Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 1 | #include "aos/events/logging/log_namer.h" |
| 2 | |
| 3 | #include <functional> |
| 4 | #include <map> |
| 5 | #include <memory> |
| 6 | #include <string_view> |
| 7 | #include <vector> |
| 8 | |
| 9 | #include "absl/strings/str_cat.h" |
| 10 | #include "aos/events/logging/logfile_utils.h" |
| 11 | #include "aos/events/logging/logger_generated.h" |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 12 | #include "aos/flatbuffer_merge.h" |
Austin Schuh | 4385b14 | 2021-03-14 21:31:13 -0700 | [diff] [blame] | 13 | #include "aos/uuid.h" |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 14 | #include "flatbuffers/flatbuffers.h" |
| 15 | #include "glog/logging.h" |
| 16 | |
| 17 | namespace aos { |
| 18 | namespace logger { |
| 19 | |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame^] | 20 | NewDataWriter::NewDataWriter(LogNamer *log_namer, const Node *node, |
| 21 | std::function<void(NewDataWriter *)> reopen, |
| 22 | std::function<void(NewDataWriter *)> close) |
| 23 | : node_(node), |
| 24 | node_index_(configuration::GetNodeIndex(log_namer->configuration_, node)), |
| 25 | log_namer_(log_namer), |
| 26 | reopen_(std::move(reopen)), |
| 27 | close_(std::move(close)) { |
| 28 | reopen_(this); |
| 29 | } |
| 30 | |
| 31 | NewDataWriter::~NewDataWriter() { |
| 32 | if (writer) { |
| 33 | Close(); |
| 34 | } |
| 35 | } |
| 36 | |
| 37 | void NewDataWriter::Rotate() { |
| 38 | ++parts_index_; |
| 39 | reopen_(this); |
| 40 | header_written_ = false; |
| 41 | QueueHeader(log_namer_->MakeHeader(node_index_, source_node_boot_uuid_, |
| 42 | parts_uuid(), parts_index_)); |
| 43 | } |
| 44 | |
| 45 | void NewDataWriter::Reboot() { |
| 46 | parts_uuid_ = UUID::Random(); |
| 47 | ++parts_index_; |
| 48 | reopen_(this); |
| 49 | header_written_ = false; |
| 50 | } |
| 51 | |
| 52 | void NewDataWriter::QueueSizedFlatbuffer(flatbuffers::FlatBufferBuilder *fbb, |
| 53 | const UUID &source_node_boot_uuid, |
| 54 | aos::monotonic_clock::time_point now) { |
| 55 | // TODO(austin): Handle remote nodes changing too, not just the source node. |
| 56 | if (source_node_boot_uuid_ != source_node_boot_uuid) { |
| 57 | if (header_written_) { |
| 58 | Reboot(); |
| 59 | } |
| 60 | |
| 61 | QueueHeader(log_namer_->MakeHeader(node_index_, source_node_boot_uuid, |
| 62 | parts_uuid(), parts_index_)); |
| 63 | } |
| 64 | CHECK_EQ(source_node_boot_uuid_, source_node_boot_uuid); |
| 65 | CHECK(header_written_) << ": Attempting to write message before header to " |
| 66 | << writer->filename(); |
| 67 | writer->QueueSizedFlatbuffer(fbb, now); |
| 68 | } |
| 69 | |
| 70 | void NewDataWriter::QueueHeader( |
| 71 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> &&header) { |
| 72 | CHECK(!header_written_) << ": Attempting to write duplicate header to " |
| 73 | << writer->filename(); |
| 74 | CHECK(header.message().has_source_node_boot_uuid()); |
| 75 | source_node_boot_uuid_ = |
| 76 | UUID::FromString(header.message().source_node_boot_uuid()); |
| 77 | // TODO(austin): This triggers a dummy allocation that we don't need as part |
| 78 | // of releasing. Can we skip it? |
| 79 | writer->QueueSizedFlatbuffer(header.Release()); |
| 80 | header_written_ = true; |
| 81 | } |
| 82 | |
| 83 | void NewDataWriter::Close() { |
| 84 | CHECK(writer); |
| 85 | close_(this); |
| 86 | writer.reset(); |
| 87 | header_written_ = false; |
| 88 | } |
| 89 | |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 90 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> LogNamer::MakeHeader( |
| 91 | size_t node_index, const UUID &source_node_boot_uuid, |
| 92 | const UUID &parts_uuid, int parts_index) const { |
| 93 | const Node *const source_node = |
| 94 | configuration::GetNode(configuration_, node_index); |
| 95 | CHECK_EQ(LogFileHeader::MiniReflectTypeTable()->num_elems, 18u); |
| 96 | flatbuffers::FlatBufferBuilder fbb; |
| 97 | fbb.ForceDefaults(true); |
| 98 | |
| 99 | flatbuffers::Offset<flatbuffers::String> config_sha256_offset; |
| 100 | flatbuffers::Offset<aos::Configuration> configuration_offset; |
| 101 | if (header_.message().has_configuration()) { |
| 102 | CHECK(!header_.message().has_configuration_sha256()); |
| 103 | configuration_offset = |
| 104 | CopyFlatBuffer(header_.message().configuration(), &fbb); |
