aos/events/logging/log_namer.cc

#include "aos/events/logging/log_namer.h"

#include <functional>
#include <map>
#include <memory>
#include <string_view>
#include <vector>

#include "absl/strings/str_cat.h"
#include "aos/events/logging/logfile_utils.h"
#include "aos/events/logging/logger_generated.h"
#include "aos/uuid.h"
#include "flatbuffers/flatbuffers.h"
#include "glog/logging.h"

namespace aos {
namespace logger {

void LogNamer::UpdateHeader(
    aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header,
    const UUID &uuid, int parts_index) const {
  header->mutable_message()->mutate_parts_index(parts_index);
  CHECK_EQ(UUID::kStringSize,
           header->mutable_message()->mutable_parts_uuid()->size());
  uuid.CopyTo(reinterpret_cast<char *>(
      header->mutable_message()->mutable_parts_uuid()->Data()));
}

void LocalLogNamer::WriteHeader(
    aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header,
    const Node *node) {
  CHECK_EQ(node, this->node());
  UpdateHeader(header, data_writer_.uuid(), data_writer_.part_number);
  data_writer_.writer->QueueSpan(header->span());
}

NewDataWriter *LocalLogNamer::MakeWriter(const Channel *channel) {
  CHECK(configuration::ChannelIsSendableOnNode(channel, node()))
      << ": " << configuration::CleanedChannelToString(channel);
  return &data_writer_;
}

void LocalLogNamer::Rotate(
    const Node *node,
    aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header) {
  CHECK(node == this->node());
  data_writer_.Rotate();
  UpdateHeader(header, data_writer_.uuid(), data_writer_.part_number);
  data_writer_.writer->QueueSpan(header->span());
}

void LocalLogNamer::WriteConfiguration(
    aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header,
    std::string_view config_sha256) {
  const std::string filename = absl::StrCat(base_name_, config_sha256, ".bfbs");

  std::unique_ptr<DetachedBufferWriter> writer =
      std::make_unique<DetachedBufferWriter>(
          filename, std::make_unique<aos::logger::DummyEncoder>());
  writer->QueueSizedFlatbuffer(header->Release());
}

void LocalLogNamer::Reboot(
    const Node * /*node*/,
    aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> * /*header*/) {
  LOG(FATAL) << "Can't reboot a single node.";
}

NewDataWriter *LocalLogNamer::MakeTimestampWriter(const Channel *channel) {
  CHECK(configuration::ChannelIsReadableOnNode(channel, node_))
      << ": Message is not delivered to this node.";
  CHECK(node_ != nullptr) << ": Can't log timestamps in a single node world";
  CHECK(configuration::ConnectionDeliveryTimeIsLoggedOnNode(channel, node_,
                                                            node_))
      << ": Delivery times aren't logged for this channel on this node.";
  return &data_writer_;
}

NewDataWriter *LocalLogNamer::MakeForwardedTimestampWriter(
    const Channel * /*channel*/, const Node * /*node*/) {
  LOG(FATAL) << "Can't log forwarded timestamps in a singe log file.";
  return nullptr;
}

MultiNodeLogNamer::MultiNodeLogNamer(std::string_view base_name,
                                     const Configuration *configuration,
                                     const Node *node)
    : LogNamer(node),
      base_name_(base_name),
      old_base_name_(),
      configuration_(configuration) {}

