blob: 8c6d936b4b2f88bf42fa3b6667dbe0393dafb68a [file] [log] [blame]
James Kuszmaul38735e82019-12-07 16:42:06 -08001#include "aos/events/logging/logger.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -08002
3#include <fcntl.h>
Austin Schuh4c4e0092019-12-22 16:18:03 -08004#include <limits.h>
Austin Schuhe309d2a2019-11-29 13:25:21 -08005#include <sys/stat.h>
6#include <sys/types.h>
7#include <sys/uio.h>
8#include <vector>
9
Austin Schuh2f8fd752020-09-01 22:38:28 -070010#include "absl/strings/escaping.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080011#include "absl/types/span.h"
12#include "aos/events/event_loop.h"
Austin Schuhf6f9bf32020-10-11 14:37:43 -070013#include "aos/events/logging/logfile_sorting.h"
James Kuszmaul38735e82019-12-07 16:42:06 -080014#include "aos/events/logging/logger_generated.h"
Austin Schuh64fab802020-09-09 22:47:47 -070015#include "aos/events/logging/uuid.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080016#include "aos/flatbuffer_merge.h"
Austin Schuh0ca1fd32020-12-18 22:53:05 -080017#include "aos/network/multinode_timestamp_filter.h"
Austin Schuh0de30f32020-12-06 12:44:28 -080018#include "aos/network/remote_message_generated.h"
19#include "aos/network/remote_message_schema.h"
Austin Schuh288479d2019-12-18 19:47:52 -080020#include "aos/network/team_number.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080021#include "aos/time/time.h"
Brian Silvermanae7c0332020-09-30 16:58:23 -070022#include "aos/util/file.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080023#include "flatbuffers/flatbuffers.h"
Austin Schuh8c399962020-12-25 21:51:45 -080024#include "openssl/sha.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080025
Austin Schuh15649d62019-12-28 16:36:38 -080026DEFINE_bool(skip_missing_forwarding_entries, false,
27 "If true, drop any forwarding entries with missing data. If "
28 "false, CHECK.");
Austin Schuhe309d2a2019-11-29 13:25:21 -080029
Austin Schuh0ca1fd32020-12-18 22:53:05 -080030DECLARE_bool(timestamps_to_csv);
Austin Schuh8bd96322020-02-13 21:18:22 -080031
Austin Schuh2f8fd752020-09-01 22:38:28 -070032DEFINE_bool(skip_order_validation, false,
33 "If true, ignore any out of orderness in replay");
34
Austin Schuhf0688662020-12-19 15:37:45 -080035DEFINE_double(
36 time_estimation_buffer_seconds, 2.0,
37 "The time to buffer ahead in the log file to accurately reconstruct time.");
38
Austin Schuhe309d2a2019-11-29 13:25:21 -080039namespace aos {
40namespace logger {
Austin Schuh0afc4d12020-10-19 11:42:04 -070041namespace {
Austin Schuh8c399962020-12-25 21:51:45 -080042std::string Sha256(const absl::Span<const uint8_t> str) {
43 unsigned char hash[SHA256_DIGEST_LENGTH];
44 SHA256_CTX sha256;
45 SHA256_Init(&sha256);
46 SHA256_Update(&sha256, str.data(), str.size());
47 SHA256_Final(hash, &sha256);
48 std::stringstream ss;
49 for (int i = 0; i < SHA256_DIGEST_LENGTH; i++) {
50 ss << std::hex << std::setw(2) << std::setfill('0') << (int)hash[i];
51 }
52 return ss.str();
53}
54
Austin Schuh315b96b2020-12-11 21:21:12 -080055std::string LogFileVectorToString(std::vector<LogFile> log_files) {
56 std::stringstream ss;
Austin Schuh297d2352021-01-21 19:02:17 -080057 for (const auto &f : log_files) {
Austin Schuh315b96b2020-12-11 21:21:12 -080058 ss << f << "\n";
59 }
60 return ss.str();
61}
62
Austin Schuh0de30f32020-12-06 12:44:28 -080063// Copies the channel, removing the schema as we go. If new_name is provided,
64// it is used instead of the name inside the channel. If new_type is provided,
65// it is used instead of the type in the channel.
66flatbuffers::Offset<Channel> CopyChannel(const Channel *c,
67 std::string_view new_name,
68 std::string_view new_type,
69 flatbuffers::FlatBufferBuilder *fbb) {
70 flatbuffers::Offset<flatbuffers::String> name_offset =
71 fbb->CreateSharedString(new_name.empty() ? c->name()->string_view()
72 : new_name);
73 flatbuffers::Offset<flatbuffers::String> type_offset =
74 fbb->CreateSharedString(new_type.empty() ? c->type()->str() : new_type);
75 flatbuffers::Offset<flatbuffers::String> source_node_offset =
76 c->has_source_node() ? fbb->CreateSharedString(c->source_node()->str())
77 : 0;
78
79 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Connection>>>
80 destination_nodes_offset =
81 aos::RecursiveCopyVectorTable(c->destination_nodes(), fbb);
82
83 flatbuffers::Offset<
84 flatbuffers::Vector<flatbuffers::Offset<flatbuffers::String>>>
85 logger_nodes_offset = aos::CopyVectorSharedString(c->logger_nodes(), fbb);
86
87 Channel::Builder channel_builder(*fbb);
88 channel_builder.add_name(name_offset);
89 channel_builder.add_type(type_offset);
90 if (c->has_frequency()) {
91 channel_builder.add_frequency(c->frequency());
92 }
93 if (c->has_max_size()) {
94 channel_builder.add_max_size(c->max_size());
95 }
96 if (c->has_num_senders()) {
97 channel_builder.add_num_senders(c->num_senders());
98 }
99 if (c->has_num_watchers()) {
100 channel_builder.add_num_watchers(c->num_watchers());
101 }
102 if (!source_node_offset.IsNull()) {
103 channel_builder.add_source_node(source_node_offset);
104 }
105 if (!destination_nodes_offset.IsNull()) {
106 channel_builder.add_destination_nodes(destination_nodes_offset);
107 }
108 if (c->has_logger()) {
109 channel_builder.add_logger(c->logger());
110 }
111 if (!logger_nodes_offset.IsNull()) {
112 channel_builder.add_logger_nodes(logger_nodes_offset);
113 }
114 if (c->has_read_method()) {
115 channel_builder.add_read_method(c->read_method());
116 }
117 if (c->has_num_readers()) {
118 channel_builder.add_num_readers(c->num_readers());
119 }
120 return channel_builder.Finish();
121}
122
Austin Schuhe309d2a2019-11-29 13:25:21 -0800123namespace chrono = std::chrono;
Austin Schuh0de30f32020-12-06 12:44:28 -0800124using message_bridge::RemoteMessage;
Austin Schuh0afc4d12020-10-19 11:42:04 -0700125} // namespace
Austin Schuhe309d2a2019-11-29 13:25:21 -0800126
Brian Silverman1f345222020-09-24 21:14:48 -0700127Logger::Logger(EventLoop *event_loop, const Configuration *configuration,
128 std::function<bool(const Channel *)> should_log)
Austin Schuhe309d2a2019-11-29 13:25:21 -0800129 : event_loop_(event_loop),
Austin Schuh0c297012020-09-16 18:41:59 -0700130 configuration_(configuration),
131 name_(network::GetHostname()),
Brian Silverman1f345222020-09-24 21:14:48 -0700132 timer_handler_(event_loop_->AddTimer(
133 [this]() { DoLogData(event_loop_->monotonic_now()); })),
Austin Schuh2f8fd752020-09-01 22:38:28 -0700134 server_statistics_fetcher_(
135 configuration::MultiNode(event_loop_->configuration())
136 ? event_loop_->MakeFetcher<message_bridge::ServerStatistics>(
137 "/aos")
138 : aos::Fetcher<message_bridge::ServerStatistics>()) {
Brian Silverman1f345222020-09-24 21:14:48 -0700139 VLOG(1) << "Creating logger for " << FlatbufferToJson(event_loop_->node());
Austin Schuh2f8fd752020-09-01 22:38:28 -0700140
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700141 // Find all the nodes which are logging timestamps on our node. This may
142 // over-estimate if should_log is specified.
143 std::vector<const Node *> timestamp_logger_nodes =
144 configuration::TimestampNodes(configuration_, event_loop_->node());
Austin Schuh2f8fd752020-09-01 22:38:28 -0700145
146 std::map<const Channel *, const Node *> timestamp_logger_channels;
147
148 // Now that we have all the nodes accumulated, make remote timestamp loggers
149 // for them.
150 for (const Node *node : timestamp_logger_nodes) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700151 // Note: since we are doing a find using the event loop channel, we need to
152 // make sure this channel pointer is part of the event loop configuration,
153 // not configuration_. This only matters when configuration_ !=
154 // event_loop->configuration();
Austin Schuh2f8fd752020-09-01 22:38:28 -0700155 const Channel *channel = configuration::GetChannel(
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700156 event_loop->configuration(),
Austin Schuh2f8fd752020-09-01 22:38:28 -0700157 absl::StrCat("/aos/remote_timestamps/", node->name()->string_view()),
Austin Schuh0de30f32020-12-06 12:44:28 -0800158 RemoteMessage::GetFullyQualifiedName(), event_loop_->name(),
Austin Schuh2f8fd752020-09-01 22:38:28 -0700159 event_loop_->node());
160
161 CHECK(channel != nullptr)
162 << ": Remote timestamps are logged on "
163 << event_loop_->node()->name()->string_view()
164 << " but can't find channel /aos/remote_timestamps/"
165 << node->name()->string_view();
Brian Silverman1f345222020-09-24 21:14:48 -0700166 if (!should_log(channel)) {
167 continue;
168 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700169 timestamp_logger_channels.insert(std::make_pair(channel, node));
170 }
171
Brian Silvermand90905f2020-09-23 14:42:56 -0700172 const size_t our_node_index =
173 configuration::GetNodeIndex(configuration_, event_loop_->node());
Austin Schuh2f8fd752020-09-01 22:38:28 -0700174
Brian Silverman1f345222020-09-24 21:14:48 -0700175 for (size_t channel_index = 0;
176 channel_index < configuration_->channels()->size(); ++channel_index) {
177 const Channel *const config_channel =
178 configuration_->channels()->Get(channel_index);
Austin Schuh0c297012020-09-16 18:41:59 -0700179 // The MakeRawFetcher method needs a channel which is in the event loop
180 // configuration() object, not the configuration_ object. Go look that up
181 // from the config.
182 const Channel *channel = aos::configuration::GetChannel(
183 event_loop_->configuration(), config_channel->name()->string_view(),
184 config_channel->type()->string_view(), "", event_loop_->node());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700185 CHECK(channel != nullptr)
186 << ": Failed to look up channel "
187 << aos::configuration::CleanedChannelToString(config_channel);
Brian Silverman1f345222020-09-24 21:14:48 -0700188 if (!should_log(channel)) {
189 continue;
190 }
Austin Schuh0c297012020-09-16 18:41:59 -0700191
Austin Schuhe309d2a2019-11-29 13:25:21 -0800192 FetcherStruct fs;
Brian Silverman1f345222020-09-24 21:14:48 -0700193 fs.channel_index = channel_index;
194 fs.channel = channel;
195
Austin Schuh6f3babe2020-01-26 20:34:50 -0800196 const bool is_local =
197 configuration::ChannelIsSendableOnNode(channel, event_loop_->node());
198
Austin Schuh15649d62019-12-28 16:36:38 -0800199 const bool is_readable =
200 configuration::ChannelIsReadableOnNode(channel, event_loop_->node());
Brian Silverman1f345222020-09-24 21:14:48 -0700201 const bool is_logged = configuration::ChannelMessageIsLoggedOnNode(
202 channel, event_loop_->node());
203 const bool log_message = is_logged && is_readable;
Austin Schuh15649d62019-12-28 16:36:38 -0800204
Brian Silverman1f345222020-09-24 21:14:48 -0700205 bool log_delivery_times = false;
206 if (event_loop_->node() != nullptr) {
207 log_delivery_times = configuration::ConnectionDeliveryTimeIsLoggedOnNode(
208 channel, event_loop_->node(), event_loop_->node());
209 }
Austin Schuh15649d62019-12-28 16:36:38 -0800210
Austin Schuh0de30f32020-12-06 12:44:28 -0800211 // Now, detect a RemoteMessage timestamp logger where we should just log the
Austin Schuh2f8fd752020-09-01 22:38:28 -0700212 // contents to a file directly.
213 const bool log_contents = timestamp_logger_channels.find(channel) !=
214 timestamp_logger_channels.end();
Austin Schuh2f8fd752020-09-01 22:38:28 -0700215
216 if (log_message || log_delivery_times || log_contents) {
Austin Schuh15649d62019-12-28 16:36:38 -0800217 fs.fetcher = event_loop->MakeRawFetcher(channel);
218 VLOG(1) << "Logging channel "
219 << configuration::CleanedChannelToString(channel);
220
221 if (log_delivery_times) {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800222 VLOG(1) << " Delivery times";
Brian Silverman1f345222020-09-24 21:14:48 -0700223 fs.wants_timestamp_writer = true;
Austin Schuh315b96b2020-12-11 21:21:12 -0800224 fs.timestamp_node_index = our_node_index;
Austin Schuh15649d62019-12-28 16:36:38 -0800225 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800226 if (log_message) {
227 VLOG(1) << " Data";
Brian Silverman1f345222020-09-24 21:14:48 -0700228 fs.wants_writer = true;
Austin Schuh6f3babe2020-01-26 20:34:50 -0800229 if (!is_local) {
Austin Schuh315b96b2020-12-11 21:21:12 -0800230 const Node *source_node = configuration::GetNode(
231 configuration_, channel->source_node()->string_view());
232 fs.data_node_index =
233 configuration::GetNodeIndex(configuration_, source_node);
Austin Schuh6f3babe2020-01-26 20:34:50 -0800234 fs.log_type = LogType::kLogRemoteMessage;
Austin Schuh315b96b2020-12-11 21:21:12 -0800235 } else {
236 fs.data_node_index = our_node_index;
Austin Schuh6f3babe2020-01-26 20:34:50 -0800237 }
238 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700239 if (log_contents) {
240 VLOG(1) << "Timestamp logger channel "
241 << configuration::CleanedChannelToString(channel);
Brian Silverman1f345222020-09-24 21:14:48 -0700242 fs.timestamp_node = timestamp_logger_channels.find(channel)->second;
243 fs.wants_contents_writer = true;
Austin Schuh315b96b2020-12-11 21:21:12 -0800244 fs.contents_node_index =
Brian Silverman1f345222020-09-24 21:14:48 -0700245 configuration::GetNodeIndex(configuration_, fs.timestamp_node);
Austin Schuh2f8fd752020-09-01 22:38:28 -0700246 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800247 fetchers_.emplace_back(std::move(fs));
Austin Schuh15649d62019-12-28 16:36:38 -0800248 }
Brian Silverman1f345222020-09-24 21:14:48 -0700249 }
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700250
251 // When we are logging remote timestamps, we need to be able to translate from
252 // the channel index that the event loop uses to the channel index in the
253 // config in the log file.
