR;
    };
    const auto messageresult_inflight = [](const MessageResult r) {
      return r.status == MessageStatus::InFlight && r.err_code == RD_KAFKA_RESP_ERR_NO_ERROR;
    };
    std::vector<size_t> flow_files_in_flight;
    std::ostringstream oss;
    if (interrupted_) { oss << "interrupted, "; }
    for (size_t ffi = 0; ffi < flow_files_.size(); ++ffi) {
      const auto& [flow_file_error, messages] = flow_files_[ffi];
      if (!flow_file_error && ranges::all_of(messages, messageresult_ok)) {
        continue;  // don't log the happy path to reduce log spam
      }
      if (!flow_file_error && ranges::all_of(messages, messageresult_inflight)) {
        flow_files_in_flight.push_back(ffi);
        continue;  // don't log fully in-flight flow files here, log them at the end instead
      }
      oss << '[' << ffi << "]: {";
      if (flow_file_error) { oss << "error, "; }
      for (size_t msgi = 0; msgi < messages.size(); ++msgi) {
        const auto& msg = messages[msgi];
        if (messageresult_ok(msg)) { continue; }
        oss << '<' << msgi << ">: (msg " << to_string(msg.status) << ", " << rd_kafka_err2str(msg.err_code) << "), ";
      }
      oss << "}, ";
    }
    oss << "in-flight (" << flow_files_in_flight.size() << "): " << utils::string::join(",", flow_files_in_flight);
    return oss.str();
  }

 public:
  explicit Messages(std::shared_ptr<core::logging::Logger> logger) : logger_{std::move(logger)} {}

  void waitForCompletion() {
    std::unique_lock<std::mutex> lock(mutex_);
    cv_.wait(lock, [this, &lock] {
      if (logger_->should_log(core::logging::LOG_LEVEL::trace)) { logger_->log_trace("{}", logStatus(lock)); }
      return interrupted_ || ranges::all_of(this->flow_files_, [](const FlowFileResult& flow_file) {
        return flow_file.flow_file_error || ranges::all_of(flow_file.messages, [](const MessageResult& message) {
          return message.status != MessageStatus::InFlight;
        });
      });
    });
  }

  template<typename Func>
  auto modifyResult(size_t index, Func fun) -> decltype(fun(flow_files_.at(index))) {
    std::unique_lock<std::mutex> lock(mutex_);
    const auto notifier = gsl::finally([this] { cv_.notify_all(); });
    try {
      return fun(flow_files_.at(index));
    } catch (const std::exception& ex) {
      logger_->log_warn("Messages::modifyResult exception: {}", ex.what());
      throw;
    }
  }

  size_t addFlowFile() {
    std::lock_guard<std::mutex> lock(mutex_);
    flow_files_.emplace_back();
    return flow_files_.size() - 1;
  }

  template<typename Func>
  auto iterateFlowFiles(Func fun) -> std::void_t<decltype(fun(size_t{0}, flow_files_.front()))> {
    std::lock_guard<std::mutex> lock(mutex_);
    for (size_t index = 0U; index < flow_files_.size(); index++) { fun(index, flow_files_[index]); }
  }

  void interrupt() {
    std::unique_lock<std::mutex> lock(mutex_);
    interrupted_ = true;
    cv_.notify_all();
    gsl_Ensures(interrupted_);
  }

  bool wasInterrupted() {
    std::lock_guard<std::mutex> lock(mutex_);
    return interrupted_;
  }
};

namespace {
class ReadCallback {
  void allocate_message_object(const size_t segment_num) const {
    messages_->modifyResult(flow_file_index_, [segment_num](FlowFileResult& flow_file) {
      // allocate message object to be filled in by the callback in produce()
      if (flow_file.messages.size() < segment_num + 1) { flow_file.messages.resize(segment_num + 1); }
      gsl_Ensures(flow_file.messages.size() > segment_num);
    });
  }

  static utils::rd_kafka_headers_unique_ptr make_headers(const core::FlowFile& flow_file,
      const std::optional<utils::Regex>& attribute_name_regex) {
    utils::rd_kafka_headers_unique_ptr result{rd_kafka_headers_new(8)};
    if (!result) { throw std::bad_alloc{}; }