MultiNodeLogNamer::~MultiNodeLogNamer() {
  if (!ran_out_of_space_) {
    // This handles renaming temporary files etc.
    Close();
  }
}

void MultiNodeLogNamer::WriteHeader(
    aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header,
    const Node *node) {
  if (node == this->node()) {
    if (!data_writer_) {
      OpenDataWriter();
    }
    UpdateHeader(header, data_writer_->uuid(), data_writer_->part_number);
    data_writer_->writer->QueueSpan(header->span());
  } else {
    for (std::pair<const Channel *const, NewDataWriter> &data_writer :
         data_writers_) {
      if (node == data_writer.second.node) {
        UpdateHeader(header, data_writer.second.uuid(),
                     data_writer.second.part_number);
        data_writer.second.writer->QueueSpan(header->span());
      }
    }
  }
}

void MultiNodeLogNamer::Rotate(
    const Node *node,
    aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header) {
  DoRotate(node, header, false);
}

void MultiNodeLogNamer::Reboot(
    const Node *node,
    aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header) {
  DoRotate(node, header, true);
}

void MultiNodeLogNamer::DoRotate(
    const Node *node,
    aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header,
    bool reboot) {
  if (node == this->node()) {
    if (data_writer_) {
      if (reboot) {
        data_writer_->Reboot();
      } else {
        data_writer_->Rotate();
      }
      // TODO(austin): Move this logic down once we have a better ownership
      // model for the header.
      UpdateHeader(header, data_writer_->uuid(), data_writer_->part_number);
      data_writer_->writer->QueueSpan(header->span());
    }
  } else {
    for (std::pair<const Channel *const, NewDataWriter> &data_writer :
         data_writers_) {
      if (node == data_writer.second.node) {
        if (reboot) {
          data_writer.second.Reboot();
        } else {
          data_writer.second.Rotate();
        }
        UpdateHeader(header, data_writer.second.uuid(),
                     data_writer.second.part_number);
        data_writer.second.writer->QueueSpan(header->span());
      }
    }
  }
}

void MultiNodeLogNamer::WriteConfiguration(
    aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header,
    std::string_view config_sha256) {
  if (ran_out_of_space_) {
    return;
  }

  const std::string_view separator = base_name_.back() == '/' ? "" : "_";
  const std::string filename = absl::StrCat(
      base_name_, separator, config_sha256, ".bfbs", extension_, temp_suffix_);

  std::unique_ptr<DetachedBufferWriter> writer =
      std::make_unique<DetachedBufferWriter>(filename, encoder_factory_());

  writer->QueueSizedFlatbuffer(header->Release());

  if (!writer->ran_out_of_space()) {
    all_filenames_.emplace_back(filename);
  }
  CloseWriter(&writer);
}

NewDataWriter *MultiNodeLogNamer::MakeWriter(const Channel *channel) {
  // See if we can read the data on this node at all.
  const bool is_readable =
      configuration::ChannelIsReadableOnNode(channel, this->node());
  if (!is_readable) {
    return nullptr;
  }

  // Then, see if we are supposed to log the data here.
  const bool log_message =
      configuration::ChannelMessageIsLoggedOnNode(channel, this->node());

  if (!log_message) {
    return nullptr;
  }

  // Now, sort out if this is data generated on this node, or not.  It is
  // generated if it is sendable on this node.
  if (configuration::ChannelIsSendableOnNode(channel, this->node())) {
    if (!data_writer_) {
      OpenDataWriter();
    }
    return data_writer_.get();
  }

  // Ok, we have data that is being forwarded to us that we are supposed to
  // log.  It needs to be logged with send timestamps, but be sorted enough
  // to be able to be processed.
  CHECK(data_writers_.find(channel) == data_writers_.end());

  // Track that this node is being logged.
  const Node *source_node = configuration::GetNode(
      configuration_, channel->source_node()->string_view());

  if (std::find(nodes_.begin(), nodes_.end(), source_node) == nodes_.end()) {
    nodes_.emplace_back(source_node);
  }

  NewDataWriter data_writer(
      [this, channel](NewDataWriter *data_writer) {
        OpenWriter(channel, data_writer);
      },
      [this](NewDataWriter *data_writer) {
        CloseWriter(&data_writer->writer);
      });
  data_writer.node = source_node;

  return &(
      data_writers_.emplace(channel, std::move(data_writer)).first->second);
}