| 105 | } else { |
| 106 | CHECK(!header_.message().has_configuration()); |
| 107 | CHECK(header_.message().has_configuration_sha256()); |
| 108 | config_sha256_offset = fbb.CreateString( |
| 109 | header_.message().configuration_sha256()->string_view()); |
| 110 | } |
| 111 | |
| 112 | CHECK(header_.message().has_name()); |
| 113 | const flatbuffers::Offset<flatbuffers::String> name_offset = |
| 114 | fbb.CreateString(header_.message().name()->string_view()); |
| 115 | |
| 116 | CHECK(header_.message().has_log_event_uuid()); |
| 117 | const flatbuffers::Offset<flatbuffers::String> log_event_uuid_offset = |
| 118 | fbb.CreateString(header_.message().log_event_uuid()->string_view()); |
| 119 | |
| 120 | CHECK(header_.message().has_logger_instance_uuid()); |
| 121 | const flatbuffers::Offset<flatbuffers::String> logger_instance_uuid_offset = |
| 122 | fbb.CreateString(header_.message().logger_instance_uuid()->string_view()); |
| 123 | |
| 124 | flatbuffers::Offset<flatbuffers::String> log_start_uuid_offset; |
| 125 | if (header_.message().has_log_start_uuid()) { |
| 126 | log_start_uuid_offset = |
| 127 | fbb.CreateString(header_.message().log_start_uuid()->string_view()); |
| 128 | } |
| 129 | |
| 130 | CHECK(header_.message().has_logger_node_boot_uuid()); |
| 131 | const flatbuffers::Offset<flatbuffers::String> logger_node_boot_uuid_offset = |
| 132 | fbb.CreateString( |
| 133 | header_.message().logger_node_boot_uuid()->string_view()); |
| 134 | |
| 135 | CHECK_NE(source_node_boot_uuid, UUID::Zero()); |
| 136 | const flatbuffers::Offset<flatbuffers::String> source_node_boot_uuid_offset = |
| 137 | source_node_boot_uuid.PackString(&fbb); |
| 138 | |
| 139 | const flatbuffers::Offset<flatbuffers::String> parts_uuid_offset = |
| 140 | parts_uuid.PackString(&fbb); |
| 141 | |
| 142 | flatbuffers::Offset<Node> node_offset; |
| 143 | flatbuffers::Offset<Node> logger_node_offset; |
| 144 | |
| 145 | if (configuration::MultiNode(configuration_)) { |
| 146 | node_offset = RecursiveCopyFlatBuffer(source_node, &fbb); |
| 147 | logger_node_offset = RecursiveCopyFlatBuffer(node_, &fbb); |
| 148 | } |
| 149 | |
| 150 | aos::logger::LogFileHeader::Builder log_file_header_builder(fbb); |
| 151 | |
| 152 | log_file_header_builder.add_name(name_offset); |
| 153 | |
| 154 | // Only add the node if we are running in a multinode configuration. |
| 155 | if (!logger_node_offset.IsNull()) { |
| 156 | log_file_header_builder.add_node(node_offset); |
| 157 | log_file_header_builder.add_logger_node(logger_node_offset); |
| 158 | } |
| 159 | |
| 160 | if (!configuration_offset.IsNull()) { |
| 161 | log_file_header_builder.add_configuration(configuration_offset); |
| 162 | } |
| 163 | log_file_header_builder.add_max_out_of_order_duration( |
| 164 | header_.message().max_out_of_order_duration()); |
| 165 | |
| 166 | log_file_header_builder.add_monotonic_start_time( |
| 167 | std::chrono::duration_cast<std::chrono::nanoseconds>( |
| 168 | node_states_[node_index].monotonic_start_time.time_since_epoch()) |
| 169 | .count()); |
| 170 | if (source_node == node_) { |
| 171 | log_file_header_builder.add_realtime_start_time( |
| 172 | std::chrono::duration_cast<std::chrono::nanoseconds>( |
| 173 | node_states_[node_index].realtime_start_time.time_since_epoch()) |
| 174 | .count()); |
| 175 | } else { |
| 176 | // Fill out the legacy start times. Since these were implemented to never |
| 177 | // change on reboot, they aren't very helpful in tracking what happened. |
| 178 | log_file_header_builder.add_logger_monotonic_start_time( |
| 179 | std::chrono::duration_cast<std::chrono::nanoseconds>( |
| 180 | node_states_[node_index] |
| 181 | .logger_monotonic_start_time.time_since_epoch()) |
| 182 | .count()); |
| 183 | log_file_header_builder.add_logger_realtime_start_time( |
| 184 | std::chrono::duration_cast<std::chrono::nanoseconds>( |
| 185 | node_states_[node_index] |
| 186 | .logger_realtime_start_time.time_since_epoch()) |
| 187 | .count()); |
| 188 | } |
| 189 | |
| 190 | // TODO(austin): Add more useful times. When was this part started? What do |
| 191 | // we know about both the logger and remote then? |
| 192 | |
| 193 | log_file_header_builder.add_log_event_uuid(log_event_uuid_offset); |
| 194 | log_file_header_builder.add_logger_instance_uuid(logger_instance_uuid_offset); |
