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)) { |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 28 | state_.resize(configuration::NodesCount(log_namer->configuration_)); |
| 29 | CHECK_LT(node_index_, state_.size()); |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 30 | } |
| 31 | |
| 32 | NewDataWriter::~NewDataWriter() { |
| 33 | if (writer) { |
| 34 | Close(); |
| 35 | } |
| 36 | } |
| 37 | |
| 38 | void NewDataWriter::Rotate() { |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 39 | // No need to rotate if nothing has been written. |
| 40 | if (header_written_) { |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 41 | VLOG(1) << "Rotated " << filename(); |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 42 | ++parts_index_; |
| 43 | reopen_(this); |
| 44 | header_written_ = false; |
| 45 | QueueHeader(MakeHeader()); |
| 46 | } |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 47 | } |
| 48 | |
| 49 | void NewDataWriter::Reboot() { |
| 50 | parts_uuid_ = UUID::Random(); |
| 51 | ++parts_index_; |
| 52 | reopen_(this); |
| 53 | header_written_ = false; |
| 54 | } |
| 55 | |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 56 | void NewDataWriter::UpdateRemote( |
| 57 | const size_t remote_node_index, const UUID &remote_node_boot_uuid, |
| 58 | const monotonic_clock::time_point monotonic_remote_time, |
| 59 | const monotonic_clock::time_point monotonic_event_time, |
| 60 | const bool reliable) { |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 61 | // Trigger rotation if anything in the header changes. |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 62 | bool rotate = false; |
| 63 | CHECK_LT(remote_node_index, state_.size()); |
| 64 | State &state = state_[remote_node_index]; |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 65 | |
| 66 | // Did the remote boot UUID change? |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 67 | if (state.boot_uuid != remote_node_boot_uuid) { |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 68 | VLOG(1) << filename() << " Remote " << remote_node_index << " updated to " |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 69 | << remote_node_boot_uuid << " from " << state.boot_uuid; |
| 70 | state.boot_uuid = remote_node_boot_uuid; |
| 71 | state.oldest_remote_monotonic_timestamp = monotonic_clock::max_time; |
| 72 | state.oldest_local_monotonic_timestamp = monotonic_clock::max_time; |
| 73 | state.oldest_remote_unreliable_monotonic_timestamp = |
| 74 | monotonic_clock::max_time; |
| 75 | state.oldest_local_unreliable_monotonic_timestamp = |
| 76 | monotonic_clock::max_time; |
| 77 | rotate = true; |
| 78 | } |
| 79 | |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 80 | |
| 81 | // Did the unreliable timestamps change? |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 82 | if (!reliable) { |
| 83 | if (state.oldest_remote_unreliable_monotonic_timestamp > |
| 84 | monotonic_remote_time) { |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 85 | VLOG(1) << filename() << " Remote " << remote_node_index |
| 86 | << " oldest_remote_unreliable_monotonic_timestamp updated from " |
| 87 | << state.oldest_remote_unreliable_monotonic_timestamp << " to " |
| 88 | << monotonic_remote_time; |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 89 | state.oldest_remote_unreliable_monotonic_timestamp = |
| 90 | monotonic_remote_time; |
| 91 | state.oldest_local_unreliable_monotonic_timestamp = monotonic_event_time; |
| 92 | rotate = true; |
| 93 | } |
| 94 | } |
| 95 | |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 96 | // Did any of the timestamps change? |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 97 | if (state.oldest_remote_monotonic_timestamp > monotonic_remote_time) { |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 98 | VLOG(1) << filename() << " Remote " << remote_node_index |
| 99 | << " oldest_remote_monotonic_timestamp updated from " |
| 100 | << state.oldest_remote_monotonic_timestamp << " to " |
| 101 | << monotonic_remote_time; |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 102 | state.oldest_remote_monotonic_timestamp = monotonic_remote_time; |
| 103 | state.oldest_local_monotonic_timestamp = monotonic_event_time; |
| 104 | rotate = true; |
| 105 | } |
| 106 | |
| 107 | if (rotate) { |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 108 | Rotate(); |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | void NewDataWriter::QueueMessage(flatbuffers::FlatBufferBuilder *fbb, |
| 113 | const UUID &source_node_boot_uuid, |
| 114 | aos::monotonic_clock::time_point now) { |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 115 | // Trigger a reboot if we detect the boot UUID change. |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 116 | if (state_[node_index_].boot_uuid != source_node_boot_uuid) { |
| 117 | state_[node_index_].boot_uuid = source_node_boot_uuid; |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 118 | if (header_written_) { |
| 119 | Reboot(); |
| 120 | } |
| 121 | |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 122 | QueueHeader(MakeHeader()); |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 123 | } |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 124 | |
| 125 | // If the start time has changed for this node, trigger a rotation. |