    for (const auto& [attribute_key, attribute_value]: flow_file.getAttributes()) {
      if (attribute_name_regex && utils::regexMatch(attribute_key, *attribute_name_regex)) {
        rd_kafka_header_add(result.get(),
            attribute_key.c_str(),
            gsl::narrow<ssize_t>(attribute_key.size()),
            attribute_value.c_str(),
            gsl::narrow<ssize_t>(attribute_value.size()));
      }
    }
    return result;
  }

  rd_kafka_resp_err_t produce(const size_t segment_num, std::vector<std::byte>& buffer, const size_t buflen) const {
    const std::shared_ptr<PublishKafka::Messages> messages_ptr_copy = this->messages_;
    const auto flow_file_index_copy = this->flow_file_index_;
    const auto logger = logger_;
    const auto produce_callback = [messages_ptr_copy, flow_file_index_copy, segment_num, logger](rd_kafka_t* /*rk*/,
                                      const rd_kafka_message_t* rkmessage) {
      messages_ptr_copy->modifyResult(flow_file_index_copy,
          [segment_num, rkmessage, logger, flow_file_index_copy](FlowFileResult& flow_file) {
            auto& [status, err_code] = flow_file.messages.at(segment_num);
            err_code = rkmessage->err;
            status = err_code == 0 ? MessageStatus::Success : MessageStatus::Error;
            if (err_code != RD_KAFKA_RESP_ERR_NO_ERROR) {
              logger->log_warn("delivery callback, flow file #{}/segment #{}: {}",
                  flow_file_index_copy,
                  segment_num,
                  rd_kafka_err2str(err_code));
            } else {
              logger->log_debug("delivery callback, flow file #{}/segment #{}: success", flow_file_index_copy, segment_num);
            }
          });
    };
    // release()d below, deallocated in PublishKafka::messageDeliveryCallback
    auto callback_ptr =
        std::make_unique<std::function<void(rd_kafka_t*, const rd_kafka_message_t*)>>(std::move(produce_callback));

    allocate_message_object(segment_num);

    auto hdrs_copy = utils::rd_kafka_headers_unique_ptr{rd_kafka_headers_copy(hdrs.get())};
    const auto err = rd_kafka_producev(rk_,
        RD_KAFKA_V_RKT(rkt_),
        RD_KAFKA_V_PARTITION(RD_KAFKA_PARTITION_UA),
        RD_KAFKA_V_MSGFLAGS(RD_KAFKA_MSG_F_COPY),
        RD_KAFKA_V_VALUE(buffer.data(), buflen),
        RD_KAFKA_V_HEADERS(hdrs_copy.get()),
        RD_KAFKA_V_KEY(key_.c_str(), key_.size()),
        RD_KAFKA_V_OPAQUE(callback_ptr.get()),
        RD_KAFKA_V_END);
    if (err == RD_KAFKA_RESP_ERR_NO_ERROR) {
      // in case of failure, messageDeliveryCallback is not called and callback_ptr will delete the callback
      // in case of success, messageDeliveryCallback takes ownership of the callback, so we no longer need to delete it
      std::ignore = hdrs_copy.release();
      // in case of success, rd_kafka_producev takes ownership of the headers, so we no longer need to delete it
      std::ignore = callback_ptr.release();
    }
    logger_->log_trace("produce enqueued flow file #{}/segment #{}: {}",
        flow_file_index_,
        segment_num,
        rd_kafka_err2str(err));
    return err;
  }

 public:
  ReadCallback(const uint64_t max_seg_size, std::string key, rd_kafka_topic_t* const rkt, rd_kafka_t* const rk,
      const core::FlowFile& flowFile, const std::optional<utils::Regex>& attributeNameRegex,
      std::shared_ptr<PublishKafka::Messages> messages, const size_t flow_file_index, const bool fail_empty_flow_files,
      std::shared_ptr<core::logging::Logger> logger)
      : flow_size_(flowFile.getSize()),
        max_seg_size_(max_seg_size == 0 || flow_size_ < max_seg_size ? flow_size_ : max_seg_size),
        key_(std::move(key)),
        rkt_(rkt),
        rk_(rk),
        hdrs(make_headers(flowFile, attributeNameRegex)),
        messages_(std::move(messages)),
        flow_file_index_(flow_file_index),
        fail_empty_flow_files_(fail_empty_flow_files),
        logger_(std::move(logger)) {}