254 event_loop_to_logged_channel_index_.resize(
255 event_loop->configuration()->channels()->size(), -1);
256 for (size_t event_loop_channel_index = 0;
257 event_loop_channel_index <
258 event_loop->configuration()->channels()->size();
259 ++event_loop_channel_index) {
260 const Channel *event_loop_channel =
261 event_loop->configuration()->channels()->Get(event_loop_channel_index);
262
263 const Channel *logged_channel = aos::configuration::GetChannel(
264 configuration_, event_loop_channel->name()->string_view(),
265 event_loop_channel->type()->string_view(), "",
266 configuration::GetNode(configuration_, event_loop_->node()));
267
268 if (logged_channel != nullptr) {
269 event_loop_to_logged_channel_index_[event_loop_channel_index] =
270 configuration::ChannelIndex(configuration_, logged_channel);
271 }
272 }
Brian Silverman1f345222020-09-24 21:14:48 -0700273}
274
275Logger::~Logger() {
276 if (log_namer_) {
277 // If we are replaying a log file, or in simulation, we want to force the
278 // last bit of data to be logged. The easiest way to deal with this is to
279 // poll everything as we go to destroy the class, ie, shut down the logger,
280 // and write it to disk.
281 StopLogging(event_loop_->monotonic_now());
282 }
283}
284
Brian Silvermanae7c0332020-09-30 16:58:23 -0700285void Logger::StartLogging(std::unique_ptr<LogNamer> log_namer,
286 std::string_view log_start_uuid) {
Brian Silverman1f345222020-09-24 21:14:48 -0700287 CHECK(!log_namer_) << ": Already logging";
288 log_namer_ = std::move(log_namer);
Austin Schuh8c399962020-12-25 21:51:45 -0800289
290 std::string config_sha256;
291 if (separate_config_) {
292 flatbuffers::FlatBufferBuilder fbb;
293 flatbuffers::Offset<aos::Configuration> configuration_offset =
294 CopyFlatBuffer(configuration_, &fbb);
295 LogFileHeader::Builder log_file_header_builder(fbb);
296 log_file_header_builder.add_configuration(configuration_offset);
297 fbb.FinishSizePrefixed(log_file_header_builder.Finish());
298 aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> config_header(
299 fbb.Release());
300 config_sha256 = Sha256(config_header.span());
301 LOG(INFO) << "Config sha256 of " << config_sha256;
302 log_namer_->WriteConfiguration(&config_header, config_sha256);
303 }
304
Brian Silvermanae7c0332020-09-30 16:58:23 -0700305 log_event_uuid_ = UUID::Random();
306 log_start_uuid_ = log_start_uuid;
Brian Silverman1f345222020-09-24 21:14:48 -0700307 VLOG(1) << "Starting logger for " << FlatbufferToJson(event_loop_->node());
308
309 // We want to do as much work as possible before the initial Fetch. Time
310 // between that and actually starting to log opens up the possibility of
311 // falling off the end of the queue during that time.
312
313 for (FetcherStruct &f : fetchers_) {
314 if (f.wants_writer) {
315 f.writer = log_namer_->MakeWriter(f.channel);
316 }
317 if (f.wants_timestamp_writer) {
318 f.timestamp_writer = log_namer_->MakeTimestampWriter(f.channel);
319 }
320 if (f.wants_contents_writer) {
321 f.contents_writer = log_namer_->MakeForwardedTimestampWriter(
322 f.channel, CHECK_NOTNULL(f.timestamp_node));
323 }
Austin Schuhe309d2a2019-11-29 13:25:21 -0800324 }
325
Brian Silverman1f345222020-09-24 21:14:48 -0700326 CHECK(node_state_.empty());
Austin Schuh0c297012020-09-16 18:41:59 -0700327 node_state_.resize(configuration::MultiNode(configuration_)
328 ? configuration_->nodes()->size()
Austin Schuh2f8fd752020-09-01 22:38:28 -0700329 : 1u);
Austin Schuhe309d2a2019-11-29 13:25:21 -0800330
Austin Schuh2f8fd752020-09-01 22:38:28 -0700331 for (const Node *node : log_namer_->nodes()) {
Brian Silvermand90905f2020-09-23 14:42:56 -0700332 const int node_index = configuration::GetNodeIndex(configuration_, node);
Austin Schuhe309d2a2019-11-29 13:25:21 -0800333
Austin Schuh816e5d62021-01-05 23:42:20 -0800334 node_state_[node_index].log_file_header = MakeHeader(node, config_sha256);
Austin Schuh2f8fd752020-09-01 22:38:28 -0700335 }
Austin Schuhe309d2a2019-11-29 13:25:21 -0800336
Austin Schuh2f8fd752020-09-01 22:38:28 -0700337 // Grab data from each channel right before we declare the log file started
338 // so we can capture the latest message on each channel. This lets us have
339 // non periodic messages with configuration that now get logged.
340 for (FetcherStruct &f : fetchers_) {
Brian Silvermancb805822020-10-06 17:43:35 -0700341 const auto start = event_loop_->monotonic_now();
342 const bool got_new = f.fetcher->Fetch();
343 const auto end = event_loop_->monotonic_now();
344 RecordFetchResult(start, end, got_new, &f);
345
346 // If there is a message, we want to write it.
347 f.written = f.fetcher->context().data == nullptr;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700348 }
349
350 // Clear out any old timestamps in case we are re-starting logging.
351 for (size_t i = 0; i < node_state_.size(); ++i) {
Austin Schuh315b96b2020-12-11 21:21:12 -0800352 SetStartTime(i, monotonic_clock::min_time, realtime_clock::min_time,
353 monotonic_clock::min_time, realtime_clock::min_time);
Austin Schuh2f8fd752020-09-01 22:38:28 -0700354 }
355
356 WriteHeader();
357
358 LOG(INFO) << "Logging node as " << FlatbufferToJson(event_loop_->node())
359 << " start_time " << last_synchronized_time_;
360
Austin Schuh315b96b2020-12-11 21:21:12 -0800361 // Force logging up until the start of the log file now, so the messages at
362 // the start are always ordered before the rest of the messages.
363 // Note: this ship may have already sailed, but we don't have to make it
364 // worse.
365 // TODO(austin): Test...
366 LogUntil(last_synchronized_time_);
367
Austin Schuh2f8fd752020-09-01 22:38:28 -0700368 timer_handler_->Setup(event_loop_->monotonic_now() + polling_period_,
369 polling_period_);
370}
371
Brian Silverman1f345222020-09-24 21:14:48 -0700372std::unique_ptr<LogNamer> Logger::StopLogging(
373 aos::monotonic_clock::time_point end_time) {
374 CHECK(log_namer_) << ": Not logging right now";
375
376 if (end_time != aos::monotonic_clock::min_time) {
377 LogUntil(end_time);
378 }
379 timer_handler_->Disable();
380
381 for (FetcherStruct &f : fetchers_) {
382 f.writer = nullptr;
383 f.timestamp_writer = nullptr;
384 f.contents_writer = nullptr;
385 }
386 node_state_.clear();
387
Brian Silvermanae7c0332020-09-30 16:58:23 -0700388 log_event_uuid_ = UUID::Zero();
389 log_start_uuid_ = std::string();
390
Brian Silverman1f345222020-09-24 21:14:48 -0700391 return std::move(log_namer_);
392}
393
Austin Schuhfa895892020-01-07 20:07:41 -0800394void Logger::WriteHeader() {
Austin Schuh0c297012020-09-16 18:41:59 -0700395 if (configuration::MultiNode(configuration_)) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700396 server_statistics_fetcher_.Fetch();
397 }
398
399 aos::monotonic_clock::time_point monotonic_start_time =
400 event_loop_->monotonic_now();
401 aos::realtime_clock::time_point realtime_start_time =
402 event_loop_->realtime_now();
403
404 // We need to pick a point in time to declare the log file "started". This
405 // starts here. It needs to be after everything is fetched so that the
406 // fetchers are all pointed at the most recent message before the start
407 // time.
408 last_synchronized_time_ = monotonic_start_time;
409
Austin Schuh6f3babe2020-01-26 20:34:50 -0800410 for (const Node *node : log_namer_->nodes()) {
Brian Silvermand90905f2020-09-23 14:42:56 -0700411 const int node_index = configuration::GetNodeIndex(configuration_, node);
Austin Schuh2f8fd752020-09-01 22:38:28 -0700412 MaybeUpdateTimestamp(node, node_index, monotonic_start_time,
413 realtime_start_time);
Austin Schuh315b96b2020-12-11 21:21:12 -0800414 MaybeWriteHeader(node_index, node);
Austin Schuh6f3babe2020-01-26 20:34:50 -0800415 }
416}
Austin Schuh8bd96322020-02-13 21:18:22 -0800417
Austin Schuh315b96b2020-12-11 21:21:12 -0800418void Logger::MaybeWriteHeader(int node_index) {
419 if (configuration::MultiNode(configuration_)) {
420 return MaybeWriteHeader(node_index,
421 configuration_->nodes()->Get(node_index));
422 } else {
423 return MaybeWriteHeader(node_index, nullptr);
424 }
425}
426
427void Logger::MaybeWriteHeader(int node_index, const Node *node) {
428 // This function is responsible for writing the header when the header both
429 // has valid data, and when it needs to be written.
430 if (node_state_[node_index].header_written &&
431 node_state_[node_index].header_valid) {
432 // The header has been written and is valid, nothing to do.
433 return;
434 }
435 if (!node_state_[node_index].has_source_node_boot_uuid) {
436 // Can't write a header if we don't have the boot UUID.
437 return;
438 }
439
440 // WriteHeader writes the first header in a log file. We want to do this only
441 // once.
442 //
443 // Rotate rewrites the same header with a new part ID, but keeps the same part
444 // UUID. We don't want that when things reboot, because that implies that
445 // parts go together across a reboot.
446 //
447 // Reboot resets the parts UUID. So, once we've written a header the first
448 // time, we want to use Reboot to rotate the log and reset the parts UUID.
449 //
450 // header_valid is cleared whenever the remote reboots.
451 if (node_state_[node_index].header_written) {
452 log_namer_->Reboot(node, &node_state_[node_index].log_file_header);
453 } else {
454 log_namer_->WriteHeader(&node_state_[node_index].log_file_header, node);
455
456 node_state_[node_index].header_written = true;
457 }
458 node_state_[node_index].header_valid = true;
459}
460
Austin Schuh2f8fd752020-09-01 22:38:28 -0700461void Logger::WriteMissingTimestamps() {
Austin Schuh0c297012020-09-16 18:41:59 -0700462 if (configuration::MultiNode(configuration_)) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700463 server_statistics_fetcher_.Fetch();
464 } else {
465 return;
466 }
467
468 if (server_statistics_fetcher_.get() == nullptr) {
469 return;
470 }
471
472 for (const Node *node : log_namer_->nodes()) {
Brian Silvermand90905f2020-09-23 14:42:56 -0700473 const int node_index = configuration::GetNodeIndex(configuration_, node);
Austin Schuh2f8fd752020-09-01 22:38:28 -0700474 if (MaybeUpdateTimestamp(
475 node, node_index,
476 server_statistics_fetcher_.context().monotonic_event_time,
477 server_statistics_fetcher_.context().realtime_event_time)) {
Austin Schuh315b96b2020-12-11 21:21:12 -0800478 CHECK(node_state_[node_index].header_written);
479 CHECK(node_state_[node_index].header_valid);
Austin Schuh64fab802020-09-09 22:47:47 -0700480 log_namer_->Rotate(node, &node_state_[node_index].log_file_header);
Austin Schuh315b96b2020-12-11 21:21:12 -0800481 } else {
482 MaybeWriteHeader(node_index, node);
Austin Schuh2f8fd752020-09-01 22:38:28 -0700483 }
484 }
485}
486
Austin Schuh315b96b2020-12-11 21:21:12 -0800487void Logger::SetStartTime(
488 size_t node_index, aos::monotonic_clock::time_point monotonic_start_time,
489 aos::realtime_clock::time_point realtime_start_time,
490 aos::monotonic_clock::time_point logger_monotonic_start_time,
491 aos::realtime_clock::time_point logger_realtime_start_time) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700492 node_state_[node_index].monotonic_start_time = monotonic_start_time;
493 node_state_[node_index].realtime_start_time = realtime_start_time;
494 node_state_[node_index]
495 .log_file_header.mutable_message()
496 ->mutate_monotonic_start_time(
497 std::chrono::duration_cast<std::chrono::nanoseconds>(
498 monotonic_start_time.time_since_epoch())
499 .count());
Austin Schuh315b96b2020-12-11 21:21:12 -0800500
501 // Add logger start times if they are available in the log file header.
502 if (node_state_[node_index]
503 .log_file_header.mutable_message()
504 ->has_logger_monotonic_start_time()) {
505 node_state_[node_index]
506 .log_file_header.mutable_message()
507 ->mutate_logger_monotonic_start_time(
508 std::chrono::duration_cast<std::chrono::nanoseconds>(
509 logger_monotonic_start_time.time_since_epoch())
510 .count());
511 }
512
513 if (node_state_[node_index]
514 .log_file_header.mutable_message()
515 ->has_logger_realtime_start_time()) {
516 node_state_[node_index]
517 .log_file_header.mutable_message()
518 ->mutate_logger_realtime_start_time(
519 std::chrono::duration_cast<std::chrono::nanoseconds>(
520 logger_realtime_start_time.time_since_epoch())
521 .count());
522 }
523
Austin Schuh2f8fd752020-09-01 22:38:28 -0700524 if (node_state_[node_index]
525 .log_file_header.mutable_message()
526 ->has_realtime_start_time()) {
527 node_state_[node_index]
528 .log_file_header.mutable_message()
529 ->mutate_realtime_start_time(
530 std::chrono::duration_cast<std::chrono::nanoseconds>(
531 realtime_start_time.time_since_epoch())
532 .count());
533 }
534}
535
536bool Logger::MaybeUpdateTimestamp(
537 const Node *node, int node_index,
538 aos::monotonic_clock::time_point monotonic_start_time,
539 aos::realtime_clock::time_point realtime_start_time) {
Brian Silverman87ac0402020-09-17 14:47:01 -0700540 // Bail early if the start times are already set.
Austin Schuh2f8fd752020-09-01 22:38:28 -0700541 if (node_state_[node_index].monotonic_start_time !=
542 monotonic_clock::min_time) {
543 return false;
544 }
Austin Schuh315b96b2020-12-11 21:21:12 -0800545 if (event_loop_->node() == node ||
546 !configuration::MultiNode(configuration_)) {
547 // There are no offsets to compute for ourself, so always succeed.
548 SetStartTime(node_index, monotonic_start_time, realtime_start_time,
549 monotonic_start_time, realtime_start_time);
550 node_state_[node_index].SetBootUUID(event_loop_->boot_uuid().string_view());
Austin Schuh2f8fd752020-09-01 22:38:28 -0700551 return true;
Austin Schuh315b96b2020-12-11 21:21:12 -0800552 } else if (server_statistics_fetcher_.get() != nullptr) {
553 // We must be a remote node now. Look for the connection and see if it is
554 // connected.
555
556 for (const message_bridge::ServerConnection *connection :
557 *server_statistics_fetcher_->connections()) {
558 if (connection->node()->name()->string_view() !=
559 node->name()->string_view()) {
560 continue;
561 }
562
563 if (connection->state() != message_bridge::State::CONNECTED) {
564 VLOG(1) << node->name()->string_view()
565 << " is not connected, can't start it yet.";
566 break;
567 }
568
569 // Update the boot UUID as soon as we know we are connected.