| 126 | if (log_namer_->monotonic_start_time(node_index_, source_node_boot_uuid) != |
| 127 | monotonic_start_time_) { |
| 128 | CHECK(header_written_); |
| 129 | Rotate(); |
| 130 | } |
| 131 | |
| 132 | CHECK_EQ(log_namer_->monotonic_start_time(node_index_, source_node_boot_uuid), |
| 133 | monotonic_start_time_); |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 134 | CHECK_EQ(state_[node_index_].boot_uuid, source_node_boot_uuid); |
milind-u | a50344f | 2021-08-25 18:22:20 -0700 | [diff] [blame] | 135 | CHECK(writer); |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 136 | CHECK(header_written_) << ": Attempting to write message before header to " |
| 137 | << writer->filename(); |
| 138 | writer->QueueSizedFlatbuffer(fbb, now); |
| 139 | } |
| 140 | |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 141 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> |
| 142 | NewDataWriter::MakeHeader() { |
| 143 | const size_t logger_node_index = log_namer_->logger_node_index(); |
| 144 | const UUID &logger_node_boot_uuid = log_namer_->logger_node_boot_uuid(); |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 145 | if (state_[logger_node_index].boot_uuid == UUID::Zero()) { |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 146 | VLOG(1) << filename() << " Logger node is " << logger_node_index |
| 147 | << " and uuid is " << logger_node_boot_uuid; |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 148 | state_[logger_node_index].boot_uuid = logger_node_boot_uuid; |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 149 | } else { |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 150 | CHECK_EQ(state_[logger_node_index].boot_uuid, logger_node_boot_uuid); |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 151 | } |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 152 | return log_namer_->MakeHeader(node_index_, state_, parts_uuid(), |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 153 | parts_index_); |
| 154 | } |
| 155 | |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 156 | void NewDataWriter::QueueHeader( |
| 157 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> &&header) { |
| 158 | CHECK(!header_written_) << ": Attempting to write duplicate header to " |
| 159 | << writer->filename(); |
| 160 | CHECK(header.message().has_source_node_boot_uuid()); |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 161 | CHECK_EQ(state_[node_index_].boot_uuid, |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 162 | UUID::FromString(header.message().source_node_boot_uuid())); |
Austin Schuh | 510dc62 | 2021-08-06 18:47:30 -0700 | [diff] [blame] | 163 | if (!writer) { |
| 164 | reopen_(this); |
| 165 | } |
| 166 | |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 167 | VLOG(1) << "Writing to " << filename() << " " |
| 168 | << aos::FlatbufferToJson( |
| 169 | header, {.multi_line = false, .max_vector_size = 100}); |
| 170 | |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 171 | // TODO(austin): This triggers a dummy allocation that we don't need as part |
| 172 | // of releasing. Can we skip it? |
Austin Schuh | 510dc62 | 2021-08-06 18:47:30 -0700 | [diff] [blame] | 173 | CHECK(writer); |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 174 | writer->QueueSizedFlatbuffer(header.Release()); |
| 175 | header_written_ = true; |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 176 | monotonic_start_time_ = log_namer_->monotonic_start_time( |
| 177 | node_index_, state_[node_index_].boot_uuid); |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 178 | } |
| 179 | |
| 180 | void NewDataWriter::Close() { |
| 181 | CHECK(writer); |
| 182 | close_(this); |
| 183 | writer.reset(); |
| 184 | header_written_ = false; |
| 185 | } |
| 186 | |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 187 | LogNamer::NodeState *LogNamer::GetNodeState(size_t node_index, |
| 188 | const UUID &boot_uuid) { |
| 189 | auto it = node_states_.find(std::make_pair(node_index, boot_uuid)); |
| 190 | if (it == node_states_.end()) { |
| 191 | it = |
| 192 | node_states_.emplace(std::make_pair(node_index, boot_uuid), NodeState()) |
| 193 | .first; |
| 194 | } |
| 195 | return &it->second; |
| 196 | } |
| 197 | |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 198 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> LogNamer::MakeHeader( |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 199 | size_t node_index, const std::vector<NewDataWriter::State> &state, |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 200 | const UUID &parts_uuid, int parts_index) { |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 201 | const UUID &source_node_boot_uuid = state[node_index].boot_uuid; |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 202 | const Node *const source_node = |
| 203 | configuration::GetNode(configuration_, node_index); |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 204 | CHECK_EQ(LogFileHeader::MiniReflectTypeTable()->num_elems, 24u); |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 205 | flatbuffers::FlatBufferBuilder fbb; |
| 206 | fbb.ForceDefaults(true); |
| 207 | |
| 208 | flatbuffers::Offset<flatbuffers::String> config_sha256_offset; |
| 209 | flatbuffers::Offset<aos::Configuration> configuration_offset; |
| 210 | if (header_.message().has_configuration()) { |