  ReadCallback(ReadCallback&&) = delete;
  ReadCallback(const ReadCallback&) = delete;
  ReadCallback& operator=(ReadCallback&&) = delete;
  ReadCallback& operator=(const ReadCallback&) = delete;
  ~ReadCallback() = default;

  int64_t operator()(const std::shared_ptr<io::InputStream>& stream) {
    std::vector<std::byte> buffer;

    buffer.resize(max_seg_size_);
    read_size_ = 0;
    status_ = 0;
    called_ = true;

    gsl_Expects(max_seg_size_ != 0 || (flow_size_ == 0 && "max_seg_size_ == 0 implies flow_size_ == 0"));
    // ^^ therefore checking max_seg_size_ == 0 handles both division by zero and flow_size_ == 0 cases
    const size_t reserved_msg_capacity = max_seg_size_ == 0 ? 1 : utils::intdiv_ceil(flow_size_, max_seg_size_);
    messages_->modifyResult(flow_file_index_,
        [reserved_msg_capacity](FlowFileResult& flow_file) { flow_file.messages.reserve(reserved_msg_capacity); });

    // If the flow file is empty, we still want to send the message, unless the user wants to fail_empty_flow_files_
    if (flow_size_ == 0 && !fail_empty_flow_files_) {
      const auto err = produce(0, buffer, 0);
      if (err != RD_KAFKA_RESP_ERR_NO_ERROR) {
        status_ = -1;
        error_ = rd_kafka_err2str(err);
      }
      return 0;
    }

    for (size_t segment_num = 0; read_size_ < flow_size_; ++segment_num) {
      const auto readRet = stream->read(buffer);
      if (io::isError(readRet)) {
        status_ = -1;
        error_ = "Failed to read from stream";
        return read_size_;
      }
      if (readRet == 0) { break; }

      if (const auto err = produce(segment_num, buffer, readRet)) {
        messages_->modifyResult(flow_file_index_, [segment_num, err](FlowFileResult& flow_file) {
          auto& [status, err_code] = flow_file.messages.at(segment_num);
          status = MessageStatus::Error;
          err_code = err;
        });
        status_ = -1;
        error_ = rd_kafka_err2str(err);
        return read_size_;
      }
      read_size_ += gsl::narrow<uint32_t>(readRet);
    }
    return read_size_;
  }

  const uint64_t flow_size_ = 0;
  const uint64_t max_seg_size_ = 0;
  const std::string key_;
  rd_kafka_topic_t* const rkt_ = nullptr;
  rd_kafka_t* const rk_ = nullptr;
  const gsl::not_null<utils::rd_kafka_headers_unique_ptr> hdrs;
  const std::shared_ptr<PublishKafka::Messages> messages_;
  const size_t flow_file_index_;
  int status_ = 0;
  std::string error_;
  uint32_t read_size_ = 0;
  bool called_ = false;
  const bool fail_empty_flow_files_ = true;
  const std::shared_ptr<core::logging::Logger> logger_;
};

/**
 * Message delivery report callback using the richer rd_kafka_message_t object.
 */
void messageDeliveryCallback(rd_kafka_t* rk, const rd_kafka_message_t* rkmessage, void* /*opaque*/) {
  if (rkmessage->_private == nullptr) { return; }
  // allocated in ReadCallback::produce
  const auto* const func = static_cast<std::function<void(rd_kafka_t*, const rd_kafka_message_t*)>*>(rkmessage->_private);
  try {
    (*func)(rk, rkmessage);
  } catch (...) {}
  delete func;  // NOLINT(cppcoreguidelines-owning-memory)
}
}  // namespace

void PublishKafka::initialize() {
  setSupportedProperties(Properties);
  setSupportedRelationships(Relationships);
}

void PublishKafka::onSchedule(core::ProcessContext& context, core::ProcessSessionFactory&) {
  interrupted_ = false;

  // Try to get a KafkaConnection
  std::string client_id = utils::parseProperty(context, ClientName);
  std::string brokers = utils::parseProperty(context, SeedBrokers);
  // Get some properties not (only) used directly to set up librdkafka

  // Batch Size
  batch_size_ = utils::parseU64Property(context, BatchSize);
  logger_->log_debug("PublishKafka: Batch Size [{}]", batch_size_);