| 195 | if (!log_start_uuid_offset.IsNull()) { |
| 196 | log_file_header_builder.add_log_start_uuid(log_start_uuid_offset); |
| 197 | } |
| 198 | log_file_header_builder.add_logger_node_boot_uuid( |
| 199 | logger_node_boot_uuid_offset); |
| 200 | log_file_header_builder.add_source_node_boot_uuid( |
| 201 | source_node_boot_uuid_offset); |
| 202 | |
| 203 | log_file_header_builder.add_parts_uuid(parts_uuid_offset); |
| 204 | log_file_header_builder.add_parts_index(parts_index); |
| 205 | |
| 206 | if (!config_sha256_offset.IsNull()) { |
| 207 | log_file_header_builder.add_configuration_sha256(config_sha256_offset); |
| 208 | } |
| 209 | |
| 210 | fbb.FinishSizePrefixed(log_file_header_builder.Finish()); |
| 211 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> result( |
| 212 | fbb.Release()); |
| 213 | |
| 214 | CHECK(result.Verify()) << ": Built a corrupted header."; |
| 215 | |
| 216 | return result; |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 217 | } |
| 218 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 219 | NewDataWriter *LocalLogNamer::MakeWriter(const Channel *channel) { |
Austin Schuh | df57647 | 2020-10-19 09:39:37 -0700 | [diff] [blame] | 220 | CHECK(configuration::ChannelIsSendableOnNode(channel, node())) |
| 221 | << ": " << configuration::CleanedChannelToString(channel); |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 222 | return &data_writer_; |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 223 | } |
| 224 | |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 225 | void LocalLogNamer::Rotate(const Node *node) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 226 | CHECK(node == this->node()); |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 227 | data_writer_.Rotate(); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 228 | } |
Austin Schuh | 8c39996 | 2020-12-25 21:51:45 -0800 | [diff] [blame] | 229 | |
| 230 | void LocalLogNamer::WriteConfiguration( |
| 231 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header, |
| 232 | std::string_view config_sha256) { |
| 233 | const std::string filename = absl::StrCat(base_name_, config_sha256, ".bfbs"); |
| 234 | |
| 235 | std::unique_ptr<DetachedBufferWriter> writer = |
| 236 | std::make_unique<DetachedBufferWriter>( |
| 237 | filename, std::make_unique<aos::logger::DummyEncoder>()); |
| 238 | writer->QueueSizedFlatbuffer(header->Release()); |
| 239 | } |
| 240 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 241 | NewDataWriter *LocalLogNamer::MakeTimestampWriter(const Channel *channel) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 242 | CHECK(configuration::ChannelIsReadableOnNode(channel, node_)) |
| 243 | << ": Message is not delivered to this node."; |
| 244 | CHECK(node_ != nullptr) << ": Can't log timestamps in a single node world"; |
| 245 | CHECK(configuration::ConnectionDeliveryTimeIsLoggedOnNode(channel, node_, |
| 246 | node_)) |
| 247 | << ": Delivery times aren't logged for this channel on this node."; |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 248 | return &data_writer_; |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 249 | } |
| 250 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 251 | NewDataWriter *LocalLogNamer::MakeForwardedTimestampWriter( |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 252 | const Channel * /*channel*/, const Node * /*node*/) { |
| 253 | LOG(FATAL) << "Can't log forwarded timestamps in a singe log file."; |
| 254 | return nullptr; |
| 255 | } |
| 256 | |
| 257 | MultiNodeLogNamer::MultiNodeLogNamer(std::string_view base_name, |
| 258 | const Configuration *configuration, |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 259 | const Node *node) |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 260 | : LogNamer(configuration, node), base_name_(base_name), old_base_name_() {} |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 261 | |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 262 | MultiNodeLogNamer::~MultiNodeLogNamer() { |
| 263 | if (!ran_out_of_space_) { |
| 264 | // This handles renaming temporary files etc. |
| 265 | Close(); |
| 266 | } |
| 267 | } |