570 if (!connection->has_boot_uuid()) {
571 VLOG(1) << "Missing boot_uuid for node " << aos::FlatbufferToJson(node);
572 break;
573 }
574
575 if (!node_state_[node_index].has_source_node_boot_uuid ||
576 node_state_[node_index].source_node_boot_uuid !=
577 connection->boot_uuid()->string_view()) {
578 node_state_[node_index].SetBootUUID(
579 connection->boot_uuid()->string_view());
580 }
581
582 if (!connection->has_monotonic_offset()) {
583 VLOG(1) << "Missing monotonic offset for setting start time for node "
584 << aos::FlatbufferToJson(node);
585 break;
586 }
587
588 // Found it and it is connected. Compensate and go.
589 SetStartTime(node_index,
590 monotonic_start_time +
591 std::chrono::nanoseconds(connection->monotonic_offset()),
592 realtime_start_time, monotonic_start_time,
593 realtime_start_time);
594 return true;
595 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700596 }
597 return false;
598}
599
600aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> Logger::MakeHeader(
Austin Schuh8c399962020-12-25 21:51:45 -0800601 const Node *node, std::string_view config_sha256) {
Austin Schuhfa895892020-01-07 20:07:41 -0800602 // Now write the header with this timestamp in it.
603 flatbuffers::FlatBufferBuilder fbb;
Austin Schuhd7b15da2020-02-17 15:06:11 -0800604 fbb.ForceDefaults(true);
Austin Schuhfa895892020-01-07 20:07:41 -0800605
Austin Schuh8c399962020-12-25 21:51:45 -0800606 flatbuffers::Offset<aos::Configuration> configuration_offset;
607 if (!separate_config_) {
608 configuration_offset = CopyFlatBuffer(configuration_, &fbb);
609 } else {
610 CHECK(!config_sha256.empty());
611 }
Austin Schuhfa895892020-01-07 20:07:41 -0800612
Brian Silvermanae7c0332020-09-30 16:58:23 -0700613 const flatbuffers::Offset<flatbuffers::String> name_offset =
Austin Schuh0c297012020-09-16 18:41:59 -0700614 fbb.CreateString(name_);
Austin Schuhfa895892020-01-07 20:07:41 -0800615
Brian Silvermanae7c0332020-09-30 16:58:23 -0700616 CHECK(log_event_uuid_ != UUID::Zero());
617 const flatbuffers::Offset<flatbuffers::String> log_event_uuid_offset =
618 fbb.CreateString(log_event_uuid_.string_view());
Austin Schuh64fab802020-09-09 22:47:47 -0700619
Brian Silvermanae7c0332020-09-30 16:58:23 -0700620 const flatbuffers::Offset<flatbuffers::String> logger_instance_uuid_offset =
621 fbb.CreateString(logger_instance_uuid_.string_view());
622
623 flatbuffers::Offset<flatbuffers::String> log_start_uuid_offset;
624 if (!log_start_uuid_.empty()) {
625 log_start_uuid_offset = fbb.CreateString(log_start_uuid_);
626 }
627
Austin Schuh8c399962020-12-25 21:51:45 -0800628 flatbuffers::Offset<flatbuffers::String> config_sha256_offset;
629 if (!config_sha256.empty()) {
630 config_sha256_offset = fbb.CreateString(config_sha256);
631 }
632
Austin Schuh315b96b2020-12-11 21:21:12 -0800633 const flatbuffers::Offset<flatbuffers::String> logger_node_boot_uuid_offset =
634 fbb.CreateString(event_loop_->boot_uuid().string_view());
635
636 const flatbuffers::Offset<flatbuffers::String> source_node_boot_uuid_offset =
637 fbb.CreateString(event_loop_->boot_uuid().string_view());
Brian Silvermanae7c0332020-09-30 16:58:23 -0700638
639 const flatbuffers::Offset<flatbuffers::String> parts_uuid_offset =
Austin Schuh64fab802020-09-09 22:47:47 -0700640 fbb.CreateString("00000000-0000-4000-8000-000000000000");
641
Austin Schuhfa895892020-01-07 20:07:41 -0800642 flatbuffers::Offset<Node> node_offset;
Brian Silverman80993c22020-10-01 15:05:19 -0700643 flatbuffers::Offset<Node> logger_node_offset;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700644
Austin Schuh0c297012020-09-16 18:41:59 -0700645 if (configuration::MultiNode(configuration_)) {
Austin Schuha4fc60f2020-11-01 23:06:47 -0800646 node_offset = RecursiveCopyFlatBuffer(node, &fbb);
647 logger_node_offset = RecursiveCopyFlatBuffer(event_loop_->node(), &fbb);
Austin Schuhfa895892020-01-07 20:07:41 -0800648 }
649
650 aos::logger::LogFileHeader::Builder log_file_header_builder(fbb);
651
Austin Schuh64fab802020-09-09 22:47:47 -0700652 log_file_header_builder.add_name(name_offset);
Austin Schuhfa895892020-01-07 20:07:41 -0800653
654 // Only add the node if we are running in a multinode configuration.
Austin Schuh6f3babe2020-01-26 20:34:50 -0800655 if (node != nullptr) {
Austin Schuhfa895892020-01-07 20:07:41 -0800656 log_file_header_builder.add_node(node_offset);
Brian Silverman80993c22020-10-01 15:05:19 -0700657 log_file_header_builder.add_logger_node(logger_node_offset);
Austin Schuhfa895892020-01-07 20:07:41 -0800658 }
659
Austin Schuh8c399962020-12-25 21:51:45 -0800660 if (!configuration_offset.IsNull()) {
661 log_file_header_builder.add_configuration(configuration_offset);
662 }
Austin Schuhfa895892020-01-07 20:07:41 -0800663 // The worst case theoretical out of order is the polling period times 2.
664 // One message could get logged right after the boundary, but be for right
665 // before the next boundary. And the reverse could happen for another
666 // message. Report back 3x to be extra safe, and because the cost isn't
667 // huge on the read side.
668 log_file_header_builder.add_max_out_of_order_duration(
Brian Silverman1f345222020-09-24 21:14:48 -0700669 std::chrono::nanoseconds(3 * polling_period_).count());
Austin Schuhfa895892020-01-07 20:07:41 -0800670
671 log_file_header_builder.add_monotonic_start_time(
672 std::chrono::duration_cast<std::chrono::nanoseconds>(
Austin Schuh2f8fd752020-09-01 22:38:28 -0700673 monotonic_clock::min_time.time_since_epoch())
Austin Schuhfa895892020-01-07 20:07:41 -0800674 .count());
Austin Schuh2f8fd752020-09-01 22:38:28 -0700675 if (node == event_loop_->node()) {
676 log_file_header_builder.add_realtime_start_time(
677 std::chrono::duration_cast<std::chrono::nanoseconds>(
678 realtime_clock::min_time.time_since_epoch())
679 .count());
Austin Schuh315b96b2020-12-11 21:21:12 -0800680 } else {
681 log_file_header_builder.add_logger_monotonic_start_time(
682 std::chrono::duration_cast<std::chrono::nanoseconds>(
683 monotonic_clock::min_time.time_since_epoch())
684 .count());
685 log_file_header_builder.add_logger_realtime_start_time(
686 std::chrono::duration_cast<std::chrono::nanoseconds>(
687 realtime_clock::min_time.time_since_epoch())
688 .count());
Austin Schuh6f3babe2020-01-26 20:34:50 -0800689 }
690
Brian Silvermanae7c0332020-09-30 16:58:23 -0700691 log_file_header_builder.add_log_event_uuid(log_event_uuid_offset);
692 log_file_header_builder.add_logger_instance_uuid(logger_instance_uuid_offset);
693 if (!log_start_uuid_offset.IsNull()) {
694 log_file_header_builder.add_log_start_uuid(log_start_uuid_offset);
695 }
Austin Schuh315b96b2020-12-11 21:21:12 -0800696 log_file_header_builder.add_logger_node_boot_uuid(
697 logger_node_boot_uuid_offset);
698 log_file_header_builder.add_source_node_boot_uuid(
699 source_node_boot_uuid_offset);
Austin Schuh64fab802020-09-09 22:47:47 -0700700
701 log_file_header_builder.add_parts_uuid(parts_uuid_offset);
702 log_file_header_builder.add_parts_index(0);
703
Austin Schuh8c399962020-12-25 21:51:45 -0800704 log_file_header_builder.add_configuration_sha256(0);
705
706 if (!config_sha256_offset.IsNull()) {
707 log_file_header_builder.add_configuration_sha256(config_sha256_offset);
708 }
709
Austin Schuh2f8fd752020-09-01 22:38:28 -0700710 fbb.FinishSizePrefixed(log_file_header_builder.Finish());
Austin Schuha4fc60f2020-11-01 23:06:47 -0800711 aos::SizePrefixedFlatbufferDetachedBuffer<LogFileHeader> result(
712 fbb.Release());
713
714 CHECK(result.Verify()) << ": Built a corrupted header.";
715
716 return result;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700717}
718
Brian Silvermancb805822020-10-06 17:43:35 -0700719void Logger::ResetStatisics() {
720 max_message_fetch_time_ = std::chrono::nanoseconds::zero();
721 max_message_fetch_time_channel_ = -1;
722 max_message_fetch_time_size_ = -1;
723 total_message_fetch_time_ = std::chrono::nanoseconds::zero();
724 total_message_fetch_count_ = 0;
725 total_message_fetch_bytes_ = 0;
726 total_nop_fetch_time_ = std::chrono::nanoseconds::zero();
727 total_nop_fetch_count_ = 0;
728 max_copy_time_ = std::chrono::nanoseconds::zero();
729 max_copy_time_channel_ = -1;
730 max_copy_time_size_ = -1;
731 total_copy_time_ = std::chrono::nanoseconds::zero();
732 total_copy_count_ = 0;
733 total_copy_bytes_ = 0;
734}
735
Austin Schuh2f8fd752020-09-01 22:38:28 -0700736void Logger::Rotate() {
737 for (const Node *node : log_namer_->nodes()) {
Brian Silvermand90905f2020-09-23 14:42:56 -0700738 const int node_index = configuration::GetNodeIndex(configuration_, node);
Austin Schuh64fab802020-09-09 22:47:47 -0700739 log_namer_->Rotate(node, &node_state_[node_index].log_file_header);
Austin Schuh2f8fd752020-09-01 22:38:28 -0700740 }
741}
742
743void Logger::LogUntil(monotonic_clock::time_point t) {
Austin Schuh315b96b2020-12-11 21:21:12 -0800744 // Grab the latest ServerStatistics message. This will always have the
745 // oppertunity to be >= to the current time, so it will always represent any
746 // reboots which may have happened.
Austin Schuh2f8fd752020-09-01 22:38:28 -0700747 WriteMissingTimestamps();
748
749 // Write each channel to disk, one at a time.
750 for (FetcherStruct &f : fetchers_) {
751 while (true) {
752 if (f.written) {
Brian Silvermancb805822020-10-06 17:43:35 -0700753 const auto start = event_loop_->monotonic_now();
754 const bool got_new = f.fetcher->FetchNext();
755 const auto end = event_loop_->monotonic_now();
756 RecordFetchResult(start, end, got_new, &f);
757 if (!got_new) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700758 VLOG(2) << "No new data on "
759 << configuration::CleanedChannelToString(
760 f.fetcher->channel());
761 break;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700762 }
Brian Silvermancb805822020-10-06 17:43:35 -0700763 f.written = false;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700764 }
765
Austin Schuh2f8fd752020-09-01 22:38:28 -0700766 // TODO(james): Write tests to exercise this logic.
Brian Silvermancb805822020-10-06 17:43:35 -0700767 if (f.fetcher->context().monotonic_event_time >= t) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700768 break;
769 }
Brian Silvermancb805822020-10-06 17:43:35 -0700770 if (f.writer != nullptr) {
771 // Write!
772 const auto start = event_loop_->monotonic_now();
773 flatbuffers::FlatBufferBuilder fbb(f.fetcher->context().size +
774 max_header_size_);
775 fbb.ForceDefaults(true);
776
777 fbb.FinishSizePrefixed(PackMessage(&fbb, f.fetcher->context(),
778 f.channel_index, f.log_type));
779 const auto end = event_loop_->monotonic_now();
780 RecordCreateMessageTime(start, end, &f);
781
782 VLOG(2) << "Writing data as node "
783 << FlatbufferToJson(event_loop_->node()) << " for channel "
784 << configuration::CleanedChannelToString(f.fetcher->channel())
785 << " to " << f.writer->filename() << " data "
786 << FlatbufferToJson(
787 flatbuffers::GetSizePrefixedRoot<MessageHeader>(
788 fbb.GetBufferPointer()));
789
790 max_header_size_ = std::max(max_header_size_,
791 fbb.GetSize() - f.fetcher->context().size);
Austin Schuh315b96b2020-12-11 21:21:12 -0800792 CHECK(node_state_[f.data_node_index].header_valid)
793 << ": Can't write data before the header on channel "
794 << configuration::CleanedChannelToString(f.fetcher->channel());
Brian Silvermancb805822020-10-06 17:43:35 -0700795 f.writer->QueueSizedFlatbuffer(&fbb);
796 }
797
798 if (f.timestamp_writer != nullptr) {
799 // And now handle timestamps.
800 const auto start = event_loop_->monotonic_now();
801 flatbuffers::FlatBufferBuilder fbb;
802 fbb.ForceDefaults(true);
803
804 fbb.FinishSizePrefixed(PackMessage(&fbb, f.fetcher->context(),
805 f.channel_index,
806 LogType::kLogDeliveryTimeOnly));
807 const auto end = event_loop_->monotonic_now();
808 RecordCreateMessageTime(start, end, &f);
809
810 VLOG(2) << "Writing timestamps as node "
811 << FlatbufferToJson(event_loop_->node()) << " for channel "
812 << configuration::CleanedChannelToString(f.fetcher->channel())
813 << " to " << f.timestamp_writer->filename() << " timestamp "
814 << FlatbufferToJson(
815 flatbuffers::GetSizePrefixedRoot<MessageHeader>(
816 fbb.GetBufferPointer()));
817
Austin Schuh315b96b2020-12-11 21:21:12 -0800818 CHECK(node_state_[f.timestamp_node_index].header_valid)
819 << ": Can't write data before the header on channel "
820 << configuration::CleanedChannelToString(f.fetcher->channel());
Brian Silvermancb805822020-10-06 17:43:35 -0700821 f.timestamp_writer->QueueSizedFlatbuffer(&fbb);
822 }
823
824 if (f.contents_writer != nullptr) {
825 const auto start = event_loop_->monotonic_now();
826 // And now handle the special message contents channel. Copy the
827 // message into a FlatBufferBuilder and save it to disk.
828 // TODO(austin): We can be more efficient here when we start to
829 // care...