| 211 | CHECK(!header_.message().has_configuration_sha256()); |
| 212 | configuration_offset = |
| 213 | CopyFlatBuffer(header_.message().configuration(), &fbb); |
| 214 | } else { |
| 215 | CHECK(!header_.message().has_configuration()); |
| 216 | CHECK(header_.message().has_configuration_sha256()); |
| 217 | config_sha256_offset = fbb.CreateString( |
| 218 | header_.message().configuration_sha256()->string_view()); |
| 219 | } |
| 220 | |
| 221 | CHECK(header_.message().has_name()); |
| 222 | const flatbuffers::Offset<flatbuffers::String> name_offset = |
| 223 | fbb.CreateString(header_.message().name()->string_view()); |
| 224 | |
| 225 | CHECK(header_.message().has_log_event_uuid()); |
| 226 | const flatbuffers::Offset<flatbuffers::String> log_event_uuid_offset = |
| 227 | fbb.CreateString(header_.message().log_event_uuid()->string_view()); |
| 228 | |
| 229 | CHECK(header_.message().has_logger_instance_uuid()); |
| 230 | const flatbuffers::Offset<flatbuffers::String> logger_instance_uuid_offset = |
| 231 | fbb.CreateString(header_.message().logger_instance_uuid()->string_view()); |
| 232 | |
| 233 | flatbuffers::Offset<flatbuffers::String> log_start_uuid_offset; |
| 234 | if (header_.message().has_log_start_uuid()) { |
| 235 | log_start_uuid_offset = |
| 236 | fbb.CreateString(header_.message().log_start_uuid()->string_view()); |
| 237 | } |
| 238 | |
| 239 | CHECK(header_.message().has_logger_node_boot_uuid()); |
| 240 | const flatbuffers::Offset<flatbuffers::String> logger_node_boot_uuid_offset = |
| 241 | fbb.CreateString( |
| 242 | header_.message().logger_node_boot_uuid()->string_view()); |
| 243 | |
| 244 | CHECK_NE(source_node_boot_uuid, UUID::Zero()); |
| 245 | const flatbuffers::Offset<flatbuffers::String> source_node_boot_uuid_offset = |
| 246 | source_node_boot_uuid.PackString(&fbb); |
| 247 | |
| 248 | const flatbuffers::Offset<flatbuffers::String> parts_uuid_offset = |
| 249 | parts_uuid.PackString(&fbb); |
| 250 | |
| 251 | flatbuffers::Offset<Node> node_offset; |
| 252 | flatbuffers::Offset<Node> logger_node_offset; |
| 253 | |
| 254 | if (configuration::MultiNode(configuration_)) { |
| 255 | node_offset = RecursiveCopyFlatBuffer(source_node, &fbb); |
| 256 | logger_node_offset = RecursiveCopyFlatBuffer(node_, &fbb); |
| 257 | } |
| 258 | |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 259 | std::vector<flatbuffers::Offset<flatbuffers::String>> boot_uuid_offsets; |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 260 | boot_uuid_offsets.reserve(state.size()); |
| 261 | for (const NewDataWriter::State &state : state) { |
| 262 | if (state.boot_uuid != UUID::Zero()) { |
| 263 | boot_uuid_offsets.emplace_back(state.boot_uuid.PackString(&fbb)); |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 264 | } else { |
| 265 | boot_uuid_offsets.emplace_back(fbb.CreateString("")); |
| 266 | } |
| 267 | } |
| 268 | |
| 269 | flatbuffers::Offset< |
| 270 | flatbuffers::Vector<flatbuffers::Offset<flatbuffers::String>>> |
| 271 | boot_uuids_offset = fbb.CreateVector(boot_uuid_offsets); |
| 272 | |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 273 | int64_t *oldest_remote_monotonic_timestamps; |
| 274 | flatbuffers::Offset<flatbuffers::Vector<int64_t>> |
| 275 | oldest_remote_monotonic_timestamps_offset = fbb.CreateUninitializedVector( |
| 276 | state.size(), &oldest_remote_monotonic_timestamps); |
| 277 | |
| 278 | int64_t *oldest_local_monotonic_timestamps; |
| 279 | flatbuffers::Offset<flatbuffers::Vector<int64_t>> |
| 280 | oldest_local_monotonic_timestamps_offset = fbb.CreateUninitializedVector( |
| 281 | state.size(), &oldest_local_monotonic_timestamps); |
| 282 | |
| 283 | int64_t *oldest_remote_unreliable_monotonic_timestamps; |
| 284 | flatbuffers::Offset<flatbuffers::Vector<int64_t>> |
| 285 | oldest_remote_unreliable_monotonic_timestamps_offset = |
| 286 | fbb.CreateUninitializedVector( |
| 287 | state.size(), &oldest_remote_unreliable_monotonic_timestamps); |
| 288 | |
| 289 | int64_t *oldest_local_unreliable_monotonic_timestamps; |
| 290 | flatbuffers::Offset<flatbuffers::Vector<int64_t>> |
| 291 | oldest_local_unreliable_monotonic_timestamps_offset = |
| 292 | fbb.CreateUninitializedVector( |
| 293 | state.size(), &oldest_local_unreliable_monotonic_timestamps); |
| 294 | |
| 295 | for (size_t i = 0; i < state.size(); ++i) { |
Austin Schuh | 5ae8f4a | 2021-09-11 19:09:50 -0700 | [diff] [blame^] | 296 | if (state[i].boot_uuid == UUID::Zero()) { |
| 297 | CHECK_EQ(state[i].oldest_remote_monotonic_timestamp, |
| 298 | monotonic_clock::max_time); |
| 299 | CHECK_EQ(state[i].oldest_local_monotonic_timestamp, |
| 300 | monotonic_clock::max_time); |
| 301 | CHECK_EQ(state[i].oldest_remote_unreliable_monotonic_timestamp, |
| 302 | monotonic_clock::max_time); |
| 303 | CHECK_EQ(state[i].oldest_local_unreliable_monotonic_timestamp, |
| 304 | monotonic_clock::max_time); |
| 305 | } |
| 306 | |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 307 | oldest_remote_monotonic_timestamps[i] = |
| 308 | state[i].oldest_remote_monotonic_timestamp.time_since_epoch().count(); |
| 309 | oldest_local_monotonic_timestamps[i] = |
| 310 | state[i].oldest_local_monotonic_timestamp.time_since_epoch().count(); |
| 311 | oldest_remote_unreliable_monotonic_timestamps[i] = |
| 312 | state[i] |