| 268 | |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame^] | 269 | void MultiNodeLogNamer::Rotate(const Node *node) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 270 | if (node == this->node()) { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 271 | if (data_writer_) { |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame^] | 272 | data_writer_->Rotate(); |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 273 | } |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 274 | } else { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 275 | for (std::pair<const Channel *const, NewDataWriter> &data_writer : |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 276 | data_writers_) { |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame^] | 277 | if (node == data_writer.second.node()) { |
| 278 | data_writer.second.Rotate(); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 279 | } |
| 280 | } |
| 281 | } |
| 282 | } |
| 283 | |
Austin Schuh | 8c39996 | 2020-12-25 21:51:45 -0800 | [diff] [blame] | 284 | void MultiNodeLogNamer::WriteConfiguration( |
| 285 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header, |
| 286 | std::string_view config_sha256) { |
| 287 | if (ran_out_of_space_) { |
| 288 | return; |
| 289 | } |
| 290 | |
| 291 | const std::string_view separator = base_name_.back() == '/' ? "" : "_"; |
| 292 | const std::string filename = absl::StrCat( |
| 293 | base_name_, separator, config_sha256, ".bfbs", extension_, temp_suffix_); |
| 294 | |
| 295 | std::unique_ptr<DetachedBufferWriter> writer = |
| 296 | std::make_unique<DetachedBufferWriter>(filename, encoder_factory_()); |
| 297 | |
| 298 | writer->QueueSizedFlatbuffer(header->Release()); |
| 299 | |
| 300 | if (!writer->ran_out_of_space()) { |
| 301 | all_filenames_.emplace_back(filename); |
| 302 | } |
| 303 | CloseWriter(&writer); |
| 304 | } |
| 305 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 306 | NewDataWriter *MultiNodeLogNamer::MakeWriter(const Channel *channel) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 307 | // See if we can read the data on this node at all. |
| 308 | const bool is_readable = |
| 309 | configuration::ChannelIsReadableOnNode(channel, this->node()); |
| 310 | if (!is_readable) { |
| 311 | return nullptr; |
| 312 | } |
| 313 | |
| 314 | // Then, see if we are supposed to log the data here. |
| 315 | const bool log_message = |
| 316 | configuration::ChannelMessageIsLoggedOnNode(channel, this->node()); |
| 317 | |
| 318 | if (!log_message) { |
| 319 | return nullptr; |
| 320 | } |
| 321 | |
| 322 | // Now, sort out if this is data generated on this node, or not. It is |
| 323 | // generated if it is sendable on this node. |
| 324 | if (configuration::ChannelIsSendableOnNode(channel, this->node())) { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 325 | if (!data_writer_) { |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 326 | OpenDataWriter(); |
| 327 | } |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 328 | return data_writer_.get(); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 329 | } |
| 330 | |
| 331 | // Ok, we have data that is being forwarded to us that we are supposed to |
| 332 | // log. It needs to be logged with send timestamps, but be sorted enough |
| 333 | // to be able to be processed. |
| 334 | CHECK(data_writers_.find(channel) == data_writers_.end()); |
| 335 | |
| 336 | // Track that this node is being logged. |
| 337 | const Node *source_node = configuration::GetNode( |
| 338 | configuration_, channel->source_node()->string_view()); |
| 339 | |
| 340 | if (std::find(nodes_.begin(), nodes_.end(), source_node) == nodes_.end()) { |
| 341 | nodes_.emplace_back(source_node); |
| 342 | } |
| 343 | |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame^] | 344 | NewDataWriter data_writer(this, source_node, |
| 345 | [this, channel](NewDataWriter *data_writer) { |
| 346 | OpenWriter(channel, data_writer); |
| 347 | }, |
| 348 | [this](NewDataWriter *data_writer) { |
| 349 | CloseWriter(&data_writer->writer); |
| 350 | }); |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 351 | return &( |
| 352 | data_writers_.emplace(channel, std::move(data_writer)).first->second); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 353 | } |
| 354 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 355 | NewDataWriter *MultiNodeLogNamer::MakeForwardedTimestampWriter( |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 356 | const Channel *channel, const Node *node) { |