830 flatbuffers::FlatBufferBuilder fbb;
831 fbb.ForceDefaults(true);
832
Austin Schuh0de30f32020-12-06 12:44:28 -0800833 const RemoteMessage *msg =
834 flatbuffers::GetRoot<RemoteMessage>(f.fetcher->context().data);
Brian Silvermancb805822020-10-06 17:43:35 -0700835
Austin Schuh315b96b2020-12-11 21:21:12 -0800836 CHECK(msg->has_boot_uuid()) << ": " << aos::FlatbufferToJson(msg);
837 if (!node_state_[f.contents_node_index].has_source_node_boot_uuid ||
838 node_state_[f.contents_node_index].source_node_boot_uuid !=
839 msg->boot_uuid()->string_view()) {
840 node_state_[f.contents_node_index].SetBootUUID(
841 msg->boot_uuid()->string_view());
842
843 MaybeWriteHeader(f.contents_node_index);
844 }
845
Brian Silvermancb805822020-10-06 17:43:35 -0700846 logger::MessageHeader::Builder message_header_builder(fbb);
847
848 // TODO(austin): This needs to check the channel_index and confirm
849 // that it should be logged before squirreling away the timestamp to
850 // disk. We don't want to log irrelevant timestamps.
851
852 // Note: this must match the same order as MessageBridgeServer and
853 // PackMessage. We want identical headers to have identical
854 // on-the-wire formats to make comparing them easier.
855
856 // Translate from the channel index that the event loop uses to the
857 // channel index in the log file.
858 message_header_builder.add_channel_index(
859 event_loop_to_logged_channel_index_[msg->channel_index()]);
860
861 message_header_builder.add_queue_index(msg->queue_index());
862 message_header_builder.add_monotonic_sent_time(
863 msg->monotonic_sent_time());
864 message_header_builder.add_realtime_sent_time(
865 msg->realtime_sent_time());
866
867 message_header_builder.add_monotonic_remote_time(
868 msg->monotonic_remote_time());
869 message_header_builder.add_realtime_remote_time(
870 msg->realtime_remote_time());
871 message_header_builder.add_remote_queue_index(
872 msg->remote_queue_index());
873
Austin Schuh969cd602021-01-03 00:09:45 -0800874 message_header_builder.add_monotonic_timestamp_time(
875 f.fetcher->context()
876 .monotonic_event_time.time_since_epoch()
877 .count());
878
Brian Silvermancb805822020-10-06 17:43:35 -0700879 fbb.FinishSizePrefixed(message_header_builder.Finish());
880 const auto end = event_loop_->monotonic_now();
881 RecordCreateMessageTime(start, end, &f);
882
Austin Schuh315b96b2020-12-11 21:21:12 -0800883 CHECK(node_state_[f.contents_node_index].header_valid)
884 << ": Can't write data before the header on channel "
885 << configuration::CleanedChannelToString(f.fetcher->channel());
Brian Silvermancb805822020-10-06 17:43:35 -0700886 f.contents_writer->QueueSizedFlatbuffer(&fbb);
887 }
888
889 f.written = true;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700890 }
891 }
892 last_synchronized_time_ = t;
Austin Schuhfa895892020-01-07 20:07:41 -0800893}
894
Brian Silverman1f345222020-09-24 21:14:48 -0700895void Logger::DoLogData(const monotonic_clock::time_point end_time) {
896 // We want to guarantee that messages aren't out of order by more than
Austin Schuhe309d2a2019-11-29 13:25:21 -0800897 // max_out_of_order_duration. To do this, we need sync points. Every write
898 // cycle should be a sync point.
Austin Schuhe309d2a2019-11-29 13:25:21 -0800899
900 do {
901 // Move the sync point up by at most polling_period. This forces one sync
902 // per iteration, even if it is small.
Brian Silverman1f345222020-09-24 21:14:48 -0700903 LogUntil(std::min(last_synchronized_time_ + polling_period_, end_time));
904
905 on_logged_period_();
Austin Schuhe309d2a2019-11-29 13:25:21 -0800906
Austin Schuhe309d2a2019-11-29 13:25:21 -0800907 // If we missed cycles, we could be pretty far behind. Spin until we are
908 // caught up.
Brian Silverman1f345222020-09-24 21:14:48 -0700909 } while (last_synchronized_time_ + polling_period_ < end_time);
Austin Schuhe309d2a2019-11-29 13:25:21 -0800910}
911
Brian Silvermancb805822020-10-06 17:43:35 -0700912void Logger::RecordFetchResult(aos::monotonic_clock::time_point start,
913 aos::monotonic_clock::time_point end,
914 bool got_new, FetcherStruct *fetcher) {
915 const auto duration = end - start;
916 if (!got_new) {
917 ++total_nop_fetch_count_;
918 total_nop_fetch_time_ += duration;
919 return;
920 }
921 ++total_message_fetch_count_;
922 total_message_fetch_bytes_ += fetcher->fetcher->context().size;
923 total_message_fetch_time_ += duration;
924 if (duration > max_message_fetch_time_) {
925 max_message_fetch_time_ = duration;
926 max_message_fetch_time_channel_ = fetcher->channel_index;
927 max_message_fetch_time_size_ = fetcher->fetcher->context().size;
928 }
929}
930
931void Logger::RecordCreateMessageTime(aos::monotonic_clock::time_point start,
932 aos::monotonic_clock::time_point end,
933 FetcherStruct *fetcher) {
934 const auto duration = end - start;
935 total_copy_time_ += duration;
936 ++total_copy_count_;
937 total_copy_bytes_ += fetcher->fetcher->context().size;
938 if (duration > max_copy_time_) {
939 max_copy_time_ = duration;
940 max_copy_time_channel_ = fetcher->channel_index;
941 max_copy_time_size_ = fetcher->fetcher->context().size;
942 }
943}
944
Austin Schuh11d43732020-09-21 17:28:30 -0700945std::vector<std::vector<std::string>> ToLogReaderVector(
946 const std::vector<LogFile> &log_files) {
947 std::vector<std::vector<std::string>> result;
948 for (const LogFile &log_file : log_files) {
949 for (const LogParts &log_parts : log_file.parts) {
950 std::vector<std::string> parts;
951 for (const std::string &part : log_parts.parts) {
952 parts.emplace_back(part);
953 }
954 result.emplace_back(std::move(parts));
955 }
Austin Schuh5212cad2020-09-09 23:12:09 -0700956 }
957 return result;
958}
959
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800960LogReader::LogReader(std::string_view filename,
961 const Configuration *replay_configuration)
Austin Schuh287d43d2020-12-04 20:19:33 -0800962 : LogReader(SortParts({std::string(filename)}), replay_configuration) {}
Austin Schuhfa895892020-01-07 20:07:41 -0800963
Austin Schuh287d43d2020-12-04 20:19:33 -0800964LogReader::LogReader(std::vector<LogFile> log_files,
Austin Schuhfa895892020-01-07 20:07:41 -0800965 const Configuration *replay_configuration)
Austin Schuh287d43d2020-12-04 20:19:33 -0800966 : log_files_(std::move(log_files)),
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800967 replay_configuration_(replay_configuration) {
Austin Schuh0ca51f32020-12-25 21:51:45 -0800968 CHECK_GT(log_files_.size(), 0u);
969 {
970 // Validate that we have the same config everwhere. This will be true if
971 // all the parts were sorted together and the configs match.
972 const Configuration *config = nullptr;
Austin Schuh297d2352021-01-21 19:02:17 -0800973 for (const LogFile &log_file : log_files_) {
974 if (log_file.config.get() == nullptr) {
975 LOG(FATAL) << "Couldn't find a config in " << log_file;
976 }
Austin Schuh0ca51f32020-12-25 21:51:45 -0800977 if (config == nullptr) {
978 config = log_file.config.get();
979 } else {
980 CHECK_EQ(config, log_file.config.get());
981 }
982 }
983 }
Austin Schuhdda74ec2021-01-03 19:30:37 -0800984
Austin Schuh6331ef92020-01-07 18:28:09 -0800985 MakeRemappedConfig();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800986
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700987 // Remap all existing remote timestamp channels. They will be recreated, and
988 // the data logged isn't relevant anymore.
Austin Schuh3c5dae52020-10-06 18:55:18 -0700989 for (const Node *node : configuration::GetNodes(logged_configuration())) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700990 std::vector<const Node *> timestamp_logger_nodes =
991 configuration::TimestampNodes(logged_configuration(), node);
992 for (const Node *remote_node : timestamp_logger_nodes) {
993 const std::string channel = absl::StrCat(
994 "/aos/remote_timestamps/", remote_node->name()->string_view());
Austin Schuh0de30f32020-12-06 12:44:28 -0800995 // See if the log file is an old log with MessageHeader channels in it, or
996 // a newer log with RemoteMessage. If we find an older log, rename the
997 // type too along with the name.
998 if (HasChannel<MessageHeader>(channel, node)) {
999 CHECK(!HasChannel<RemoteMessage>(channel, node))
1000 << ": Can't have both a MessageHeader and RemoteMessage remote "
1001 "timestamp channel.";
James Kuszmaul4f106fb2021-01-05 20:53:02 -08001002 // In theory, we should check NOT_LOGGED like RemoteMessage and be more
1003 // careful about updating the config, but there are fewer and fewer logs
1004 // with MessageHeader remote messages, so it isn't worth the effort.
Austin Schuh0de30f32020-12-06 12:44:28 -08001005 RemapLoggedChannel<MessageHeader>(channel, node, "/original",
1006 "aos.message_bridge.RemoteMessage");
1007 } else {
1008 CHECK(HasChannel<RemoteMessage>(channel, node))
1009 << ": Failed to find {\"name\": \"" << channel << "\", \"type\": \""
1010 << RemoteMessage::GetFullyQualifiedName() << "\"} for node "
1011 << node->name()->string_view();
James Kuszmaul4f106fb2021-01-05 20:53:02 -08001012 // Only bother to remap if there's something on the channel. We can
1013 // tell if the channel was marked NOT_LOGGED or not. This makes the
1014 // config not change un-necesarily when we replay a log with NOT_LOGGED
1015 // messages.
1016 if (HasLoggedChannel<RemoteMessage>(channel, node)) {
1017 RemapLoggedChannel<RemoteMessage>(channel, node);
1018 }
Austin Schuh0de30f32020-12-06 12:44:28 -08001019 }
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001020 }
1021 }
1022
Austin Schuh6aa77be2020-02-22 21:06:40 -08001023 if (replay_configuration) {
1024 CHECK_EQ(configuration::MultiNode(configuration()),
1025 configuration::MultiNode(replay_configuration))
Austin Schuh2f8fd752020-09-01 22:38:28 -07001026 << ": Log file and replay config need to both be multi or single "
1027 "node.";
Austin Schuh6aa77be2020-02-22 21:06:40 -08001028 }
1029
Austin Schuh6f3babe2020-01-26 20:34:50 -08001030 if (!configuration::MultiNode(configuration())) {
Austin Schuh287d43d2020-12-04 20:19:33 -08001031 states_.emplace_back(std::make_unique<State>(
1032 std::make_unique<TimestampMapper>(FilterPartsForNode(log_files_, ""))));
Austin Schuh8bd96322020-02-13 21:18:22 -08001033 } else {
Austin Schuh6aa77be2020-02-22 21:06:40 -08001034 if (replay_configuration) {
James Kuszmaul46d82582020-05-09 19:50:09 -07001035 CHECK_EQ(logged_configuration()->nodes()->size(),
Austin Schuh6aa77be2020-02-22 21:06:40 -08001036 replay_configuration->nodes()->size())
Austin Schuh2f8fd752020-09-01 22:38:28 -07001037 << ": Log file and replay config need to have matching nodes "
1038 "lists.";
James Kuszmaul46d82582020-05-09 19:50:09 -07001039 for (const Node *node : *logged_configuration()->nodes()) {
1040 if (configuration::GetNode(replay_configuration, node) == nullptr) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001041 LOG(FATAL) << "Found node " << FlatbufferToJson(node)
1042 << " in logged config that is not present in the replay "
1043 "config.";
James Kuszmaul46d82582020-05-09 19:50:09 -07001044 }
1045 }
Austin Schuh6aa77be2020-02-22 21:06:40 -08001046 }
Austin Schuh8bd96322020-02-13 21:18:22 -08001047 states_.resize(configuration()->nodes()->size());
Austin Schuh6f3babe2020-01-26 20:34:50 -08001048 }
Austin Schuhe309d2a2019-11-29 13:25:21 -08001049}
1050
Austin Schuh6aa77be2020-02-22 21:06:40 -08001051LogReader::~LogReader() {
Austin Schuh39580f12020-08-01 14:44:08 -07001052 if (event_loop_factory_unique_ptr_) {
1053 Deregister();
1054 } else if (event_loop_factory_ != nullptr) {
1055 LOG(FATAL) << "Must call Deregister before the SimulatedEventLoopFactory "
1056 "is destroyed";
1057 }
Austin Schuh2f8fd752020-09-01 22:38:28 -07001058 // Zero out some buffers. It's easy to do use-after-frees on these, so make
1059 // it more obvious.
Austin Schuh39580f12020-08-01 14:44:08 -07001060 if (remapped_configuration_buffer_) {
1061 remapped_configuration_buffer_->Wipe();
1062 }
Austin Schuh8bd96322020-02-13 21:18:22 -08001063}
Austin Schuhe309d2a2019-11-29 13:25:21 -08001064
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001065const Configuration *LogReader::logged_configuration() const {
Austin Schuh0ca51f32020-12-25 21:51:45 -08001066 return log_files_[0].config.get();
Austin Schuhe309d2a2019-11-29 13:25:21 -08001067}
1068
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001069const Configuration *LogReader::configuration() const {
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001070 return remapped_configuration_;
1071}
1072
Austin Schuh07676622021-01-21 18:59:17 -08001073std::vector<const Node *> LogReader::LoggedNodes() const {
1074 return configuration::GetNodes(logged_configuration());
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001075}
Austin Schuh15649d62019-12-28 16:36:38 -08001076
Austin Schuh11d43732020-09-21 17:28:30 -07001077monotonic_clock::time_point LogReader::monotonic_start_time(
1078 const Node *node) const {
Austin Schuh8bd96322020-02-13 21:18:22 -08001079 State *state =
1080 states_[configuration::GetNodeIndex(configuration(), node)].get();
1081 CHECK(state != nullptr) << ": Unknown node " << FlatbufferToJson(node);
1082
Austin Schuh858c9f32020-08-31 16:56:12 -07001083 return state->monotonic_start_time();
Austin Schuhe309d2a2019-11-29 13:25:21 -08001084}
1085
Austin Schuh11d43732020-09-21 17:28:30 -07001086realtime_clock::time_point LogReader::realtime_start_time(
1087 const Node *node) const {
Austin Schuh8bd96322020-02-13 21:18:22 -08001088 State *state =
1089 states_[configuration::GetNodeIndex(configuration(), node)].get();
1090 CHECK(state != nullptr) << ": Unknown node " << FlatbufferToJson(node);
1091
Austin Schuh858c9f32020-08-31 16:56:12 -07001092 return state->realtime_start_time();
Austin Schuhe309d2a2019-11-29 13:25:21 -08001093}
1094
James Kuszmaul84ff3e52020-01-03 19:48:53 -08001095void LogReader::Register() {
1096 event_loop_factory_unique_ptr_ =
Austin Schuhac0771c2020-01-07 18:36:30 -08001097 std::make_unique<SimulatedEventLoopFactory>(configuration());
James Kuszmaul84ff3e52020-01-03 19:48:53 -08001098 Register(event_loop_factory_unique_ptr_.get());
1099}
1100
Austin Schuh92547522019-12-28 14:33:43 -08001101void LogReader::Register(SimulatedEventLoopFactory *event_loop_factory) {
Austin Schuh92547522019-12-28 14:33:43 -08001102 event_loop_factory_ = event_loop_factory;
Austin Schuhe5bbd9e2020-09-21 17:29:20 -07001103 remapped_configuration_ = event_loop_factory_->configuration();
Austin Schuh0ca1fd32020-12-18 22:53:05 -08001104 filters_ =
1105 std::make_unique<message_bridge::MultiNodeNoncausalOffsetEstimator>(
Austin Schuhba20ea72021-01-21 16:47:01 -08001106 event_loop_factory_->configuration(), logged_configuration(),
Austin Schuhfe3fb342021-01-16 18:50:37 -08001107 FLAGS_skip_order_validation,
1108 chrono::duration_cast<chrono::nanoseconds>(
1109 chrono::duration<double>(FLAGS_time_estimation_buffer_seconds)));
Austin Schuh92547522019-12-28 14:33:43 -08001110
Brian Silvermand90905f2020-09-23 14:42:56 -07001111 for (const Node *node : configuration::GetNodes(configuration())) {
Austin Schuh8bd96322020-02-13 21:18:22 -08001112 const size_t node_index =
1113 configuration::GetNodeIndex(configuration(), node);
Austin Schuh287d43d2020-12-04 20:19:33 -08001114 std::vector<LogParts> filtered_parts = FilterPartsForNode(
1115 log_files_, node != nullptr ? node->name()->string_view() : "");
Austin Schuh315b96b2020-12-11 21:21:12 -08001116
1117 // Confirm that all the parts are from the same boot if there are enough
1118 // parts to not be from the same boot.