| 313 | .oldest_remote_unreliable_monotonic_timestamp.time_since_epoch() |
| 314 | .count(); |
| 315 | oldest_local_unreliable_monotonic_timestamps[i] = |
| 316 | state[i] |
| 317 | .oldest_local_unreliable_monotonic_timestamp.time_since_epoch() |
| 318 | .count(); |
| 319 | } |
| 320 | |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 321 | aos::logger::LogFileHeader::Builder log_file_header_builder(fbb); |
| 322 | |
| 323 | log_file_header_builder.add_name(name_offset); |
| 324 | |
| 325 | // Only add the node if we are running in a multinode configuration. |
| 326 | if (!logger_node_offset.IsNull()) { |
| 327 | log_file_header_builder.add_node(node_offset); |
| 328 | log_file_header_builder.add_logger_node(logger_node_offset); |
| 329 | } |
| 330 | |
| 331 | if (!configuration_offset.IsNull()) { |
| 332 | log_file_header_builder.add_configuration(configuration_offset); |
| 333 | } |
| 334 | log_file_header_builder.add_max_out_of_order_duration( |
| 335 | header_.message().max_out_of_order_duration()); |
| 336 | |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 337 | NodeState *node_state = GetNodeState(node_index, source_node_boot_uuid); |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 338 | log_file_header_builder.add_monotonic_start_time( |
| 339 | std::chrono::duration_cast<std::chrono::nanoseconds>( |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 340 | node_state->monotonic_start_time.time_since_epoch()) |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 341 | .count()); |
| 342 | if (source_node == node_) { |
| 343 | log_file_header_builder.add_realtime_start_time( |
| 344 | std::chrono::duration_cast<std::chrono::nanoseconds>( |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 345 | node_state->realtime_start_time.time_since_epoch()) |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 346 | .count()); |
| 347 | } else { |
| 348 | // Fill out the legacy start times. Since these were implemented to never |
| 349 | // change on reboot, they aren't very helpful in tracking what happened. |
| 350 | log_file_header_builder.add_logger_monotonic_start_time( |
| 351 | std::chrono::duration_cast<std::chrono::nanoseconds>( |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 352 | node_state->logger_monotonic_start_time.time_since_epoch()) |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 353 | .count()); |
| 354 | log_file_header_builder.add_logger_realtime_start_time( |
| 355 | std::chrono::duration_cast<std::chrono::nanoseconds>( |
Austin Schuh | 58646e2 | 2021-08-23 23:51:46 -0700 | [diff] [blame] | 356 | node_state->logger_realtime_start_time.time_since_epoch()) |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 357 | .count()); |
| 358 | } |
| 359 | |
| 360 | // TODO(austin): Add more useful times. When was this part started? What do |
| 361 | // we know about both the logger and remote then? |
| 362 | |
| 363 | log_file_header_builder.add_log_event_uuid(log_event_uuid_offset); |
| 364 | log_file_header_builder.add_logger_instance_uuid(logger_instance_uuid_offset); |
| 365 | if (!log_start_uuid_offset.IsNull()) { |
| 366 | log_file_header_builder.add_log_start_uuid(log_start_uuid_offset); |
| 367 | } |
| 368 | log_file_header_builder.add_logger_node_boot_uuid( |
| 369 | logger_node_boot_uuid_offset); |
| 370 | log_file_header_builder.add_source_node_boot_uuid( |
| 371 | source_node_boot_uuid_offset); |
| 372 | |
| 373 | log_file_header_builder.add_parts_uuid(parts_uuid_offset); |
| 374 | log_file_header_builder.add_parts_index(parts_index); |
| 375 | |
| 376 | if (!config_sha256_offset.IsNull()) { |
| 377 | log_file_header_builder.add_configuration_sha256(config_sha256_offset); |
| 378 | } |
| 379 | |
Austin Schuh | e46492f | 2021-07-31 19:49:41 -0700 | [diff] [blame] | 380 | log_file_header_builder.add_boot_uuids(boot_uuids_offset); |
Austin Schuh | a499cea | 2021-07-31 19:49:53 -0700 | [diff] [blame] | 381 | log_file_header_builder.add_logger_part_monotonic_start_time( |
| 382 | std::chrono::duration_cast<std::chrono::nanoseconds>( |
| 383 | event_loop_->monotonic_now().time_since_epoch()) |
| 384 | .count()); |
| 385 | log_file_header_builder.add_logger_part_realtime_start_time( |
| 386 | std::chrono::duration_cast<std::chrono::nanoseconds>( |
| 387 | event_loop_->realtime_now().time_since_epoch()) |
| 388 | .count()); |
Austin Schuh | 72211ae | 2021-08-05 14:02:30 -0700 | [diff] [blame] | 389 | log_file_header_builder.add_oldest_remote_monotonic_timestamps( |
| 390 | oldest_remote_monotonic_timestamps_offset); |
| 391 | log_file_header_builder.add_oldest_local_monotonic_timestamps( |
| 392 | oldest_local_monotonic_timestamps_offset); |
| 393 | log_file_header_builder.add_oldest_remote_unreliable_monotonic_timestamps( |
| 394 | oldest_remote_unreliable_monotonic_timestamps_offset); |
| 395 | log_file_header_builder.add_oldest_local_unreliable_monotonic_timestamps( |
| 396 | oldest_local_unreliable_monotonic_timestamps_offset); |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 397 | fbb.FinishSizePrefixed(log_file_header_builder.Finish()); |
| 398 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> result( |
| 399 | fbb.Release()); |
| 400 | |
| 401 | CHECK(result.Verify()) << ": Built a corrupted header."; |
| 402 | |
| 403 | return result; |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 404 | } |