NewDataWriter *MultiNodeLogNamer::MakeForwardedTimestampWriter(
    const Channel *channel, const Node *node) {
  // See if we can read the data on this node at all.
  const bool is_readable =
      configuration::ChannelIsReadableOnNode(channel, this->node());
  CHECK(is_readable) << ": " << configuration::CleanedChannelToString(channel);

  CHECK(data_writers_.find(channel) == data_writers_.end());

  if (std::find(nodes_.begin(), nodes_.end(), node) == nodes_.end()) {
    nodes_.emplace_back(node);
  }

  NewDataWriter data_writer(
      [this, channel](NewDataWriter *data_writer) {
        OpenForwardedTimestampWriter(channel, data_writer);
      },
      [this](NewDataWriter *data_writer) {
        CloseWriter(&data_writer->writer);
      });
  data_writer.node = node;

  return &(
      data_writers_.emplace(channel, std::move(data_writer)).first->second);
}

NewDataWriter *MultiNodeLogNamer::MakeTimestampWriter(const Channel *channel) {
  bool log_delivery_times = false;
  if (this->node() != nullptr) {
    log_delivery_times = configuration::ConnectionDeliveryTimeIsLoggedOnNode(
        channel, this->node(), this->node());
  }
  if (!log_delivery_times) {
    return nullptr;
  }

  if (!data_writer_) {
    OpenDataWriter();
  }
  return data_writer_.get();
}

void MultiNodeLogNamer::Close() {
  data_writers_.clear();
  data_writer_.reset();
}

void MultiNodeLogNamer::ResetStatistics() {
  for (std::pair<const Channel *const, NewDataWriter> &data_writer :
       data_writers_) {
    if (!data_writer.second.writer) continue;
    data_writer.second.writer->ResetStatistics();
  }
  if (data_writer_) {
    data_writer_->writer->ResetStatistics();
  }
  max_write_time_ = std::chrono::nanoseconds::zero();
  max_write_time_bytes_ = -1;
  max_write_time_messages_ = -1;
  total_write_time_ = std::chrono::nanoseconds::zero();
  total_write_count_ = 0;
  total_write_messages_ = 0;
  total_write_bytes_ = 0;
}

void MultiNodeLogNamer::OpenForwardedTimestampWriter(
    const Channel *channel, NewDataWriter *data_writer) {
  std::string filename =
      absl::StrCat("timestamps", channel->name()->string_view(), "/",
                   channel->type()->string_view(), ".part",
                   data_writer->part_number, ".bfbs", extension_);
  CreateBufferWriter(filename, &data_writer->writer);
}

void MultiNodeLogNamer::OpenWriter(const Channel *channel,
                                   NewDataWriter *data_writer) {
  const std::string filename = absl::StrCat(
      CHECK_NOTNULL(channel->source_node())->string_view(), "_data",
      channel->name()->string_view(), "/", channel->type()->string_view(),
      ".part", data_writer->part_number, ".bfbs", extension_);
  CreateBufferWriter(filename, &data_writer->writer);
}

void MultiNodeLogNamer::OpenDataWriter() {
  if (!data_writer_) {
    data_writer_ = std::make_unique<NewDataWriter>(
        [this](NewDataWriter *writer) {
          std::string name;
          if (node() != nullptr) {
            name = absl::StrCat(name, node()->name()->string_view(), "_");
          }
          absl::StrAppend(&name, "data.part", writer->part_number, ".bfbs",
                          extension_);
          CreateBufferWriter(name, &writer->writer);
        },
        [this](NewDataWriter *data_writer) {
          CloseWriter(&data_writer->writer);
        });
  }
}