1119 if (filtered_parts.size() > 1u) {
1120 for (size_t i = 1; i < filtered_parts.size(); ++i) {
1121 CHECK_EQ(filtered_parts[i].source_boot_uuid,
1122 filtered_parts[0].source_boot_uuid)
1123 << ": Found parts from different boots "
1124 << LogFileVectorToString(log_files_);
1125 }
James Kuszmaul4f106fb2021-01-05 20:53:02 -08001126 if (!filtered_parts[0].source_boot_uuid.empty()) {
1127 event_loop_factory_->GetNodeEventLoopFactory(node)->set_boot_uuid(
1128 filtered_parts[0].source_boot_uuid);
1129 }
Austin Schuh315b96b2020-12-11 21:21:12 -08001130 }
1131
Austin Schuh287d43d2020-12-04 20:19:33 -08001132 states_[node_index] = std::make_unique<State>(
1133 filtered_parts.size() == 0u
1134 ? nullptr
1135 : std::make_unique<TimestampMapper>(std::move(filtered_parts)));
Austin Schuh8bd96322020-02-13 21:18:22 -08001136 State *state = states_[node_index].get();
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001137 state->set_event_loop(state->SetNodeEventLoopFactory(
Austin Schuh858c9f32020-08-31 16:56:12 -07001138 event_loop_factory_->GetNodeEventLoopFactory(node)));
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001139
1140 state->SetChannelCount(logged_configuration()->channels()->size());
Austin Schuhcde938c2020-02-02 17:30:07 -08001141 }
Austin Schuh87dd3832021-01-01 23:07:31 -08001142 event_loop_factory_->SetTimeConverter(filters_.get());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001143
Austin Schuh287d43d2020-12-04 20:19:33 -08001144 for (const Node *node : configuration::GetNodes(configuration())) {
1145 const size_t node_index =
1146 configuration::GetNodeIndex(configuration(), node);
1147 State *state = states_[node_index].get();
1148 for (const Node *other_node : configuration::GetNodes(configuration())) {
1149 const size_t other_node_index =
1150 configuration::GetNodeIndex(configuration(), other_node);
1151 State *other_state = states_[other_node_index].get();
1152 if (other_state != state) {
1153 state->AddPeer(other_state);
1154 }
1155 }
1156 }
1157
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001158 // Register after making all the State objects so we can build references
1159 // between them.
1160 for (const Node *node : configuration::GetNodes(configuration())) {
1161 const size_t node_index =
1162 configuration::GetNodeIndex(configuration(), node);
1163 State *state = states_[node_index].get();
1164
1165 Register(state->event_loop());
1166 }
1167
James Kuszmaul46d82582020-05-09 19:50:09 -07001168 if (live_nodes_ == 0) {
1169 LOG(FATAL)
1170 << "Don't have logs from any of the nodes in the replay config--are "
1171 "you sure that the replay config matches the original config?";
1172 }
Austin Schuh6f3babe2020-01-26 20:34:50 -08001173
Austin Schuh87dd3832021-01-01 23:07:31 -08001174 filters_->CheckGraph();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001175
Austin Schuh858c9f32020-08-31 16:56:12 -07001176 for (std::unique_ptr<State> &state : states_) {
1177 state->SeedSortedMessages();
1178 }
1179
Austin Schuh2f8fd752020-09-01 22:38:28 -07001180 // We want to start the log file at the last start time of the log files
1181 // from all the nodes. Compute how long each node's simulation needs to run
1182 // to move time to this point.
Austin Schuh8bd96322020-02-13 21:18:22 -08001183 distributed_clock::time_point start_time = distributed_clock::min_time;
Austin Schuhcde938c2020-02-02 17:30:07 -08001184
Austin Schuh2f8fd752020-09-01 22:38:28 -07001185 // TODO(austin): We want an "OnStart" callback for each node rather than
1186 // running until the last node.
1187
Austin Schuh8bd96322020-02-13 21:18:22 -08001188 for (std::unique_ptr<State> &state : states_) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001189 VLOG(1) << "Start time is " << state->monotonic_start_time() << " for node "
1190 << MaybeNodeName(state->event_loop()->node()) << "now "
1191 << state->monotonic_now();
Austin Schuh287d43d2020-12-04 20:19:33 -08001192 if (state->monotonic_start_time() == monotonic_clock::min_time) {
1193 continue;
1194 }
Austin Schuh2f8fd752020-09-01 22:38:28 -07001195 // And start computing the start time on the distributed clock now that
1196 // that works.
Austin Schuh858c9f32020-08-31 16:56:12 -07001197 start_time = std::max(
1198 start_time, state->ToDistributedClock(state->monotonic_start_time()));
Austin Schuhcde938c2020-02-02 17:30:07 -08001199 }
Austin Schuh2f8fd752020-09-01 22:38:28 -07001200
Austin Schuh87dd3832021-01-01 23:07:31 -08001201 // TODO(austin): If a node doesn't have a start time, we might not queue
1202 // enough. If this happens, we'll explode with a frozen error eventually.
1203
Austin Schuh2f8fd752020-09-01 22:38:28 -07001204 CHECK_GE(start_time, distributed_clock::epoch())
1205 << ": Hmm, we have a node starting before the start of time. Offset "
1206 "everything.";
Austin Schuhcde938c2020-02-02 17:30:07 -08001207
Austin Schuh6f3babe2020-01-26 20:34:50 -08001208 // Forwarding is tracked per channel. If it is enabled, we want to turn it
1209 // off. Otherwise messages replayed will get forwarded across to the other
Austin Schuh2f8fd752020-09-01 22:38:28 -07001210 // nodes, and also replayed on the other nodes. This may not satisfy all
1211 // our users, but it'll start the discussion.
Austin Schuh6f3babe2020-01-26 20:34:50 -08001212 if (configuration::MultiNode(event_loop_factory_->configuration())) {
1213 for (size_t i = 0; i < logged_configuration()->channels()->size(); ++i) {
1214 const Channel *channel = logged_configuration()->channels()->Get(i);
1215 const Node *node = configuration::GetNode(
1216 configuration(), channel->source_node()->string_view());
1217
Austin Schuh8bd96322020-02-13 21:18:22 -08001218 State *state =
1219 states_[configuration::GetNodeIndex(configuration(), node)].get();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001220
1221 const Channel *remapped_channel =
Austin Schuh858c9f32020-08-31 16:56:12 -07001222 RemapChannel(state->event_loop(), channel);
Austin Schuh6f3babe2020-01-26 20:34:50 -08001223
1224 event_loop_factory_->DisableForwarding(remapped_channel);
1225 }
Austin Schuh4c3b9702020-08-30 11:34:55 -07001226
1227 // If we are replaying a log, we don't want a bunch of redundant messages
1228 // from both the real message bridge and simulated message bridge.
1229 event_loop_factory_->DisableStatistics();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001230 }
1231
Austin Schuhcde938c2020-02-02 17:30:07 -08001232 // While we are starting the system up, we might be relying on matching data
1233 // to timestamps on log files where the timestamp log file starts before the
1234 // data. In this case, it is reasonable to expect missing data.
Austin Schuhdda74ec2021-01-03 19:30:37 -08001235 {
1236 const bool prior_ignore_missing_data = ignore_missing_data_;
1237 ignore_missing_data_ = true;
1238 VLOG(1) << "Running until " << start_time << " in Register";
1239 event_loop_factory_->RunFor(start_time.time_since_epoch());
1240 VLOG(1) << "At start time";
1241 // Now that we are running for real, missing data means that the log file is
1242 // corrupted or went wrong.
1243 ignore_missing_data_ = prior_ignore_missing_data;
1244 }
Austin Schuh92547522019-12-28 14:33:43 -08001245
Austin Schuh8bd96322020-02-13 21:18:22 -08001246 for (std::unique_ptr<State> &state : states_) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001247 // Make the RT clock be correct before handing it to the user.
1248 if (state->realtime_start_time() != realtime_clock::min_time) {
1249 state->SetRealtimeOffset(state->monotonic_start_time(),
1250 state->realtime_start_time());
1251 }
1252 VLOG(1) << "Start time is " << state->monotonic_start_time() << " for node "
1253 << MaybeNodeName(state->event_loop()->node()) << "now "
1254 << state->monotonic_now();
1255 }
1256
1257 if (FLAGS_timestamps_to_csv) {
Austin Schuh0ca1fd32020-12-18 22:53:05 -08001258 filters_->Start(event_loop_factory);
Austin Schuh8bd96322020-02-13 21:18:22 -08001259 }
1260}
1261
Austin Schuh2f8fd752020-09-01 22:38:28 -07001262message_bridge::NoncausalOffsetEstimator *LogReader::GetFilter(
Austin Schuh8bd96322020-02-13 21:18:22 -08001263 const Node *node_a, const Node *node_b) {
Austin Schuh0ca1fd32020-12-18 22:53:05 -08001264 if (filters_) {
1265 return filters_->GetFilter(node_a, node_b);
Austin Schuh8bd96322020-02-13 21:18:22 -08001266 }
Austin Schuh0ca1fd32020-12-18 22:53:05 -08001267 return nullptr;
Austin Schuh8bd96322020-02-13 21:18:22 -08001268}
1269
Austin Schuhe309d2a2019-11-29 13:25:21 -08001270void LogReader::Register(EventLoop *event_loop) {
Austin Schuh8bd96322020-02-13 21:18:22 -08001271 State *state =
1272 states_[configuration::GetNodeIndex(configuration(), event_loop->node())]
1273 .get();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001274
Austin Schuh858c9f32020-08-31 16:56:12 -07001275 state->set_event_loop(event_loop);
Austin Schuhe309d2a2019-11-29 13:25:21 -08001276
Tyler Chatow67ddb032020-01-12 14:30:04 -08001277 // We don't run timing reports when trying to print out logged data, because
1278 // otherwise we would end up printing out the timing reports themselves...
1279 // This is only really relevant when we are replaying into a simulation.
Austin Schuh6f3babe2020-01-26 20:34:50 -08001280 event_loop->SkipTimingReport();
1281 event_loop->SkipAosLog();
Austin Schuh39788ff2019-12-01 18:22:57 -08001282
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001283 for (size_t logged_channel_index = 0;
1284 logged_channel_index < logged_configuration()->channels()->size();
1285 ++logged_channel_index) {
1286 const Channel *channel = RemapChannel(
1287 event_loop,
1288 logged_configuration()->channels()->Get(logged_channel_index));
Austin Schuh8bd96322020-02-13 21:18:22 -08001289
Austin Schuh532656d2021-01-11 10:17:18 -08001290 if (channel->logger() == LoggerConfig::NOT_LOGGED) {
1291 continue;
1292 }
1293
Austin Schuh2f8fd752020-09-01 22:38:28 -07001294 message_bridge::NoncausalOffsetEstimator *filter = nullptr;
Austin Schuh969cd602021-01-03 00:09:45 -08001295 RemoteMessageSender *remote_timestamp_sender = nullptr;
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001296
1297 State *source_state = nullptr;
Austin Schuh8bd96322020-02-13 21:18:22 -08001298
1299 if (!configuration::ChannelIsSendableOnNode(channel, event_loop->node()) &&
1300 configuration::ChannelIsReadableOnNode(channel, event_loop->node())) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001301 // We've got a message which is being forwarded to this node.
1302 const Node *source_node = configuration::GetNode(
Austin Schuh8bd96322020-02-13 21:18:22 -08001303 event_loop->configuration(), channel->source_node()->string_view());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001304 filter = GetFilter(event_loop->node(), source_node);
Austin Schuh8bd96322020-02-13 21:18:22 -08001305
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001306 // Delivery timestamps are supposed to be logged back on the source node.
1307 // Configure remote timestamps to be sent.
1308 const bool delivery_time_is_logged =
1309 configuration::ConnectionDeliveryTimeIsLoggedOnNode(
1310 channel, event_loop->node(), source_node);
1311
1312 source_state =
1313 states_[configuration::GetNodeIndex(configuration(), source_node)]
1314 .get();
1315
1316 if (delivery_time_is_logged) {
1317 remote_timestamp_sender =
1318 source_state->RemoteTimestampSender(event_loop->node());
Austin Schuh8bd96322020-02-13 21:18:22 -08001319 }
1320 }
Austin Schuh858c9f32020-08-31 16:56:12 -07001321
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001322 state->SetChannel(
1323 logged_channel_index,
1324 configuration::ChannelIndex(event_loop->configuration(), channel),
1325 event_loop->MakeRawSender(channel), filter, remote_timestamp_sender,
1326 source_state);
Austin Schuhe309d2a2019-11-29 13:25:21 -08001327 }
1328
Austin Schuh6aa77be2020-02-22 21:06:40 -08001329 // If we didn't find any log files with data in them, we won't ever get a
1330 // callback or be live. So skip the rest of the setup.