| 405 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 406 | NewDataWriter *LocalLogNamer::MakeWriter(const Channel *channel) { |
Austin Schuh | df57647 | 2020-10-19 09:39:37 -0700 | [diff] [blame] | 407 | CHECK(configuration::ChannelIsSendableOnNode(channel, node())) |
| 408 | << ": " << configuration::CleanedChannelToString(channel); |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 409 | return &data_writer_; |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 410 | } |
| 411 | |
Austin Schuh | 7334084 | 2021-07-30 22:32:06 -0700 | [diff] [blame] | 412 | void LocalLogNamer::Rotate(const Node *node) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 413 | CHECK(node == this->node()); |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 414 | data_writer_.Rotate(); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 415 | } |
Austin Schuh | 8c39996 | 2020-12-25 21:51:45 -0800 | [diff] [blame] | 416 | |
| 417 | void LocalLogNamer::WriteConfiguration( |
| 418 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header, |
| 419 | std::string_view config_sha256) { |
| 420 | const std::string filename = absl::StrCat(base_name_, config_sha256, ".bfbs"); |
| 421 | |
| 422 | std::unique_ptr<DetachedBufferWriter> writer = |
| 423 | std::make_unique<DetachedBufferWriter>( |
| 424 | filename, std::make_unique<aos::logger::DummyEncoder>()); |
| 425 | writer->QueueSizedFlatbuffer(header->Release()); |
| 426 | } |
| 427 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 428 | NewDataWriter *LocalLogNamer::MakeTimestampWriter(const Channel *channel) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 429 | CHECK(configuration::ChannelIsReadableOnNode(channel, node_)) |
| 430 | << ": Message is not delivered to this node."; |
| 431 | CHECK(node_ != nullptr) << ": Can't log timestamps in a single node world"; |
| 432 | CHECK(configuration::ConnectionDeliveryTimeIsLoggedOnNode(channel, node_, |
| 433 | node_)) |
| 434 | << ": Delivery times aren't logged for this channel on this node."; |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 435 | return &data_writer_; |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 436 | } |
| 437 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 438 | NewDataWriter *LocalLogNamer::MakeForwardedTimestampWriter( |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 439 | const Channel * /*channel*/, const Node * /*node*/) { |
| 440 | LOG(FATAL) << "Can't log forwarded timestamps in a singe log file."; |
| 441 | return nullptr; |
| 442 | } |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 443 | MultiNodeLogNamer::MultiNodeLogNamer(std::string_view base_name, |
Austin Schuh | a499cea | 2021-07-31 19:49:53 -0700 | [diff] [blame] | 444 | EventLoop *event_loop) |
Austin Schuh | 5b728b7 | 2021-06-16 14:57:15 -0700 | [diff] [blame] | 445 | : MultiNodeLogNamer(base_name, event_loop->configuration(), event_loop, |
| 446 | event_loop->node()) {} |
| 447 | |
| 448 | MultiNodeLogNamer::MultiNodeLogNamer(std::string_view base_name, |
| 449 | const Configuration *configuration, |
| 450 | EventLoop *event_loop, const Node *node) |
| 451 | : LogNamer(configuration, event_loop, node), |
| 452 | base_name_(base_name), |
| 453 | old_base_name_() {} |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 454 | |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 455 | MultiNodeLogNamer::~MultiNodeLogNamer() { |
| 456 | if (!ran_out_of_space_) { |
| 457 | // This handles renaming temporary files etc. |
| 458 | Close(); |
| 459 | } |
| 460 | } |
| 461 | |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 462 | void MultiNodeLogNamer::Rotate(const Node *node) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 463 | if (node == this->node()) { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 464 | if (data_writer_) { |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 465 | data_writer_->Rotate(); |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 466 | } |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 467 | } else { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 468 | for (std::pair<const Channel *const, NewDataWriter> &data_writer : |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 469 | data_writers_) { |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 470 | if (node == data_writer.second.node()) { |
| 471 | data_writer.second.Rotate(); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 472 | } |
| 473 | } |
| 474 | } |
| 475 | } |
| 476 | |
Austin Schuh | 8c39996 | 2020-12-25 21:51:45 -0800 | [diff] [blame] | 477 | void MultiNodeLogNamer::WriteConfiguration( |
| 478 | aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> *header, |
| 479 | std::string_view config_sha256) { |
| 480 | if (ran_out_of_space_) { |
| 481 | return; |
| 482 | } |
| 483 | |
| 484 | const std::string_view separator = base_name_.back() == '/' ? "" : "_"; |
| 485 | const std::string filename = absl::StrCat( |
| 486 | base_name_, separator, config_sha256, ".bfbs", extension_, temp_suffix_); |
| 487 | |
| 488 | std::unique_ptr<DetachedBufferWriter> writer = |
| 489 | std::make_unique<DetachedBufferWriter>(filename, encoder_factory_()); |
| 490 | |
| 491 | writer->QueueSizedFlatbuffer(header->Release()); |
| 492 | |
| 493 | if (!writer->ran_out_of_space()) { |