| 357 | // See if we can read the data on this node at all. |
| 358 | const bool is_readable = |
| 359 | configuration::ChannelIsReadableOnNode(channel, this->node()); |
| 360 | CHECK(is_readable) << ": " << configuration::CleanedChannelToString(channel); |
| 361 | |
| 362 | CHECK(data_writers_.find(channel) == data_writers_.end()); |
| 363 | |
| 364 | if (std::find(nodes_.begin(), nodes_.end(), node) == nodes_.end()) { |
| 365 | nodes_.emplace_back(node); |
| 366 | } |
| 367 | |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame^] | 368 | NewDataWriter data_writer(this, node, |
| 369 | [this, channel](NewDataWriter *data_writer) { |
| 370 | OpenForwardedTimestampWriter(channel, |
| 371 | data_writer); |
| 372 | }, |
| 373 | [this](NewDataWriter *data_writer) { |
| 374 | CloseWriter(&data_writer->writer); |
| 375 | }); |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 376 | return &( |
| 377 | data_writers_.emplace(channel, std::move(data_writer)).first->second); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 378 | } |
| 379 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 380 | NewDataWriter *MultiNodeLogNamer::MakeTimestampWriter(const Channel *channel) { |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 381 | bool log_delivery_times = false; |
| 382 | if (this->node() != nullptr) { |
| 383 | log_delivery_times = configuration::ConnectionDeliveryTimeIsLoggedOnNode( |
| 384 | channel, this->node(), this->node()); |
| 385 | } |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 386 | if (!log_delivery_times) { |
| 387 | return nullptr; |
| 388 | } |
| 389 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 390 | if (!data_writer_) { |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 391 | OpenDataWriter(); |
| 392 | } |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 393 | return data_writer_.get(); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 394 | } |
| 395 | |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 396 | void MultiNodeLogNamer::Close() { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 397 | data_writers_.clear(); |
| 398 | data_writer_.reset(); |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 399 | } |
| 400 | |
| 401 | void MultiNodeLogNamer::ResetStatistics() { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 402 | for (std::pair<const Channel *const, NewDataWriter> &data_writer : |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 403 | data_writers_) { |
Austin Schuh | ad0cfc3 | 2020-12-21 12:34:26 -0800 | [diff] [blame] | 404 | if (!data_writer.second.writer) continue; |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 405 | data_writer.second.writer->ResetStatistics(); |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 406 | } |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 407 | if (data_writer_) { |
| 408 | data_writer_->writer->ResetStatistics(); |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 409 | } |
| 410 | max_write_time_ = std::chrono::nanoseconds::zero(); |
| 411 | max_write_time_bytes_ = -1; |
| 412 | max_write_time_messages_ = -1; |
| 413 | total_write_time_ = std::chrono::nanoseconds::zero(); |
| 414 | total_write_count_ = 0; |
| 415 | total_write_messages_ = 0; |
| 416 | total_write_bytes_ = 0; |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 417 | } |
| 418 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 419 | void MultiNodeLogNamer::OpenForwardedTimestampWriter( |
| 420 | const Channel *channel, NewDataWriter *data_writer) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 421 | std::string filename = |
Austin Schuh | e715eae | 2020-10-10 15:39:30 -0700 | [diff] [blame] | 422 | absl::StrCat("timestamps", channel->name()->string_view(), "/", |
Brian Silverman | a621f52 | 2020-09-30 16:52:43 -0700 | [diff] [blame] | 423 | channel->type()->string_view(), ".part", |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame^] | 424 | data_writer->parts_index(), ".bfbs", extension_); |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 425 | CreateBufferWriter(filename, &data_writer->writer); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 426 | } |
| 427 | |