  // Target Batch Payload Size
  target_batch_payload_size_ = utils::parseDataSizeProperty(context, TargetBatchPayloadSize);
  logger_->log_debug("PublishKafka: Target Batch Payload Size [{}]", target_batch_payload_size_);

  // Max Flow Segment Size
  max_flow_seg_size_ = utils::parseDataSizeProperty(context, MaxFlowSegSize);
  logger_->log_debug("PublishKafka: Max Flow Segment Size [{}]", max_flow_seg_size_);

  // Attributes to Send as Headers
  attributeNameRegex_ = context.getProperty(AttributeNameRegex)
    | utils::transform([](const auto pattern_str) { return utils::Regex{std::move(pattern_str)}; })
    | utils::toOptional();

  key_.brokers_ = brokers;
  key_.client_id_ = client_id;

  conn_ = std::make_unique<KafkaConnection>(key_);
  configureNewConnection(context);

  if (const auto message_key_field = context.getProperty(MessageKeyField); message_key_field && !message_key_field->empty()) {
    logger_->log_error("The {} property is set. This property is DEPRECATED and has no effect; please use Kafka Key instead.", MessageKeyField.name);
  }

  logger_->log_debug("Successfully configured PublishKafka");
}

void PublishKafka::notifyStop() {
  logger_->log_debug("notifyStop called");
  interrupted_ = true;
  {
    // Normally when we need both connection_mutex_ and messages_mutex_, we need to take connection_mutex_ first to avoid a
    // deadlock. It's not possible to do that here, because we need to interrupt the messages while onTrigger is running and
    // holding connection_mutex_. For this reason, we take messages_mutex_ only, interrupt the messages, then release the
    // lock to let a possibly running onTrigger take it and finish. After onTrigger finishes, we can take connection_mutex_
    // and close the connection without needing to wait for message finishes/timeouts in onTrigger. A possible new onTrigger
    // between our critical sections won't produce more messages because we set interrupted_ = true above.
    std::lock_guard<std::mutex> lock(messages_mutex_);
    for (auto& messages: messages_set_) { messages->interrupt(); }
  }
  std::lock_guard<std::mutex> conn_lock(connection_mutex_);
  conn_.reset();
}

bool PublishKafka::configureNewConnection(core::ProcessContext& context) {
  std::array<char, 512U> err_chars{};
  rd_kafka_conf_res_t result = RD_KAFKA_CONF_OK;
  constexpr std::string_view PREFIX_ERROR_MSG = "PublishKafka: configure error result: ";

  utils::rd_kafka_conf_unique_ptr conf_{rd_kafka_conf_new()};
  if (conf_ == nullptr) { throw Exception(PROCESS_SCHEDULE_EXCEPTION, "Failed to create rd_kafka_conf_t object"); }

  const auto* const key = conn_->getKey();

  if (key->brokers_.empty()) { throw Exception(PROCESS_SCHEDULE_EXCEPTION, "There are no brokers"); }
  result = rd_kafka_conf_set(conf_.get(), "bootstrap.servers", key->brokers_.c_str(), err_chars.data(), err_chars.size());
  logger_->log_debug("PublishKafka: bootstrap.servers [{}]", key->brokers_);
  if (result != RD_KAFKA_CONF_OK) {
    auto error_msg = utils::string::join_pack(PREFIX_ERROR_MSG, err_chars.data());
    throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
  }

  if (key->client_id_.empty()) { throw Exception(PROCESS_SCHEDULE_EXCEPTION, "Client id is empty"); }
  result = rd_kafka_conf_set(conf_.get(), "client.id", key->client_id_.c_str(), err_chars.data(), err_chars.size());
  logger_->log_debug("PublishKafka: client.id [{}]", key->client_id_);
  if (result != RD_KAFKA_CONF_OK) {
    auto error_msg = utils::string::join_pack(PREFIX_ERROR_MSG, err_chars.data());
    throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
  }

  if (const auto debug_context = context.getProperty(DebugContexts)) {
    result = rd_kafka_conf_set(conf_.get(), "debug", debug_context->c_str(), err_chars.data(), err_chars.size());
    logger_->log_debug("PublishKafka: debug [{}]", *debug_context);
    if (result != RD_KAFKA_CONF_OK) {
      auto error_msg = utils::string::join_pack(PREFIX_ERROR_MSG, err_chars.data());
      throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
    }
  }