Austin Schuh | 5b728b7 | 2021-06-16 14:57:15 -0700 | [diff] [blame] | 494 | all_filenames_.emplace_back( |
| 495 | absl::StrCat(config_sha256, ".bfbs", extension_)); |
Austin Schuh | 8c39996 | 2020-12-25 21:51:45 -0800 | [diff] [blame] | 496 | } |
| 497 | CloseWriter(&writer); |
| 498 | } |
| 499 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 500 | NewDataWriter *MultiNodeLogNamer::MakeWriter(const Channel *channel) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 501 | // See if we can read the data on this node at all. |
| 502 | const bool is_readable = |
| 503 | configuration::ChannelIsReadableOnNode(channel, this->node()); |
| 504 | if (!is_readable) { |
| 505 | return nullptr; |
| 506 | } |
| 507 | |
| 508 | // Then, see if we are supposed to log the data here. |
| 509 | const bool log_message = |
| 510 | configuration::ChannelMessageIsLoggedOnNode(channel, this->node()); |
| 511 | |
| 512 | if (!log_message) { |
| 513 | return nullptr; |
| 514 | } |
| 515 | |
| 516 | // Now, sort out if this is data generated on this node, or not. It is |
| 517 | // generated if it is sendable on this node. |
| 518 | if (configuration::ChannelIsSendableOnNode(channel, this->node())) { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 519 | if (!data_writer_) { |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 520 | OpenDataWriter(); |
| 521 | } |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 522 | return data_writer_.get(); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 523 | } |
| 524 | |
| 525 | // Ok, we have data that is being forwarded to us that we are supposed to |
| 526 | // log. It needs to be logged with send timestamps, but be sorted enough |
| 527 | // to be able to be processed. |
| 528 | CHECK(data_writers_.find(channel) == data_writers_.end()); |
| 529 | |
| 530 | // Track that this node is being logged. |
| 531 | const Node *source_node = configuration::GetNode( |
| 532 | configuration_, channel->source_node()->string_view()); |
| 533 | |
| 534 | if (std::find(nodes_.begin(), nodes_.end(), source_node) == nodes_.end()) { |
| 535 | nodes_.emplace_back(source_node); |
| 536 | } |
| 537 | |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 538 | NewDataWriter data_writer(this, source_node, |
| 539 | [this, channel](NewDataWriter *data_writer) { |
| 540 | OpenWriter(channel, data_writer); |
| 541 | }, |
| 542 | [this](NewDataWriter *data_writer) { |
| 543 | CloseWriter(&data_writer->writer); |
| 544 | }); |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 545 | return &( |
| 546 | data_writers_.emplace(channel, std::move(data_writer)).first->second); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 547 | } |
| 548 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 549 | NewDataWriter *MultiNodeLogNamer::MakeForwardedTimestampWriter( |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 550 | const Channel *channel, const Node *node) { |
| 551 | // See if we can read the data on this node at all. |
| 552 | const bool is_readable = |
| 553 | configuration::ChannelIsReadableOnNode(channel, this->node()); |
| 554 | CHECK(is_readable) << ": " << configuration::CleanedChannelToString(channel); |
| 555 | |
| 556 | CHECK(data_writers_.find(channel) == data_writers_.end()); |
| 557 | |
| 558 | if (std::find(nodes_.begin(), nodes_.end(), node) == nodes_.end()) { |
| 559 | nodes_.emplace_back(node); |
| 560 | } |
| 561 | |
Austin Schuh | 5b728b7 | 2021-06-16 14:57:15 -0700 | [diff] [blame] | 562 | NewDataWriter data_writer(this, configuration::GetNode(configuration_, node), |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 563 | [this, channel](NewDataWriter *data_writer) { |
| 564 | OpenForwardedTimestampWriter(channel, |
| 565 | data_writer); |
| 566 | }, |
| 567 | [this](NewDataWriter *data_writer) { |
| 568 | CloseWriter(&data_writer->writer); |
| 569 | }); |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 570 | return &( |
| 571 | data_writers_.emplace(channel, std::move(data_writer)).first->second); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 572 | } |
| 573 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 574 | NewDataWriter *MultiNodeLogNamer::MakeTimestampWriter(const Channel *channel) { |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 575 | bool log_delivery_times = false; |
| 576 | if (this->node() != nullptr) { |
| 577 | log_delivery_times = configuration::ConnectionDeliveryTimeIsLoggedOnNode( |
| 578 | channel, this->node(), this->node()); |
| 579 | } |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 580 | if (!log_delivery_times) { |
| 581 | return nullptr; |
| 582 | } |
| 583 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 584 | if (!data_writer_) { |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 585 | OpenDataWriter(); |
| 586 | } |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 587 | return data_writer_.get(); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 588 | } |
| 589 | |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 590 | void MultiNodeLogNamer::Close() { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 591 | data_writers_.clear(); |
| 592 | data_writer_.reset(); |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 593 | } |
| 594 | |