void MultiNodeLogNamer::CreateBufferWriter(
    std::string_view path, std::unique_ptr<DetachedBufferWriter> *destination) {
  if (ran_out_of_space_) {
    // Refuse to open any new files, which might skip data. Any existing files
    // are in the same folder, which means they're on the same filesystem, which
    // means they're probably going to run out of space and get stuck too.
    if (!destination->get()) {
      // But avoid leaving a nullptr writer if we're out of space when
      // attempting to open the first file.
      *destination = std::make_unique<DetachedBufferWriter>(
          DetachedBufferWriter::already_out_of_space_t());
    }
    return;
  }
  const std::string_view separator = base_name_.back() == '/' ? "" : "_";
  const std::string filename =
      absl::StrCat(base_name_, separator, path, temp_suffix_);
  if (!destination->get()) {
    if (ran_out_of_space_) {
      *destination = std::make_unique<DetachedBufferWriter>(
          DetachedBufferWriter::already_out_of_space_t());
      return;
    }
    *destination =
        std::make_unique<DetachedBufferWriter>(filename, encoder_factory_());
    if (!destination->get()->ran_out_of_space()) {
      all_filenames_.emplace_back(path);
    }
    return;
  }

  CloseWriter(destination);
  if (ran_out_of_space_) {
    *destination->get() =
        DetachedBufferWriter(DetachedBufferWriter::already_out_of_space_t());
    return;
  }

  *destination->get() = DetachedBufferWriter(filename, encoder_factory_());
  if (!destination->get()->ran_out_of_space()) {
    all_filenames_.emplace_back(path);
  }
}

void MultiNodeLogNamer::RenameTempFile(DetachedBufferWriter *destination) {
  if (temp_suffix_.empty()) {
    return;
  }
  std::string current_filename = std::string(destination->filename());
  CHECK(current_filename.size() > temp_suffix_.size());
  std::string final_filename =
      current_filename.substr(0, current_filename.size() - temp_suffix_.size());
  int result = rename(current_filename.c_str(), final_filename.c_str());

  // When changing the base name, we rename the log folder while there active
  // buffer writers. Therefore, the name of that active buffer may still refer
  // to the old file location rather than the new one. This minimized changes to
  // existing code.
  if (result != 0 && errno != ENOSPC && !old_base_name_.empty()) {
    auto offset = current_filename.find(old_base_name_);
    if (offset != std::string::npos) {
      current_filename.replace(offset, old_base_name_.length(), base_name_);
    }
    offset = final_filename.find(old_base_name_);
    if (offset != std::string::npos) {
      final_filename.replace(offset, old_base_name_.length(), base_name_);
    }
    result = rename(current_filename.c_str(), final_filename.c_str());
  }

  if (result != 0) {
    if (errno == ENOSPC) {
      ran_out_of_space_ = true;
      return;
    } else {
      PLOG(FATAL) << "Renaming " << current_filename << " to " << final_filename
                  << " failed";
    }
  } else {
    VLOG(1) << "Renamed " << current_filename << " -> " << final_filename;
  }
}

void MultiNodeLogNamer::CloseWriter(
    std::unique_ptr<DetachedBufferWriter> *writer_pointer) {
  DetachedBufferWriter *const writer = writer_pointer->get();
  if (!writer) {
    return;
  }
  const bool was_open = writer->is_open();
  writer->Close();

  if (writer->max_write_time() > max_write_time_) {
    max_write_time_ = writer->max_write_time();
    max_write_time_bytes_ = writer->max_write_time_bytes();
    max_write_time_messages_ = writer->max_write_time_messages();
  }
  total_write_time_ += writer->total_write_time();
  total_write_count_ += writer->total_write_count();
  total_write_messages_ += writer->total_write_messages();
  total_write_bytes_ += writer->total_write_bytes();

  if (writer->ran_out_of_space()) {
    ran_out_of_space_ = true;
    writer->acknowledge_out_of_space();
  }
  if (was_open) {
    RenameTempFile(writer);
  } else {
    CHECK(access(std::string(writer->filename()).c_str(), F_OK) == -1)
        << ": File should not exist: " << writer->filename();
  }
}

}  // namespace logger
}  // namespace aos