  if (const auto max_message_size = context.getProperty(MaxMessageSize); max_message_size && !max_message_size->empty()) {
    result = rd_kafka_conf_set(conf_.get(), "message.max.bytes", max_message_size->c_str(), err_chars.data(), err_chars.size());
    logger_->log_debug("PublishKafka: message.max.bytes [{}]", *max_message_size);
    if (result != RD_KAFKA_CONF_OK) {
      auto error_msg = utils::string::join_pack(PREFIX_ERROR_MSG, err_chars.data());
      throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
    }
  }

  if (const auto queue_buffer_max_message = utils::parseOptionalU64Property(context, QueueBufferMaxMessage)) {
    if (*queue_buffer_max_message < batch_size_) {
      throw Exception(PROCESS_SCHEDULE_EXCEPTION, "Invalid configuration: Batch Size cannot be larger than Queue Max Message");
    }

    const auto value = std::to_string(*queue_buffer_max_message);
    result = rd_kafka_conf_set(conf_.get(), "queue.buffering.max.messages", value.c_str(), err_chars.data(), err_chars.size());
    logger_->log_debug("PublishKafka: queue.buffering.max.messages [{}]", value);
    if (result != RD_KAFKA_CONF_OK) {
      auto error_msg = utils::string::join_pack(PREFIX_ERROR_MSG, err_chars.data());
      throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
    }
  }

  if (const auto queue_buffer_max_size = utils::parseOptionalDataSizeProperty(context, QueueBufferMaxSize)) {
    auto valInt = *queue_buffer_max_size / 1024;
    auto valueConf = std::to_string(valInt);
    result = rd_kafka_conf_set(conf_.get(), "queue.buffering.max.kbytes", valueConf.c_str(), err_chars.data(), err_chars.size());
    logger_->log_debug("PublishKafka: queue.buffering.max.kbytes [{}]", valueConf);
    if (result != RD_KAFKA_CONF_OK) {
      auto error_msg = utils::string::join_pack(PREFIX_ERROR_MSG, err_chars.data());
      throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
    }
  }

  if (const auto queue_buffer_max_time = utils::parseOptionalDurationProperty(context, QueueBufferMaxTime)) {
    auto valueConf = std::to_string(queue_buffer_max_time->count());
    result = rd_kafka_conf_set(conf_.get(), "queue.buffering.max.ms", valueConf.c_str(), err_chars.data(), err_chars.size());
    logger_->log_debug("PublishKafka: queue.buffering.max.ms [{}]", valueConf);
    if (result != RD_KAFKA_CONF_OK) {
      auto error_msg = utils::string::join_pack(PREFIX_ERROR_MSG, err_chars.data());
      throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
    }
  }

  if (const auto batch_size = utils::parseOptionalU64Property(context, BatchSize)) {
    auto value = std::to_string(*batch_size);
    result = rd_kafka_conf_set(conf_.get(), "batch.num.messages", value.c_str(), err_chars.data(), err_chars.size());
    logger_->log_debug("PublishKafka: batch.num.messages [{}]", value);
    if (result != RD_KAFKA_CONF_OK) {
      auto error_msg = utils::string::join_pack(PREFIX_ERROR_MSG, err_chars.data());
      throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
    }
  }

  if (const auto compress_codec = context.getProperty(CompressCodec); compress_codec && !compress_codec->empty() && *compress_codec != "none") {
    result = rd_kafka_conf_set(conf_.get(), "compression.codec", compress_codec->c_str(), err_chars.data(), err_chars.size());
    logger_->log_debug("PublishKafka: compression.codec [{}]", *compress_codec);
    if (result != RD_KAFKA_CONF_OK) {
      auto error_msg = utils::string::join_pack(PREFIX_ERROR_MSG, err_chars.data());
      throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
    }
  }

  setKafkaAuthenticationParameters(context, gsl::make_not_null(conf_.get()));