| 428 | void MultiNodeLogNamer::OpenWriter(const Channel *channel, |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 429 | NewDataWriter *data_writer) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 430 | const std::string filename = absl::StrCat( |
Austin Schuh | e715eae | 2020-10-10 15:39:30 -0700 | [diff] [blame] | 431 | CHECK_NOTNULL(channel->source_node())->string_view(), "_data", |
Brian Silverman | a621f52 | 2020-09-30 16:52:43 -0700 | [diff] [blame] | 432 | channel->name()->string_view(), "/", channel->type()->string_view(), |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame^] | 433 | ".part", data_writer->parts_index(), ".bfbs", extension_); |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 434 | CreateBufferWriter(filename, &data_writer->writer); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 435 | } |
| 436 | |
Brian Silverman | a621f52 | 2020-09-30 16:52:43 -0700 | [diff] [blame] | 437 | void MultiNodeLogNamer::OpenDataWriter() { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 438 | if (!data_writer_) { |
| 439 | data_writer_ = std::make_unique<NewDataWriter>( |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame^] | 440 | this, node_, |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 441 | [this](NewDataWriter *writer) { |
| 442 | std::string name; |
| 443 | if (node() != nullptr) { |
| 444 | name = absl::StrCat(name, node()->name()->string_view(), "_"); |
| 445 | } |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame^] | 446 | absl::StrAppend(&name, "data.part", writer->parts_index(), ".bfbs", |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 447 | extension_); |
| 448 | CreateBufferWriter(name, &writer->writer); |
| 449 | }, |
| 450 | [this](NewDataWriter *data_writer) { |
| 451 | CloseWriter(&data_writer->writer); |
| 452 | }); |
Brian Silverman | 7af8c90 | 2020-09-29 16:14:04 -0700 | [diff] [blame] | 453 | } |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 454 | } |
| 455 | |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 456 | void MultiNodeLogNamer::CreateBufferWriter( |
Brian Silverman | a621f52 | 2020-09-30 16:52:43 -0700 | [diff] [blame] | 457 | std::string_view path, std::unique_ptr<DetachedBufferWriter> *destination) { |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 458 | if (ran_out_of_space_) { |
| 459 | // Refuse to open any new files, which might skip data. Any existing files |
| 460 | // are in the same folder, which means they're on the same filesystem, which |
| 461 | // means they're probably going to run out of space and get stuck too. |
Austin Schuh | a426f1f | 2021-03-31 22:27:41 -0700 | [diff] [blame] | 462 | if (!destination->get()) { |
| 463 | // But avoid leaving a nullptr writer if we're out of space when |
| 464 | // attempting to open the first file. |
| 465 | *destination = std::make_unique<DetachedBufferWriter>( |
| 466 | DetachedBufferWriter::already_out_of_space_t()); |
| 467 | } |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 468 | return; |
| 469 | } |
Austin Schuh | e715eae | 2020-10-10 15:39:30 -0700 | [diff] [blame] | 470 | const std::string_view separator = base_name_.back() == '/' ? "" : "_"; |
| 471 | const std::string filename = |
| 472 | absl::StrCat(base_name_, separator, path, temp_suffix_); |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 473 | if (!destination->get()) { |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 474 | if (ran_out_of_space_) { |
| 475 | *destination = std::make_unique<DetachedBufferWriter>( |
| 476 | DetachedBufferWriter::already_out_of_space_t()); |
| 477 | return; |
| 478 | } |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 479 | *destination = |
| 480 | std::make_unique<DetachedBufferWriter>(filename, encoder_factory_()); |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 481 | if (!destination->get()->ran_out_of_space()) { |
| 482 | all_filenames_.emplace_back(path); |
| 483 | } |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 484 | return; |
| 485 | } |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 486 | |
| 487 | CloseWriter(destination); |
| 488 | if (ran_out_of_space_) { |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 489 | *destination->get() = |
| 490 | DetachedBufferWriter(DetachedBufferWriter::already_out_of_space_t()); |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 491 | return; |