| 595 | void MultiNodeLogNamer::ResetStatistics() { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 596 | for (std::pair<const Channel *const, NewDataWriter> &data_writer : |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 597 | data_writers_) { |
Austin Schuh | ad0cfc3 | 2020-12-21 12:34:26 -0800 | [diff] [blame] | 598 | if (!data_writer.second.writer) continue; |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 599 | data_writer.second.writer->ResetStatistics(); |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 600 | } |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 601 | if (data_writer_) { |
| 602 | data_writer_->writer->ResetStatistics(); |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 603 | } |
| 604 | max_write_time_ = std::chrono::nanoseconds::zero(); |
| 605 | max_write_time_bytes_ = -1; |
| 606 | max_write_time_messages_ = -1; |
| 607 | total_write_time_ = std::chrono::nanoseconds::zero(); |
| 608 | total_write_count_ = 0; |
| 609 | total_write_messages_ = 0; |
| 610 | total_write_bytes_ = 0; |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 611 | } |
| 612 | |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 613 | void MultiNodeLogNamer::OpenForwardedTimestampWriter( |
| 614 | const Channel *channel, NewDataWriter *data_writer) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 615 | std::string filename = |
Austin Schuh | e715eae | 2020-10-10 15:39:30 -0700 | [diff] [blame] | 616 | absl::StrCat("timestamps", channel->name()->string_view(), "/", |
Brian Silverman | a621f52 | 2020-09-30 16:52:43 -0700 | [diff] [blame] | 617 | channel->type()->string_view(), ".part", |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 618 | data_writer->parts_index(), ".bfbs", extension_); |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 619 | CreateBufferWriter(filename, &data_writer->writer); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 620 | } |
| 621 | |
| 622 | void MultiNodeLogNamer::OpenWriter(const Channel *channel, |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 623 | NewDataWriter *data_writer) { |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 624 | const std::string filename = absl::StrCat( |
Austin Schuh | e715eae | 2020-10-10 15:39:30 -0700 | [diff] [blame] | 625 | CHECK_NOTNULL(channel->source_node())->string_view(), "_data", |
Brian Silverman | a621f52 | 2020-09-30 16:52:43 -0700 | [diff] [blame] | 626 | channel->name()->string_view(), "/", channel->type()->string_view(), |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 627 | ".part", data_writer->parts_index(), ".bfbs", extension_); |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 628 | CreateBufferWriter(filename, &data_writer->writer); |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 629 | } |
| 630 | |
Brian Silverman | a621f52 | 2020-09-30 16:52:43 -0700 | [diff] [blame] | 631 | void MultiNodeLogNamer::OpenDataWriter() { |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 632 | if (!data_writer_) { |
| 633 | data_writer_ = std::make_unique<NewDataWriter>( |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 634 | this, node_, |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 635 | [this](NewDataWriter *writer) { |
| 636 | std::string name; |
| 637 | if (node() != nullptr) { |
| 638 | name = absl::StrCat(name, node()->name()->string_view(), "_"); |
| 639 | } |
Austin Schuh | 572924a | 2021-07-30 22:32:12 -0700 | [diff] [blame] | 640 | absl::StrAppend(&name, "data.part", writer->parts_index(), ".bfbs", |
Austin Schuh | b8bca73 | 2021-07-30 22:32:00 -0700 | [diff] [blame] | 641 | extension_); |
| 642 | CreateBufferWriter(name, &writer->writer); |
| 643 | }, |
| 644 | [this](NewDataWriter *data_writer) { |
| 645 | CloseWriter(&data_writer->writer); |
| 646 | }); |
Brian Silverman | 7af8c90 | 2020-09-29 16:14:04 -0700 | [diff] [blame] | 647 | } |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 648 | } |
| 649 | |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 650 | void MultiNodeLogNamer::CreateBufferWriter( |
Brian Silverman | a621f52 | 2020-09-30 16:52:43 -0700 | [diff] [blame] | 651 | std::string_view path, std::unique_ptr<DetachedBufferWriter> *destination) { |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 652 | if (ran_out_of_space_) { |
| 653 | // Refuse to open any new files, which might skip data. Any existing files |
| 654 | // are in the same folder, which means they're on the same filesystem, which |
| 655 | // 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] | 656 | if (!destination->get()) { |
| 657 | // But avoid leaving a nullptr writer if we're out of space when |
| 658 | // attempting to open the first file. |
| 659 | *destination = std::make_unique<DetachedBufferWriter>( |
| 660 | DetachedBufferWriter::already_out_of_space_t()); |
| 661 | } |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 662 | return; |
| 663 | } |
Austin Schuh | e715eae | 2020-10-10 15:39:30 -0700 | [diff] [blame] | 664 | const std::string_view separator = base_name_.back() == '/' ? "" : "_"; |
| 665 | const std::string filename = |
| 666 | absl::StrCat(base_name_, separator, path, temp_suffix_); |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 667 | if (!destination->get()) { |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 668 | if (ran_out_of_space_) { |