  // Add all the dynamic properties as librdkafka configurations
  const auto& dynamic_prop_keys = context.getDynamicPropertyKeys();
  logger_->log_info("PublishKafka registering {} librdkafka dynamic properties", dynamic_prop_keys.size());

  for (const auto& prop_key : dynamic_prop_keys) {
    if (const auto dynamic_property_value = context.getDynamicProperty(prop_key, nullptr); dynamic_property_value && !dynamic_property_value->empty()) {
      logger_->log_debug("PublishKafka: DynamicProperty: [{}] -> [{}]", prop_key, *dynamic_property_value);
      result = rd_kafka_conf_set(conf_.get(), prop_key.c_str(), dynamic_property_value->c_str(), err_chars.data(), err_chars.size());
      if (result != RD_KAFKA_CONF_OK) {
        auto error_msg = utils::string::join_pack(PREFIX_ERROR_MSG, err_chars.data());
        throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
      }
    } else {
      logger_->log_warn(
          "PublishKafka Dynamic Property '{}' is empty and therefore will not "
          "be configured",
          prop_key);
    }
  }

  // Set the delivery callback
  rd_kafka_conf_set_dr_msg_cb(conf_.get(), &messageDeliveryCallback);

  // Set the logger callback
  rd_kafka_conf_set_log_cb(conf_.get(), &KafkaConnection::logCallback);

  // The producer takes ownership of the configuration, we must not free it
  utils::rd_kafka_producer_unique_ptr producer{
      rd_kafka_new(RD_KAFKA_PRODUCER, conf_.release(), err_chars.data(), err_chars.size())};
  if (producer == nullptr) {
    auto error_msg = utils::string::join_pack("Failed to create Kafka producer ", err_chars.data());
    throw Exception(PROCESS_SCHEDULE_EXCEPTION, error_msg);
  }

  conn_->setConnection(std::move(producer));

  return true;
}

bool PublishKafka::createNewTopic(core::ProcessContext& context, const std::string& topic_name,
    const std::shared_ptr<core::FlowFile>& flow_file) const {
  utils::rd_kafka_topic_conf_unique_ptr topic_conf_{rd_kafka_topic_conf_new()};
  if (topic_conf_ == nullptr) {
    logger_->log_error("Failed to create rd_kafka_topic_conf_t object");
    return false;
  }

  rd_kafka_conf_res_t result = RD_KAFKA_CONF_OK;
  std::array<char, 512U> err_chars{};

  if (auto delivery_guarantee = context.getProperty(DeliveryGuarantee, flow_file.get())) {
    /*
     * Because of a previous error in this processor, the default value of this property was "DELIVERY_ONE_NODE".
     * As this is not a valid value for "request.required.acks", the following rd_kafka_topic_conf_set call failed,
     * but because of another error, this failure was silently ignored, meaning that the default value for
     * "request.required.acks" did not change, and thus remained "-1". This means that having "DELIVERY_ONE_NODE" as
     * the value of this property actually caused the processor to wait for delivery ACKs from ALL nodes, instead
     * of just one. In order not to break configurations generated with earlier versions and keep the same behaviour
     * as they had, we have to map "DELIVERY_ONE_NODE" to "-1" here.
     */
    if (*delivery_guarantee == "DELIVERY_ONE_NODE") {
      delivery_guarantee = "-1";
      logger_->log_warn("Using DELIVERY_ONE_NODE as the Delivery Guarantee property is deprecated and is translated to -1 "
                        "(block until message is committed by all in sync replicas) for backwards compatibility. "
                        "If you want to wait for one acknowledgment use '1' as the property.");
    }
    result = rd_kafka_topic_conf_set(topic_conf_.get(), "request.required.acks", delivery_guarantee->c_str(), err_chars.data(), err_chars.size());
    logger_->log_debug("PublishKafka: request.required.acks [{}]", *delivery_guarantee);
    if (result != RD_KAFKA_CONF_OK) {
      logger_->log_error("PublishKafka: configure request.required.acks error result [{}]", err_chars.data());
      return false;
    }
  }

  if (const auto request_timeout = utils::parseOptionalDurationProperty(context, RequestTimeOut)) {
    auto valueConf = std::to_string(request_timeout->count());
    result = rd_kafka_topic_conf_set(topic_conf_.get(), "request.timeout.ms", valueConf.c_str(), err_chars.data(), err_chars.size());
    logger_->log_debug("PublishKafka: request.timeout.ms [{}]", valueConf);
    if (result != RD_KAFKA_CONF_OK) {
      logger_->log_error("PublishKafka: configure request.timeout.ms error result [{}]", err_chars.data());
      return false;
    }
  }