Austin Schuh287d43d2020-12-04 20:19:33 -08001331 if (state->OldestMessageTime() == monotonic_clock::max_time) {
Austin Schuh6aa77be2020-02-22 21:06:40 -08001332 return;
1333 }
1334
Austin Schuh858c9f32020-08-31 16:56:12 -07001335 state->set_timer_handler(event_loop->AddTimer([this, state]() {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001336 VLOG(1) << "Starting sending " << MaybeNodeName(state->event_loop()->node())
1337 << "at " << state->event_loop()->context().monotonic_event_time
1338 << " now " << state->monotonic_now();
Austin Schuh858c9f32020-08-31 16:56:12 -07001339 if (state->OldestMessageTime() == monotonic_clock::max_time) {
Austin Schuh6f3babe2020-01-26 20:34:50 -08001340 --live_nodes_;
Austin Schuh2f8fd752020-09-01 22:38:28 -07001341 VLOG(1) << MaybeNodeName(state->event_loop()->node()) << "Node down!";
James Kuszmaul71a81932020-12-15 21:08:01 -08001342 if (exit_on_finish_ && live_nodes_ == 0) {
Austin Schuh6f3babe2020-01-26 20:34:50 -08001343 event_loop_factory_->Exit();
1344 }
James Kuszmaul314f1672020-01-03 20:02:08 -08001345 return;
1346 }
Austin Schuh2f8fd752020-09-01 22:38:28 -07001347
Austin Schuhdda74ec2021-01-03 19:30:37 -08001348 TimestampedMessage timestamped_message = state->PopOldest();
Austin Schuh05b70472020-01-01 17:11:17 -08001349
Austin Schuhe309d2a2019-11-29 13:25:21 -08001350 const monotonic_clock::time_point monotonic_now =
Austin Schuh858c9f32020-08-31 16:56:12 -07001351 state->event_loop()->context().monotonic_event_time;
Austin Schuh2f8fd752020-09-01 22:38:28 -07001352 if (!FLAGS_skip_order_validation) {
Austin Schuh287d43d2020-12-04 20:19:33 -08001353 CHECK(monotonic_now == timestamped_message.monotonic_event_time)
Austin Schuh2f8fd752020-09-01 22:38:28 -07001354 << ": " << FlatbufferToJson(state->event_loop()->node()) << " Now "
1355 << monotonic_now << " trying to send "
Austin Schuh287d43d2020-12-04 20:19:33 -08001356 << timestamped_message.monotonic_event_time << " failure "
Austin Schuh2f8fd752020-09-01 22:38:28 -07001357 << state->DebugString();
Austin Schuh287d43d2020-12-04 20:19:33 -08001358 } else if (monotonic_now != timestamped_message.monotonic_event_time) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001359 LOG(WARNING) << "Check failed: monotonic_now == "
Austin Schuh287d43d2020-12-04 20:19:33 -08001360 "timestamped_message.monotonic_event_time) ("
Austin Schuh2f8fd752020-09-01 22:38:28 -07001361 << monotonic_now << " vs. "
Austin Schuh287d43d2020-12-04 20:19:33 -08001362 << timestamped_message.monotonic_event_time
Austin Schuh2f8fd752020-09-01 22:38:28 -07001363 << "): " << FlatbufferToJson(state->event_loop()->node())
1364 << " Now " << monotonic_now << " trying to send "
Austin Schuh287d43d2020-12-04 20:19:33 -08001365 << timestamped_message.monotonic_event_time << " failure "
Austin Schuh2f8fd752020-09-01 22:38:28 -07001366 << state->DebugString();
1367 }
Austin Schuhe309d2a2019-11-29 13:25:21 -08001368
Austin Schuh287d43d2020-12-04 20:19:33 -08001369 if (timestamped_message.monotonic_event_time >
Austin Schuh858c9f32020-08-31 16:56:12 -07001370 state->monotonic_start_time() ||
Austin Schuh15649d62019-12-28 16:36:38 -08001371 event_loop_factory_ != nullptr) {
Austin Schuhdda74ec2021-01-03 19:30:37 -08001372 if (timestamped_message.data.span().size() != 0u) {
1373 if (timestamped_message.monotonic_remote_time !=
1374 monotonic_clock::min_time) {
Austin Schuh8bd96322020-02-13 21:18:22 -08001375 // Confirm that the message was sent on the sending node before the
1376 // destination node (this node). As a proxy, do this by making sure
1377 // that time on the source node is past when the message was sent.
Austin Schuh87dd3832021-01-01 23:07:31 -08001378 //
1379 // TODO(austin): <= means that the cause message (which we know) could
1380 // happen after the effect even though we know they are at the same
1381 // time. I doubt anyone will notice for a bit, but we should really
1382 // fix that.
Austin Schuh2f8fd752020-09-01 22:38:28 -07001383 if (!FLAGS_skip_order_validation) {
Austin Schuh87dd3832021-01-01 23:07:31 -08001384 CHECK_LE(
Austin Schuh287d43d2020-12-04 20:19:33 -08001385 timestamped_message.monotonic_remote_time,
1386 state->monotonic_remote_now(timestamped_message.channel_index))
Austin Schuh2f8fd752020-09-01 22:38:28 -07001387 << state->event_loop()->node()->name()->string_view() << " to "
Austin Schuh287d43d2020-12-04 20:19:33 -08001388 << state->remote_node(timestamped_message.channel_index)
1389 ->name()
1390 ->string_view()
Austin Schuh315b96b2020-12-11 21:21:12 -08001391 << " while trying to send a message on "
1392 << configuration::CleanedChannelToString(
1393 logged_configuration()->channels()->Get(
1394 timestamped_message.channel_index))
Austin Schuh2f8fd752020-09-01 22:38:28 -07001395 << " " << state->DebugString();
Austin Schuh87dd3832021-01-01 23:07:31 -08001396 } else if (timestamped_message.monotonic_remote_time >
Austin Schuh287d43d2020-12-04 20:19:33 -08001397 state->monotonic_remote_now(
1398 timestamped_message.channel_index)) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001399 LOG(WARNING)
Austin Schuh287d43d2020-12-04 20:19:33 -08001400 << "Check failed: timestamped_message.monotonic_remote_time < "
1401 "state->monotonic_remote_now(timestamped_message.channel_"
1402 "index) ("
1403 << timestamped_message.monotonic_remote_time << " vs. "
1404 << state->monotonic_remote_now(
1405 timestamped_message.channel_index)
1406 << ") " << state->event_loop()->node()->name()->string_view()
1407 << " to "
1408 << state->remote_node(timestamped_message.channel_index)
1409 ->name()
1410 ->string_view()
1411 << " currently " << timestamped_message.monotonic_event_time
Austin Schuh2f8fd752020-09-01 22:38:28 -07001412 << " ("
1413 << state->ToDistributedClock(
Austin Schuh287d43d2020-12-04 20:19:33 -08001414 timestamped_message.monotonic_event_time)
Austin Schuh2f8fd752020-09-01 22:38:28 -07001415 << ") remote event time "
Austin Schuh287d43d2020-12-04 20:19:33 -08001416 << timestamped_message.monotonic_remote_time << " ("
Austin Schuh2f8fd752020-09-01 22:38:28 -07001417 << state->RemoteToDistributedClock(
Austin Schuh287d43d2020-12-04 20:19:33 -08001418 timestamped_message.channel_index,
1419 timestamped_message.monotonic_remote_time)
Austin Schuh2f8fd752020-09-01 22:38:28 -07001420 << ") " << state->DebugString();
1421 }
Austin Schuh8bd96322020-02-13 21:18:22 -08001422 }
1423
Austin Schuh15649d62019-12-28 16:36:38 -08001424 // If we have access to the factory, use it to fix the realtime time.
Austin Schuh287d43d2020-12-04 20:19:33 -08001425 state->SetRealtimeOffset(timestamped_message.monotonic_event_time,
1426 timestamped_message.realtime_event_time);
Austin Schuh15649d62019-12-28 16:36:38 -08001427
Austin Schuh2f8fd752020-09-01 22:38:28 -07001428 VLOG(1) << MaybeNodeName(state->event_loop()->node()) << "Sending "
Austin Schuh287d43d2020-12-04 20:19:33 -08001429 << timestamped_message.monotonic_event_time;
Austin Schuh2f8fd752020-09-01 22:38:28 -07001430 // TODO(austin): std::move channel_data in and make that efficient in
1431 // simulation.
Austin Schuh287d43d2020-12-04 20:19:33 -08001432 state->Send(std::move(timestamped_message));
Austin Schuhdda74ec2021-01-03 19:30:37 -08001433 } else if (!ignore_missing_data_ &&
1434 !FLAGS_skip_missing_forwarding_entries) {
1435 // We've found a timestamp without data. This likely means that we are
1436 // at the end of the log file. Record it and CHECK that in the rest of
1437 // the log file, we don't find any more data on that channel. Not all
1438 // channels will end at the same point in time since they can be in
1439 // different files.
1440 VLOG(1) << "Found the last message on channel "
1441 << timestamped_message.channel_index;
1442
1443 // Vector storing if we've seen a nullptr message or not per channel.
1444 std::vector<bool> last_message;
1445 last_message.resize(logged_configuration()->channels()->size(), false);
1446
1447 last_message[timestamped_message.channel_index] = true;
1448
1449 // Now that we found the end of one channel, artificially stop the
1450 // rest. It is confusing when part of your data gets replayed but not
1451 // all.
Austin Schuh858c9f32020-08-31 16:56:12 -07001452 while (state->OldestMessageTime() != monotonic_clock::max_time) {
Austin Schuhdda74ec2021-01-03 19:30:37 -08001453 TimestampedMessage next = state->PopOldest();
1454 // Make sure that once we have seen the last message on a channel,
1455 // data doesn't start back up again. If the user wants to play
1456 // through events like this, they can set
1457 // --skip_missing_forwarding_entries or ignore_missing_data_.
1458 CHECK_LT(next.channel_index, last_message.size());
1459 if (next.data.span().size() == 0u) {
1460 last_message[next.channel_index] = true;
1461 } else {
1462 if (last_message[next.channel_index]) {
1463 LOG(FATAL)
1464 << "Found missing data in the middle of the log file on "
1465 "channel "
1466 << next.channel_index << " Last "
1467 << last_message[next.channel_index] << state->DebugString();
1468 }
1469 }
Austin Schuh8bd96322020-02-13 21:18:22 -08001470 }
Austin Schuh92547522019-12-28 14:33:43 -08001471 }
Austin Schuhe309d2a2019-11-29 13:25:21 -08001472 } else {
Austin Schuh6f3babe2020-01-26 20:34:50 -08001473 LOG(WARNING)
1474 << "Not sending data from before the start of the log file. "
Austin Schuh287d43d2020-12-04 20:19:33 -08001475 << timestamped_message.monotonic_event_time.time_since_epoch().count()
Austin Schuh6f3babe2020-01-26 20:34:50 -08001476 << " start " << monotonic_start_time().time_since_epoch().count()
Austin Schuhd85baf82020-10-19 11:50:12 -07001477 << " "
Austin Schuh287d43d2020-12-04 20:19:33 -08001478 << FlatbufferToJson(timestamped_message.data,
Austin Schuhd85baf82020-10-19 11:50:12 -07001479 {.multi_line = false, .max_vector_size = 100});
Austin Schuhe309d2a2019-11-29 13:25:21 -08001480 }
1481
Austin Schuh858c9f32020-08-31 16:56:12 -07001482 const monotonic_clock::time_point next_time = state->OldestMessageTime();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001483 if (next_time != monotonic_clock::max_time) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001484 VLOG(1) << "Scheduling " << MaybeNodeName(state->event_loop()->node())
1485 << "wakeup for " << next_time << "("
1486 << state->ToDistributedClock(next_time)
1487 << " distributed), now is " << state->monotonic_now();
Austin Schuh858c9f32020-08-31 16:56:12 -07001488 state->Setup(next_time);
James Kuszmaul314f1672020-01-03 20:02:08 -08001489 } else {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001490 VLOG(1) << MaybeNodeName(state->event_loop()->node())
1491 << "No next message, scheduling shutdown";
1492 // Set a timer up immediately after now to die. If we don't do this,
1493 // then the senders waiting on the message we just read will never get
1494 // called.
Austin Schuheecb9282020-01-08 17:43:30 -08001495 if (event_loop_factory_ != nullptr) {
Austin Schuh858c9f32020-08-31 16:56:12 -07001496 state->Setup(monotonic_now + event_loop_factory_->send_delay() +
1497 std::chrono::nanoseconds(1));
Austin Schuheecb9282020-01-08 17:43:30 -08001498 }
Austin Schuhe309d2a2019-11-29 13:25:21 -08001499 }
Austin Schuh8bd96322020-02-13 21:18:22 -08001500
Austin Schuh2f8fd752020-09-01 22:38:28 -07001501 VLOG(1) << MaybeNodeName(state->event_loop()->node()) << "Done sending at "
1502 << state->event_loop()->context().monotonic_event_time << " now "
1503 << state->monotonic_now();
Austin Schuh858c9f32020-08-31 16:56:12 -07001504 }));
Austin Schuhe309d2a2019-11-29 13:25:21 -08001505
Austin Schuh6f3babe2020-01-26 20:34:50 -08001506 ++live_nodes_;
1507
Austin Schuh858c9f32020-08-31 16:56:12 -07001508 if (state->OldestMessageTime() != monotonic_clock::max_time) {
1509 event_loop->OnRun([state]() { state->Setup(state->OldestMessageTime()); });
Austin Schuhe309d2a2019-11-29 13:25:21 -08001510 }
1511}
1512
1513void LogReader::Deregister() {
James Kuszmaul84ff3e52020-01-03 19:48:53 -08001514 // Make sure that things get destroyed in the correct order, rather than
1515 // relying on getting the order correct in the class definition.
Austin Schuh8bd96322020-02-13 21:18:22 -08001516 for (std::unique_ptr<State> &state : states_) {
Austin Schuh858c9f32020-08-31 16:56:12 -07001517 state->Deregister();
Austin Schuhe309d2a2019-11-29 13:25:21 -08001518 }
Austin Schuh92547522019-12-28 14:33:43 -08001519
James Kuszmaul84ff3e52020-01-03 19:48:53 -08001520 event_loop_factory_unique_ptr_.reset();
1521 event_loop_factory_ = nullptr;
Austin Schuhe309d2a2019-11-29 13:25:21 -08001522}
1523
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001524void LogReader::RemapLoggedChannel(std::string_view name, std::string_view type,
Austin Schuh0de30f32020-12-06 12:44:28 -08001525 std::string_view add_prefix,
1526 std::string_view new_type) {
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001527 for (size_t ii = 0; ii < logged_configuration()->channels()->size(); ++ii) {
1528 const Channel *const channel = logged_configuration()->channels()->Get(ii);
1529 if (channel->name()->str() == name &&
1530 channel->type()->string_view() == type) {
1531 CHECK_EQ(0u, remapped_channels_.count(ii))
1532 << "Already remapped channel "
1533 << configuration::CleanedChannelToString(channel);
Austin Schuh0de30f32020-12-06 12:44:28 -08001534 RemappedChannel remapped_channel;
1535 remapped_channel.remapped_name =
1536 std::string(add_prefix) + std::string(name);
1537 remapped_channel.new_type = new_type;
1538 remapped_channels_[ii] = std::move(remapped_channel);
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001539 VLOG(1) << "Remapping channel "
1540 << configuration::CleanedChannelToString(channel)
Austin Schuh0de30f32020-12-06 12:44:28 -08001541 << " to have name " << remapped_channels_[ii].remapped_name;
Austin Schuh6331ef92020-01-07 18:28:09 -08001542 MakeRemappedConfig();
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001543 return;
1544 }
1545 }
1546 LOG(FATAL) << "Unabled to locate channel with name " << name << " and type "
1547 << type;
1548}
1549
Austin Schuh01b4c352020-09-21 23:09:39 -07001550void LogReader::RemapLoggedChannel(std::string_view name, std::string_view type,
1551 const Node *node,
Austin Schuh0de30f32020-12-06 12:44:28 -08001552 std::string_view add_prefix,
1553 std::string_view new_type) {
Austin Schuh01b4c352020-09-21 23:09:39 -07001554 VLOG(1) << "Node is " << aos::FlatbufferToJson(node);
1555 const Channel *remapped_channel =
1556 configuration::GetChannel(logged_configuration(), name, type, "", node);
1557 CHECK(remapped_channel != nullptr) << ": Failed to find {\"name\": \"" << name
1558 << "\", \"type\": \"" << type << "\"}";
1559 VLOG(1) << "Original {\"name\": \"" << name << "\", \"type\": \"" << type
1560 << "\"}";
1561 VLOG(1) << "Remapped "
1562 << aos::configuration::StrippedChannelToString(remapped_channel);
1563
1564 // We want to make /spray on node 0 go to /0/spray by snooping the maps. And
1565 // we want it to degrade if the heuristics fail to just work.