| 669 | *destination = std::make_unique<DetachedBufferWriter>( |
| 670 | DetachedBufferWriter::already_out_of_space_t()); |
| 671 | return; |
| 672 | } |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 673 | *destination = |
| 674 | std::make_unique<DetachedBufferWriter>(filename, encoder_factory_()); |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 675 | if (!destination->get()->ran_out_of_space()) { |
| 676 | all_filenames_.emplace_back(path); |
| 677 | } |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 678 | return; |
| 679 | } |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 680 | |
| 681 | CloseWriter(destination); |
| 682 | if (ran_out_of_space_) { |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 683 | *destination->get() = |
| 684 | DetachedBufferWriter(DetachedBufferWriter::already_out_of_space_t()); |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 685 | return; |
| 686 | } |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 687 | |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 688 | *destination->get() = DetachedBufferWriter(filename, encoder_factory_()); |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 689 | if (!destination->get()->ran_out_of_space()) { |
| 690 | all_filenames_.emplace_back(path); |
| 691 | } |
Brian Silverman | 0465fcf | 2020-09-24 00:29:18 -0700 | [diff] [blame] | 692 | } |
| 693 | |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 694 | void MultiNodeLogNamer::RenameTempFile(DetachedBufferWriter *destination) { |
| 695 | if (temp_suffix_.empty()) { |
| 696 | return; |
| 697 | } |
Austin Schuh | 6bb8a82 | 2021-03-31 23:04:39 -0700 | [diff] [blame] | 698 | std::string current_filename = std::string(destination->filename()); |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 699 | CHECK(current_filename.size() > temp_suffix_.size()); |
Austin Schuh | 6bb8a82 | 2021-03-31 23:04:39 -0700 | [diff] [blame] | 700 | std::string final_filename = |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 701 | current_filename.substr(0, current_filename.size() - temp_suffix_.size()); |
Austin Schuh | 6bb8a82 | 2021-03-31 23:04:39 -0700 | [diff] [blame] | 702 | int result = rename(current_filename.c_str(), final_filename.c_str()); |
| 703 | |
| 704 | // When changing the base name, we rename the log folder while there active |
| 705 | // buffer writers. Therefore, the name of that active buffer may still refer |
| 706 | // to the old file location rather than the new one. This minimized changes to |
| 707 | // existing code. |
| 708 | if (result != 0 && errno != ENOSPC && !old_base_name_.empty()) { |
| 709 | auto offset = current_filename.find(old_base_name_); |
| 710 | if (offset != std::string::npos) { |
| 711 | current_filename.replace(offset, old_base_name_.length(), base_name_); |
| 712 | } |
| 713 | offset = final_filename.find(old_base_name_); |
| 714 | if (offset != std::string::npos) { |
| 715 | final_filename.replace(offset, old_base_name_.length(), base_name_); |
| 716 | } |
| 717 | result = rename(current_filename.c_str(), final_filename.c_str()); |
| 718 | } |
| 719 | |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 720 | if (result != 0) { |
| 721 | if (errno == ENOSPC) { |
| 722 | ran_out_of_space_ = true; |
| 723 | return; |
| 724 | } else { |
| 725 | PLOG(FATAL) << "Renaming " << current_filename << " to " << final_filename |
| 726 | << " failed"; |
| 727 | } |
Austin Schuh | 6bb8a82 | 2021-03-31 23:04:39 -0700 | [diff] [blame] | 728 | } else { |
| 729 | VLOG(1) << "Renamed " << current_filename << " -> " << final_filename; |
Brian Silverman | 48deab1 | 2020-09-30 18:39:28 -0700 | [diff] [blame] | 730 | } |
| 731 | } |
| 732 | |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 733 | void MultiNodeLogNamer::CloseWriter( |
| 734 | std::unique_ptr<DetachedBufferWriter> *writer_pointer) { |
| 735 | DetachedBufferWriter *const writer = writer_pointer->get(); |
| 736 | if (!writer) { |
| 737 | return; |
| 738 | } |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 739 | const bool was_open = writer->is_open(); |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 740 | writer->Close(); |
| 741 | |
| 742 | if (writer->max_write_time() > max_write_time_) { |
| 743 | max_write_time_ = writer->max_write_time(); |
| 744 | max_write_time_bytes_ = writer->max_write_time_bytes(); |
| 745 | max_write_time_messages_ = writer->max_write_time_messages(); |
| 746 | } |
| 747 | total_write_time_ += writer->total_write_time(); |
| 748 | total_write_count_ += writer->total_write_count(); |
| 749 | total_write_messages_ += writer->total_write_messages(); |
| 750 | total_write_bytes_ += writer->total_write_bytes(); |
| 751 | |
| 752 | if (writer->ran_out_of_space()) { |
| 753 | ran_out_of_space_ = true; |
| 754 | writer->acknowledge_out_of_space(); |
| 755 | } |
Brian Silverman | a9f2ec9 | 2020-10-06 18:00:53 -0700 | [diff] [blame] | 756 | if (was_open) { |
| 757 | RenameTempFile(writer); |
| 758 | } else { |
| 759 | CHECK(access(std::string(writer->filename()).c_str(), F_OK) == -1) |
| 760 | << ": File should not exist: " << writer->filename(); |
| 761 | } |
Brian Silverman | cb80582 | 2020-10-06 17:43:35 -0700 | [diff] [blame] | 762 | } |
| 763 | |
Austin Schuh | cb5601b | 2020-09-10 15:29:59 -0700 | [diff] [blame] | 764 | } // namespace logger |
| 765 | } // namespace aos |