  if (const auto message_timeout = utils::parseOptionalDurationProperty(context, MessageTimeOut)) {
    auto valueConf = std::to_string(message_timeout->count());
    result = rd_kafka_topic_conf_set(topic_conf_.get(), "message.timeout.ms", valueConf.c_str(), err_chars.data(), err_chars.size());
    logger_->log_debug("PublishKafka: message.timeout.ms [{}]", valueConf);
    if (result != RD_KAFKA_CONF_OK) {
      logger_->log_error("PublishKafka: configure message.timeout.ms error result [{}]", err_chars.data());
      return false;
    }
  }

  // The topic takes ownership of the configuration, we must not free it
  utils::rd_kafka_topic_unique_ptr topic_reference{
      rd_kafka_topic_new(conn_->getConnection(), topic_name.c_str(), topic_conf_.release())};
  if (topic_reference == nullptr) {
    const rd_kafka_resp_err_t resp_err = rd_kafka_last_error();
    logger_->log_error("PublishKafka: failed to create topic {}, error: {}", topic_name.c_str(), rd_kafka_err2str(resp_err));
    return false;
  }

  const auto kafka_topic_ref =
      std::make_shared<KafkaTopic>(std::move(topic_reference));  // KafkaTopic takes ownership of topic_reference
  conn_->putTopic(topic_name, kafka_topic_ref);

  return true;
}

std::optional<utils::net::SslData> PublishKafka::getSslData(core::ProcessContext& context) const {
  if (auto result = KafkaProcessorBase::getSslData(context); result) { return result; }

  utils::net::SslData ssl_data;
  if (auto security_ca = context.getProperty(SecurityCA)) { ssl_data.ca_loc = *security_ca; }
  if (auto security_cert = context.getProperty(SecurityCert)) { ssl_data.cert_loc = *security_cert; }
  if (auto security_private_key = context.getProperty(SecurityPrivateKey)) { ssl_data.key_loc = *security_private_key; }
  if (auto security_private_key_pass = context.getProperty(SecurityPrivateKeyPassWord)) {
    ssl_data.key_pw = *security_private_key_pass;
  }
  return ssl_data;
}

void PublishKafka::onTrigger(core::ProcessContext& context, core::ProcessSession& session) {
  // Check whether we have been interrupted
  if (interrupted_) {
    logger_->log_info("The processor has been interrupted, not running onTrigger");
    context.yield();
    return;
  }

  std::lock_guard<std::mutex> lock_connection(connection_mutex_);
  logger_->log_debug("PublishKafka onTrigger");

  // Collect FlowFiles to process
  uint64_t actual_bytes = 0U;
  std::vector<std::shared_ptr<core::FlowFile>> flowFiles;
  for (uint64_t i = 0; i < batch_size_; i++) {
    std::shared_ptr<core::FlowFile> flowFile = session.get();
    if (flowFile == nullptr) { break; }
    actual_bytes += flowFile->getSize();
    flowFiles.emplace_back(std::move(flowFile));
    if (target_batch_payload_size_ != 0U && actual_bytes >= target_batch_payload_size_) { break; }
  }
  if (flowFiles.empty()) {
    context.yield();
    return;
  }
  logger_->log_debug("Processing {} flow files with a total size of {} B", flowFiles.size(), actual_bytes);

  auto messages = std::make_shared<Messages>(logger_);
  // We must add this to the messages set, so that it will be interrupted when notifyStop is called
  {
    std::lock_guard<std::mutex> lock(messages_mutex_);
    messages_set_.emplace(messages);
  }
  // We also have to ensure that it will be removed once we are done with it
  const auto messagesSetGuard = gsl::finally([&]() {
    std::lock_guard<std::mutex> lock(messages_mutex_);
    messages_set_.erase(messages);
  });

  // Process FlowFiles
  for (auto& flowFile: flowFiles) {
    const size_t flow_file_index = messages->addFlowFile();