1566 //
1567 // The easiest way to do this is going to be incredibly specific and verbose.
1568 // Look up /spray, to /0/spray. Then, prefix the result with /original to get
1569 // /original/0/spray. Then, create a map from /original/spray to
1570 // /original/0/spray for just the type we were asked for.
1571 if (name != remapped_channel->name()->string_view()) {
1572 MapT new_map;
1573 new_map.match = std::make_unique<ChannelT>();
1574 new_map.match->name = absl::StrCat(add_prefix, name);
1575 new_map.match->type = type;
1576 if (node != nullptr) {
1577 new_map.match->source_node = node->name()->str();
1578 }
1579 new_map.rename = std::make_unique<ChannelT>();
1580 new_map.rename->name =
1581 absl::StrCat(add_prefix, remapped_channel->name()->string_view());
1582 maps_.emplace_back(std::move(new_map));
1583 }
1584
1585 const size_t channel_index =
1586 configuration::ChannelIndex(logged_configuration(), remapped_channel);
1587 CHECK_EQ(0u, remapped_channels_.count(channel_index))
1588 << "Already remapped channel "
1589 << configuration::CleanedChannelToString(remapped_channel);
Austin Schuh0de30f32020-12-06 12:44:28 -08001590
1591 RemappedChannel remapped_channel_struct;
1592 remapped_channel_struct.remapped_name =
1593 std::string(add_prefix) +
1594 std::string(remapped_channel->name()->string_view());
1595 remapped_channel_struct.new_type = new_type;
1596 remapped_channels_[channel_index] = std::move(remapped_channel_struct);
Austin Schuh01b4c352020-09-21 23:09:39 -07001597 MakeRemappedConfig();
1598}
1599
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001600void LogReader::MakeRemappedConfig() {
Austin Schuh8bd96322020-02-13 21:18:22 -08001601 for (std::unique_ptr<State> &state : states_) {
Austin Schuh6aa77be2020-02-22 21:06:40 -08001602 if (state) {
Austin Schuh858c9f32020-08-31 16:56:12 -07001603 CHECK(!state->event_loop())
Austin Schuh6aa77be2020-02-22 21:06:40 -08001604 << ": Can't change the mapping after the events are scheduled.";
1605 }
Austin Schuh6f3babe2020-01-26 20:34:50 -08001606 }
Austin Schuhac0771c2020-01-07 18:36:30 -08001607
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001608 // If no remapping occurred and we are using the original config, then there
1609 // is nothing interesting to do here.
1610 if (remapped_channels_.empty() && replay_configuration_ == nullptr) {
Austin Schuh6f3babe2020-01-26 20:34:50 -08001611 remapped_configuration_ = logged_configuration();
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001612 return;
1613 }
1614 // Config to copy Channel definitions from. Use the specified
1615 // replay_configuration_ if it has been provided.
1616 const Configuration *const base_config = replay_configuration_ == nullptr
1617 ? logged_configuration()
1618 : replay_configuration_;
Austin Schuh0de30f32020-12-06 12:44:28 -08001619
1620 // Create a config with all the channels, but un-sorted/merged. Collect up
1621 // the schemas while we do this. Call MergeConfiguration to sort everything,
1622 // and then merge it all in together.
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001623
1624 // This is the builder that we use for the config containing all the new
1625 // channels.
Austin Schuh0de30f32020-12-06 12:44:28 -08001626 flatbuffers::FlatBufferBuilder fbb;
1627 fbb.ForceDefaults(true);
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001628 std::vector<flatbuffers::Offset<Channel>> channel_offsets;
Austin Schuh0de30f32020-12-06 12:44:28 -08001629
1630 CHECK_EQ(Channel::MiniReflectTypeTable()->num_elems, 13u)
1631 << ": Merging logic needs to be updated when the number of channel "
1632 "fields changes.";
1633
1634 // List of schemas.
1635 std::map<std::string_view, FlatbufferVector<reflection::Schema>> schema_map;
1636 // Make sure our new RemoteMessage schema is in there for old logs without it.
1637 schema_map.insert(std::make_pair(
1638 RemoteMessage::GetFullyQualifiedName(),
1639 FlatbufferVector<reflection::Schema>(FlatbufferSpan<reflection::Schema>(
1640 message_bridge::RemoteMessageSchema()))));
1641
1642 // Reconstruct the remapped channels.
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001643 for (auto &pair : remapped_channels_) {
Austin Schuh0de30f32020-12-06 12:44:28 -08001644 const Channel *const c = CHECK_NOTNULL(configuration::GetChannel(
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001645 base_config, logged_configuration()->channels()->Get(pair.first), "",
1646 nullptr));
Austin Schuh0de30f32020-12-06 12:44:28 -08001647 channel_offsets.emplace_back(
1648 CopyChannel(c, pair.second.remapped_name, "", &fbb));
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001649 }
Austin Schuh01b4c352020-09-21 23:09:39 -07001650
Austin Schuh0de30f32020-12-06 12:44:28 -08001651 // Now reconstruct the original channels, translating types as needed
1652 for (const Channel *c : *base_config->channels()) {
1653 // Search for a mapping channel.
1654 std::string_view new_type = "";
1655 for (auto &pair : remapped_channels_) {
1656 const Channel *const remapped_channel =
1657 logged_configuration()->channels()->Get(pair.first);
1658 if (remapped_channel->name()->string_view() == c->name()->string_view() &&
1659 remapped_channel->type()->string_view() == c->type()->string_view()) {
1660 new_type = pair.second.new_type;
1661 break;
1662 }
1663 }
1664
1665 // Copy everything over.
1666 channel_offsets.emplace_back(CopyChannel(c, "", new_type, &fbb));
1667
1668 // Add the schema if it doesn't exist.
1669 if (schema_map.find(c->type()->string_view()) == schema_map.end()) {
1670 CHECK(c->has_schema());
1671 schema_map.insert(std::make_pair(c->type()->string_view(),
1672 RecursiveCopyFlatBuffer(c->schema())));
1673 }
1674 }
1675
1676 // The MergeConfiguration API takes a vector, not a map. Convert.
1677 std::vector<FlatbufferVector<reflection::Schema>> schemas;
1678 while (!schema_map.empty()) {
1679 schemas.emplace_back(std::move(schema_map.begin()->second));
1680 schema_map.erase(schema_map.begin());
1681 }
1682
1683 // Create the Configuration containing the new channels that we want to add.
1684 const flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Channel>>>
1685 channels_offset =
1686 channel_offsets.empty() ? 0 : fbb.CreateVector(channel_offsets);
1687
1688 // Copy over the old maps.
Austin Schuh01b4c352020-09-21 23:09:39 -07001689 std::vector<flatbuffers::Offset<Map>> map_offsets;
Austin Schuh0de30f32020-12-06 12:44:28 -08001690 if (base_config->maps()) {
1691 for (const Map *map : *base_config->maps()) {
1692 map_offsets.emplace_back(aos::RecursiveCopyFlatBuffer(map, &fbb));
1693 }
1694 }
1695
1696 // Now create the new maps. These are second so they take effect first.
Austin Schuh01b4c352020-09-21 23:09:39 -07001697 for (const MapT &map : maps_) {
1698 const flatbuffers::Offset<flatbuffers::String> match_name_offset =
Austin Schuh0de30f32020-12-06 12:44:28 -08001699 fbb.CreateString(map.match->name);
Austin Schuh01b4c352020-09-21 23:09:39 -07001700 const flatbuffers::Offset<flatbuffers::String> match_type_offset =
Austin Schuh0de30f32020-12-06 12:44:28 -08001701 fbb.CreateString(map.match->type);
Austin Schuh01b4c352020-09-21 23:09:39 -07001702 const flatbuffers::Offset<flatbuffers::String> rename_name_offset =
Austin Schuh0de30f32020-12-06 12:44:28 -08001703 fbb.CreateString(map.rename->name);
Austin Schuh01b4c352020-09-21 23:09:39 -07001704 flatbuffers::Offset<flatbuffers::String> match_source_node_offset;
1705 if (!map.match->source_node.empty()) {
Austin Schuh0de30f32020-12-06 12:44:28 -08001706 match_source_node_offset = fbb.CreateString(map.match->source_node);
Austin Schuh01b4c352020-09-21 23:09:39 -07001707 }
Austin Schuh0de30f32020-12-06 12:44:28 -08001708 Channel::Builder match_builder(fbb);
Austin Schuh01b4c352020-09-21 23:09:39 -07001709 match_builder.add_name(match_name_offset);
1710 match_builder.add_type(match_type_offset);
1711 if (!map.match->source_node.empty()) {
1712 match_builder.add_source_node(match_source_node_offset);
1713 }
1714 const flatbuffers::Offset<Channel> match_offset = match_builder.Finish();
1715
Austin Schuh0de30f32020-12-06 12:44:28 -08001716 Channel::Builder rename_builder(fbb);
Austin Schuh01b4c352020-09-21 23:09:39 -07001717 rename_builder.add_name(rename_name_offset);
1718 const flatbuffers::Offset<Channel> rename_offset = rename_builder.Finish();
1719
Austin Schuh0de30f32020-12-06 12:44:28 -08001720 Map::Builder map_builder(fbb);
Austin Schuh01b4c352020-09-21 23:09:39 -07001721 map_builder.add_match(match_offset);
1722 map_builder.add_rename(rename_offset);
1723 map_offsets.emplace_back(map_builder.Finish());
1724 }
1725
Austin Schuh0de30f32020-12-06 12:44:28 -08001726 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Map>>>
1727 maps_offsets = map_offsets.empty() ? 0 : fbb.CreateVector(map_offsets);
Austin Schuh01b4c352020-09-21 23:09:39 -07001728
Austin Schuh0de30f32020-12-06 12:44:28 -08001729 // And copy everything else over.
1730 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Node>>>
1731 nodes_offset = aos::RecursiveCopyVectorTable(base_config->nodes(), &fbb);
1732
1733 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Application>>>
1734 applications_offset =
1735 aos::RecursiveCopyVectorTable(base_config->applications(), &fbb);
1736
1737 // Now insert everything else in unmodified.
1738 ConfigurationBuilder configuration_builder(fbb);
1739 if (!channels_offset.IsNull()) {
1740 configuration_builder.add_channels(channels_offset);
1741 }
1742 if (!maps_offsets.IsNull()) {
1743 configuration_builder.add_maps(maps_offsets);
1744 }
1745 if (!nodes_offset.IsNull()) {
1746 configuration_builder.add_nodes(nodes_offset);
1747 }
1748 if (!applications_offset.IsNull()) {
1749 configuration_builder.add_applications(applications_offset);
1750 }
1751
1752 if (base_config->has_channel_storage_duration()) {
1753 configuration_builder.add_channel_storage_duration(
1754 base_config->channel_storage_duration());
1755 }
1756
1757 CHECK_EQ(Configuration::MiniReflectTypeTable()->num_elems, 6u)
1758 << ": Merging logic needs to be updated when the number of configuration "
1759 "fields changes.";
1760
1761 fbb.Finish(configuration_builder.Finish());
1762
1763 // Clean it up and return it! By using MergeConfiguration here, we'll
1764 // actually get a deduplicated config for free too.
1765 FlatbufferDetachedBuffer<Configuration> new_merged_config =
1766 configuration::MergeConfiguration(
1767 FlatbufferDetachedBuffer<Configuration>(fbb.Release()));
1768
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001769 remapped_configuration_buffer_ =
1770 std::make_unique<FlatbufferDetachedBuffer<Configuration>>(
Austin Schuh0de30f32020-12-06 12:44:28 -08001771 configuration::MergeConfiguration(new_merged_config, schemas));
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001772
1773 remapped_configuration_ = &remapped_configuration_buffer_->message();
Austin Schuh0de30f32020-12-06 12:44:28 -08001774
1775 // TODO(austin): Lazily re-build to save CPU?
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -08001776}
1777
Austin Schuh6f3babe2020-01-26 20:34:50 -08001778const Channel *LogReader::RemapChannel(const EventLoop *event_loop,
1779 const Channel *channel) {
1780 std::string_view channel_name = channel->name()->string_view();
1781 std::string_view channel_type = channel->type()->string_view();
1782 const int channel_index =
1783 configuration::ChannelIndex(logged_configuration(), channel);
1784 // If the channel is remapped, find the correct channel name to use.