| 492 | } |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 493 | |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 494 | *destination->get() = DetachedBufferWriter(filename, encoder_factory_()); |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 495 | if (!destination->get()->ran_out_of_space()) { |
| 496 | all_filenames_.emplace_back(path); |
| 497 | } |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 498 | } |
| 499 | |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 500 | void MultiNodeLogNamer::RenameTempFile(DetachedBufferWriter *destination) { |
| 501 | if (temp_suffix_.empty()) { |
| 502 | return; |
| 503 | } |
Austin Schuh | 6bb8a82 | 2021-03-31 23:04:39 -0700 | [diff] [blame] | 504 | std::string current_filename = std::string(destination->filename()); |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 505 | CHECK(current_filename.size() > temp_suffix_.size()); |
Austin Schuh | 6bb8a82 | 2021-03-31 23:04:39 -0700 | [diff] [blame] | 506 | std::string final_filename = |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 507 | current_filename.substr(0, current_filename.size() - temp_suffix_.size()); |
Austin Schuh | 6bb8a82 | 2021-03-31 23:04:39 -0700 | [diff] [blame] | 508 | int result = rename(current_filename.c_str(), final_filename.c_str()); |
| 509 | |
| 510 | // When changing the base name, we rename the log folder while there active |
| 511 | // buffer writers. Therefore, the name of that active buffer may still refer |
| 512 | // to the old file location rather than the new one. This minimized changes to |
| 513 | // existing code. |
| 514 | if (result != 0 && errno != ENOSPC && !old_base_name_.empty()) { |
| 515 | auto offset = current_filename.find(old_base_name_); |
| 516 | if (offset != std::string::npos) { |
| 517 | current_filename.replace(offset, old_base_name_.length(), base_name_); |
| 518 | } |
| 519 | offset = final_filename.find(old_base_name_); |
| 520 | if (offset != std::string::npos) { |
| 521 | final_filename.replace(offset, old_base_name_.length(), base_name_); |
| 522 | } |
| 523 | result = rename(current_filename.c_str(), final_filename.c_str()); |
| 524 | } |
| 525 | |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 526 | if (result != 0) { |
| 527 | if (errno == ENOSPC) { |
| 528 | ran_out_of_space_ = true; |
| 529 | return; |
| 530 | } else { |
| 531 | PLOG(FATAL) << "Renaming " << current_filename << " to " << final_filename |
| 532 | << " failed"; |
| 533 | } |
Austin Schuh | 6bb8a82 | 2021-03-31 23:04:39 -0700 | [diff] [blame] | 534 | } else { |
| 535 | VLOG(1) << "Renamed " << current_filename << " -> " << final_filename; |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 536 | } |
| 537 | } |
| 538 | |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 539 | void MultiNodeLogNamer::CloseWriter( |
| 540 | std::unique_ptr<DetachedBufferWriter> *writer_pointer) { |
| 541 | DetachedBufferWriter *const writer = writer_pointer->get(); |
| 542 | if (!writer) { |
| 543 | return; |
| 544 | } |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 545 | const bool was_open = writer->is_open(); |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 546 | writer->Close(); |
| 547 | |
| 548 | if (writer->max_write_time() > max_write_time_) { |
| 549 | max_write_time_ = writer->max_write_time(); |
| 550 | max_write_time_bytes_ = writer->max_write_time_bytes(); |
| 551 | max_write_time_messages_ = writer->max_write_time_messages(); |
| 552 | } |
| 553 | total_write_time_ += writer->total_write_time(); |
| 554 | total_write_count_ += writer->total_write_count(); |
| 555 | total_write_messages_ += writer->total_write_messages(); |
| 556 | total_write_bytes_ += writer->total_write_bytes(); |
| 557 | |
| 558 | if (writer->ran_out_of_space()) { |
| 559 | ran_out_of_space_ = true; |
| 560 | writer->acknowledge_out_of_space(); |
| 561 | } |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 562 | if (was_open) { |
| 563 | RenameTempFile(writer); |
| 564 | } else { |
| 565 | CHECK(access(std::string(writer->filename()).c_str(), F_OK) == -1) |
| 566 | << ": File should not exist: " << writer->filename(); |
| 567 | } |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 568 | } |
| 569 | |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 570 | } // namespace logger |
| 571 | } // namespace aos |