    // Get Topic (FlowFile-dependent EL property)
    const auto topic = context.getProperty(Topic, flowFile.get());
    if (topic) {
      logger_->log_debug("PublishKafka: topic for flow file {} is '{}'", flowFile->getUUIDStr(), *topic);
    } else {
      logger_->log_error("Flow file {} does not have a valid Topic", flowFile->getUUIDStr());
      messages->modifyResult(flow_file_index,
          [](FlowFileResult& flow_file_result) { flow_file_result.flow_file_error = true; });
      continue;
    }

    // Add topic to the connection if needed
    if (!conn_->hasTopic(*topic)) {
      if (!createNewTopic(context, *topic, flowFile)) {
        logger_->log_error("Failed to add topic {}", *topic);
        messages->modifyResult(flow_file_index, [](FlowFileResult& flow_file_result) {
          flow_file_result.flow_file_error = true;
        });
        continue;
      }
    }

    std::string kafkaKey = context.getProperty(KafkaKey, flowFile.get()).value_or(flowFile->getUUIDStr());

    logger_->log_debug("PublishKafka: Message Key [{}]", kafkaKey);

    auto thisTopic = conn_->getTopic(*topic);
    if (thisTopic == nullptr) {
      logger_->log_error("Topic {} is invalid", *topic);
      messages->modifyResult(flow_file_index, [](FlowFileResult& flow_file_result) {
        flow_file_result.flow_file_error = true;
      });
      continue;
    }

    bool failEmptyFlowFiles = utils::parseBoolProperty(context, FailEmptyFlowFiles);

    ReadCallback callback(max_flow_seg_size_,
        kafkaKey,
        thisTopic->getTopic(),
        conn_->getConnection(),
        *flowFile,
        attributeNameRegex_,
        messages,
        flow_file_index,
        failEmptyFlowFiles,
        logger_);
    session.read(flowFile, std::ref(callback));

    if (!callback.called_) {
      // workaround: call callback since ProcessSession doesn't do so for empty flow files without resource claims
      callback(nullptr);
    }

    if (flowFile->getSize() == 0 && failEmptyFlowFiles) {
      logger_->log_debug(
          "Deprecated behavior, use connections to drop empty flow files! "
          "Failing empty flow file with uuid: {}",
          flowFile->getUUIDStr());
      messages->modifyResult(flow_file_index,
          [](FlowFileResult& flow_file_result) { flow_file_result.flow_file_error = true; });
    }

    if (callback.status_ < 0) {
      logger_->log_error("Failed to send flow to kafka topic {}, error: {}", *topic, callback.error_);
      messages->modifyResult(flow_file_index, [](FlowFileResult& flow_file_result) {
        flow_file_result.flow_file_error = true;
      });
      continue;
    }
  }

  logger_->log_trace("PublishKafka::onTrigger waitForCompletion start");
  messages->waitForCompletion();
  if (messages->wasInterrupted()) {
    logger_->log_warn(
        "Waiting for delivery confirmation was interrupted, some flow files "
        "might be routed to Failure, even if they were successfully "
        "delivered.");
  }
  logger_->log_trace("PublishKafka::onTrigger waitForCompletion finish");

  messages->iterateFlowFiles([&](size_t index, const FlowFileResult& flow_file) {
    bool success = false;
    if (flow_file.flow_file_error) {
      success = false;
    } else {
      success = true;
      for (size_t segment_num = 0; segment_num < flow_file.messages.size(); segment_num++) {
        const auto& message = flow_file.messages[segment_num];
        switch (message.status) {
          case MessageStatus::InFlight:
            success = false;
            logger_->log_error(
                "Waiting for delivery confirmation was interrupted for flow "
                "file {} segment {}",
                flowFiles[index]->getUUIDStr(),
                segment_num);
            break;
          case MessageStatus::Error:
            success = false;
            logger_->log_error("Failed to deliver flow file {} segment {}, error: {}",
                flowFiles[index]->getUUIDStr(),
                segment_num,
                rd_kafka_err2str(message.err_code));
            break;
          case MessageStatus::Success:
            logger_->log_debug("Successfully delivered flow file {} segment {}",
                flowFiles[index]->getUUIDStr(),
                segment_num);
            break;
        }
      }
    }
    if (success) {
      session.transfer(flowFiles[index], Success);
    } else {
      session.penalize(flowFiles[index]);
      session.transfer(flowFiles[index], Failure);
    }
  });
}

REGISTER_RESOURCE(PublishKafka, Processor);

}  // namespace org::apache::nifi::minifi::processors