1785 if (remapped_channels_.count(channel_index) > 0) {
Austin Schuhee711052020-08-24 16:06:09 -07001786 VLOG(3) << "Got remapped channel on "
Austin Schuh6f3babe2020-01-26 20:34:50 -08001787 << configuration::CleanedChannelToString(channel);
Austin Schuh0de30f32020-12-06 12:44:28 -08001788 channel_name = remapped_channels_[channel_index].remapped_name;
Austin Schuh6f3babe2020-01-26 20:34:50 -08001789 }
1790
Austin Schuhee711052020-08-24 16:06:09 -07001791 VLOG(2) << "Going to remap channel " << channel_name << " " << channel_type;
Austin Schuh6f3babe2020-01-26 20:34:50 -08001792 const Channel *remapped_channel = configuration::GetChannel(
1793 event_loop->configuration(), channel_name, channel_type,
1794 event_loop->name(), event_loop->node());
1795
1796 CHECK(remapped_channel != nullptr)
1797 << ": Unable to send {\"name\": \"" << channel_name << "\", \"type\": \""
1798 << channel_type << "\"} because it is not in the provided configuration.";
1799
1800 return remapped_channel;
1801}
1802
Austin Schuh287d43d2020-12-04 20:19:33 -08001803LogReader::State::State(std::unique_ptr<TimestampMapper> timestamp_mapper)
1804 : timestamp_mapper_(std::move(timestamp_mapper)) {}
1805
1806void LogReader::State::AddPeer(State *peer) {
1807 if (timestamp_mapper_ && peer->timestamp_mapper_) {
1808 timestamp_mapper_->AddPeer(peer->timestamp_mapper_.get());
1809 }
1810}
Austin Schuh858c9f32020-08-31 16:56:12 -07001811
1812EventLoop *LogReader::State::SetNodeEventLoopFactory(
1813 NodeEventLoopFactory *node_event_loop_factory) {
1814 node_event_loop_factory_ = node_event_loop_factory;
1815 event_loop_unique_ptr_ =
1816 node_event_loop_factory_->MakeEventLoop("log_reader");
1817 return event_loop_unique_ptr_.get();
1818}
1819
1820void LogReader::State::SetChannelCount(size_t count) {
1821 channels_.resize(count);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001822 remote_timestamp_senders_.resize(count);
Austin Schuh858c9f32020-08-31 16:56:12 -07001823 filters_.resize(count);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001824 channel_source_state_.resize(count);
1825 factory_channel_index_.resize(count);
1826 queue_index_map_.resize(count);
Austin Schuh858c9f32020-08-31 16:56:12 -07001827}
1828
1829void LogReader::State::SetChannel(
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001830 size_t logged_channel_index, size_t factory_channel_index,
1831 std::unique_ptr<RawSender> sender,
Austin Schuh2f8fd752020-09-01 22:38:28 -07001832 message_bridge::NoncausalOffsetEstimator *filter,
Austin Schuh969cd602021-01-03 00:09:45 -08001833 RemoteMessageSender *remote_timestamp_sender, State *source_state) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001834 channels_[logged_channel_index] = std::move(sender);
1835 filters_[logged_channel_index] = filter;
1836 remote_timestamp_senders_[logged_channel_index] = remote_timestamp_sender;
1837
1838 if (source_state) {
1839 channel_source_state_[logged_channel_index] = source_state;
1840
1841 if (remote_timestamp_sender != nullptr) {
1842 source_state->queue_index_map_[logged_channel_index] =
Austin Schuh9942bae2021-01-07 22:06:44 -08001843 std::make_unique<std::vector<State::ContiguousSentTimestamp>>();
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001844 }
1845 }
1846
1847 factory_channel_index_[logged_channel_index] = factory_channel_index;
1848}
1849
Austin Schuh287d43d2020-12-04 20:19:33 -08001850bool LogReader::State::Send(const TimestampedMessage &timestamped_message) {
1851 aos::RawSender *sender = channels_[timestamped_message.channel_index].get();
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001852 uint32_t remote_queue_index = 0xffffffff;
1853
Austin Schuh287d43d2020-12-04 20:19:33 -08001854 if (remote_timestamp_senders_[timestamped_message.channel_index] != nullptr) {
Austin Schuh9942bae2021-01-07 22:06:44 -08001855 std::vector<ContiguousSentTimestamp> *queue_index_map = CHECK_NOTNULL(
Austin Schuh287d43d2020-12-04 20:19:33 -08001856 CHECK_NOTNULL(channel_source_state_[timestamped_message.channel_index])
1857 ->queue_index_map_[timestamped_message.channel_index]
1858 .get());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001859
Austin Schuh9942bae2021-01-07 22:06:44 -08001860 struct SentTimestamp {
1861 monotonic_clock::time_point monotonic_event_time;
1862 uint32_t queue_index;
1863 } search;
1864
Austin Schuh287d43d2020-12-04 20:19:33 -08001865 search.monotonic_event_time = timestamped_message.monotonic_remote_time;
Austin Schuh287d43d2020-12-04 20:19:33 -08001866 search.queue_index = timestamped_message.remote_queue_index;
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001867
1868 // Find the sent time if available.
1869 auto element = std::lower_bound(
1870 queue_index_map->begin(), queue_index_map->end(), search,
Austin Schuh9942bae2021-01-07 22:06:44 -08001871 [](ContiguousSentTimestamp a, SentTimestamp b) {
1872 if (a.ending_monotonic_event_time < b.monotonic_event_time) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001873 return true;
1874 }
Austin Schuh9942bae2021-01-07 22:06:44 -08001875 if (a.starting_monotonic_event_time > b.monotonic_event_time) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001876 return false;
1877 }
Austin Schuh9942bae2021-01-07 22:06:44 -08001878
1879 if (a.ending_queue_index < b.queue_index) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001880 return true;
1881 }
Austin Schuh9942bae2021-01-07 22:06:44 -08001882 if (a.starting_queue_index >= b.queue_index) {
1883 return false;
1884 }
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001885
Austin Schuh9942bae2021-01-07 22:06:44 -08001886 // If it isn't clearly below or above, it is below. Since we return
1887 // the last element <, this will return a match.
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001888 return false;
1889 });
1890
1891 // TODO(austin): Be a bit more principled here, but we will want to do that
1892 // after the logger rewrite. We hit this when one node finishes, but the
1893 // other node isn't done yet. So there is no send time, but there is a
1894 // receive time.
1895 if (element != queue_index_map->end()) {
Austin Schuh9942bae2021-01-07 22:06:44 -08001896 CHECK_GE(timestamped_message.monotonic_remote_time,
1897 element->starting_monotonic_event_time);
1898 CHECK_LE(timestamped_message.monotonic_remote_time,
1899 element->ending_monotonic_event_time);
1900 CHECK_GE(timestamped_message.remote_queue_index,
1901 element->starting_queue_index);
1902 CHECK_LE(timestamped_message.remote_queue_index,
1903 element->ending_queue_index);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001904
Austin Schuh9942bae2021-01-07 22:06:44 -08001905 remote_queue_index = timestamped_message.remote_queue_index +
1906 element->actual_queue_index -
1907 element->starting_queue_index;
1908 } else {
1909 VLOG(1) << "No timestamp match in the map.";
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001910 }
1911 }
1912
1913 // Send! Use the replayed queue index here instead of the logged queue index
1914 // for the remote queue index. This makes re-logging work.
Austin Schuh287d43d2020-12-04 20:19:33 -08001915 const bool sent = sender->Send(
1916 timestamped_message.data.message().data()->Data(),
1917 timestamped_message.data.message().data()->size(),
1918 timestamped_message.monotonic_remote_time,
1919 timestamped_message.realtime_remote_time, remote_queue_index);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001920 if (!sent) return false;
1921
Austin Schuh287d43d2020-12-04 20:19:33 -08001922 if (queue_index_map_[timestamped_message.channel_index]) {
Austin Schuh9942bae2021-01-07 22:06:44 -08001923 if (queue_index_map_[timestamped_message.channel_index]->empty()) {
1924 // Nothing here, start a range with 0 length.
1925 ContiguousSentTimestamp timestamp;
1926 timestamp.starting_monotonic_event_time =
1927 timestamp.ending_monotonic_event_time =
1928 timestamped_message.monotonic_event_time;
1929 timestamp.starting_queue_index = timestamp.ending_queue_index =
1930 timestamped_message.queue_index;
1931 timestamp.actual_queue_index = sender->sent_queue_index();
1932 queue_index_map_[timestamped_message.channel_index]->emplace_back(
1933 timestamp);
1934 } else {
1935 // We've got something. See if the next timestamp is still contiguous. If
1936 // so, grow it.
1937 ContiguousSentTimestamp *back =
1938 &queue_index_map_[timestamped_message.channel_index]->back();
1939 if ((back->starting_queue_index - back->actual_queue_index) ==
1940 (timestamped_message.queue_index - sender->sent_queue_index())) {
1941 back->ending_queue_index = timestamped_message.queue_index;
1942 back->ending_monotonic_event_time =
1943 timestamped_message.monotonic_event_time;
1944 } else {
1945 // Otherwise, make a new one.
1946 ContiguousSentTimestamp timestamp;
1947 timestamp.starting_monotonic_event_time =
1948 timestamp.ending_monotonic_event_time =
1949 timestamped_message.monotonic_event_time;
1950 timestamp.starting_queue_index = timestamp.ending_queue_index =
1951 timestamped_message.queue_index;
1952 timestamp.actual_queue_index = sender->sent_queue_index();
1953 queue_index_map_[timestamped_message.channel_index]->emplace_back(
1954 timestamp);
1955 }
1956 }
1957
1958 // TODO(austin): Should we prune the map? On a many day log, I only saw the
1959 // queue index diverge a couple of elements, which would be a very small
1960 // map.
Austin Schuh287d43d2020-12-04 20:19:33 -08001961 } else if (remote_timestamp_senders_[timestamped_message.channel_index] !=
1962 nullptr) {
Austin Schuh969cd602021-01-03 00:09:45 -08001963 flatbuffers::FlatBufferBuilder fbb;
1964 fbb.ForceDefaults(true);
Austin Schuh315b96b2020-12-11 21:21:12 -08001965 flatbuffers::Offset<flatbuffers::String> boot_uuid_offset =
Austin Schuh969cd602021-01-03 00:09:45 -08001966 fbb.CreateString(event_loop_->boot_uuid().string_view());
Austin Schuh315b96b2020-12-11 21:21:12 -08001967
Austin Schuh969cd602021-01-03 00:09:45 -08001968 RemoteMessage::Builder message_header_builder(fbb);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001969
1970 message_header_builder.add_channel_index(
Austin Schuh287d43d2020-12-04 20:19:33 -08001971 factory_channel_index_[timestamped_message.channel_index]);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001972
1973 // Swap the remote and sent metrics. They are from the sender's
1974 // perspective, not the receiver's perspective.
1975 message_header_builder.add_monotonic_sent_time(
1976 sender->monotonic_sent_time().time_since_epoch().count());
1977 message_header_builder.add_realtime_sent_time(
1978 sender->realtime_sent_time().time_since_epoch().count());
1979 message_header_builder.add_queue_index(sender->sent_queue_index());
1980
1981 message_header_builder.add_monotonic_remote_time(
Austin Schuh287d43d2020-12-04 20:19:33 -08001982 timestamped_message.monotonic_remote_time.time_since_epoch().count());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001983 message_header_builder.add_realtime_remote_time(
Austin Schuh287d43d2020-12-04 20:19:33 -08001984 timestamped_message.realtime_remote_time.time_since_epoch().count());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001985
1986 message_header_builder.add_remote_queue_index(remote_queue_index);
Austin Schuh315b96b2020-12-11 21:21:12 -08001987 message_header_builder.add_boot_uuid(boot_uuid_offset);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001988
Austin Schuh969cd602021-01-03 00:09:45 -08001989 fbb.Finish(message_header_builder.Finish());
1990
1991 remote_timestamp_senders_[timestamped_message.channel_index]->Send(
1992 FlatbufferDetachedBuffer<RemoteMessage>(fbb.Release()),
1993 timestamped_message.monotonic_timestamp_time);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001994 }
1995
1996 return true;
1997}
1998
Austin Schuh969cd602021-01-03 00:09:45 -08001999LogReader::RemoteMessageSender::RemoteMessageSender(
2000 aos::Sender<message_bridge::RemoteMessage> sender, EventLoop *event_loop)
2001 : event_loop_(event_loop),
2002 sender_(std::move(sender)),
2003 timer_(event_loop->AddTimer([this]() { SendTimestamp(); })) {}
2004
2005void LogReader::RemoteMessageSender::ScheduleTimestamp() {
2006 if (remote_timestamps_.empty()) {
2007 CHECK_NOTNULL(timer_);
2008 timer_->Disable();
2009 scheduled_time_ = monotonic_clock::min_time;
2010 return;
2011 }
2012
2013 if (scheduled_time_ != remote_timestamps_.front().monotonic_timestamp_time) {
2014 CHECK_NOTNULL(timer_);
Austin Schuh816e5d62021-01-05 23:42:20 -08002015 timer_->Setup(remote_timestamps_.front().monotonic_timestamp_time);
Austin Schuh969cd602021-01-03 00:09:45 -08002016 scheduled_time_ = remote_timestamps_.front().monotonic_timestamp_time;
2017 }
2018}
2019
2020void LogReader::RemoteMessageSender::Send(
2021 FlatbufferDetachedBuffer<RemoteMessage> remote_message,
2022 monotonic_clock::time_point monotonic_timestamp_time) {
2023 // There are 2 cases. Either we have a monotonic_timestamp_time and need to
2024 // resend the timestamp at the correct time, or we don't and can send it
2025 // immediately.
2026 if (monotonic_timestamp_time == monotonic_clock::min_time) {
2027 CHECK(remote_timestamps_.empty())
2028 << ": Unsupported mix of timestamps and no timestamps.";
2029 sender_.Send(std::move(remote_message));
2030 } else {
2031 remote_timestamps_.emplace_back(std::move(remote_message),
2032 monotonic_timestamp_time);
2033 ScheduleTimestamp();
2034 }
2035}
2036
2037void LogReader::RemoteMessageSender::SendTimestamp() {
2038 CHECK_EQ(event_loop_->context().monotonic_event_time, scheduled_time_);
2039 CHECK(!remote_timestamps_.empty());
2040
2041 // Send out all timestamps at the currently scheduled time.
2042 while (remote_timestamps_.front().monotonic_timestamp_time ==
2043 scheduled_time_) {
2044 sender_.Send(std::move(remote_timestamps_.front().remote_message));
2045 remote_timestamps_.pop_front();
2046 if (remote_timestamps_.empty()) {
2047 break;
2048 }
2049 }
2050 scheduled_time_ = monotonic_clock::min_time;
2051
2052 ScheduleTimestamp();
2053}
2054
2055LogReader::RemoteMessageSender *LogReader::State::RemoteTimestampSender(
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07002056 const Node *delivered_node) {
2057 auto sender = remote_timestamp_senders_map_.find(delivered_node);
2058
2059 if (sender == remote_timestamp_senders_map_.end()) {
Austin Schuh969cd602021-01-03 00:09:45 -08002060 sender =
2061 remote_timestamp_senders_map_
2062 .emplace(delivered_node,
2063 std::make_unique<RemoteMessageSender>(
2064 event_loop()->MakeSender<RemoteMessage>(absl::StrCat(
2065 "/aos/remote_timestamps/",
2066 delivered_node->name()->string_view())),
2067 event_loop()))
2068 .first;
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07002069 }
2070
Austin Schuh969cd602021-01-03 00:09:45 -08002071 return sender->second.get();
Austin Schuh858c9f32020-08-31 16:56:12 -07002072}
2073
Austin Schuhdda74ec2021-01-03 19:30:37 -08002074TimestampedMessage LogReader::State::PopOldest() {
Austin Schuh858c9f32020-08-31 16:56:12 -07002075 CHECK_GT(sorted_messages_.size(), 0u);
2076
Austin Schuh287d43d2020-12-04 20:19:33 -08002077 std::tuple<TimestampedMessage, message_bridge::NoncausalOffsetEstimator *>
Austin Schuh858c9f32020-08-31 16:56:12 -07002078 result = std::move(sorted_messages_.front());
Austin Schuh2f8fd752020-09-01 22:38:28 -07002079 VLOG(2) << MaybeNodeName(event_loop_->node()) << "PopOldest Popping "
Austin Schuh858c9f32020-08-31 16:56:12 -07002080 << std::get<0>(result).monotonic_event_time;
2081 sorted_messages_.pop_front();
2082 SeedSortedMessages();
2083
Austin Schuh287d43d2020-12-04 20:19:33 -08002084 if (std::get<1>(result) != nullptr) {
Austin Schuh816e5d62021-01-05 23:42:20 -08002085 std::get<1>(result)->Pop(event_loop_->node(),
2086 std::get<0>(result).monotonic_event_time);
Austin Schuh2f8fd752020-09-01 22:38:28 -07002087 }
Austin Schuh287d43d2020-12-04 20:19:33 -08002088 return std::move(std::get<0>(result));
Austin Schuh858c9f32020-08-31 16:56:12 -07002089}
2090
2091monotonic_clock::time_point LogReader::State::OldestMessageTime() const {
2092 if (sorted_messages_.size() > 0) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07002093 VLOG(2) << MaybeNodeName(event_loop_->node()) << "oldest message at "
Austin Schuh858c9f32020-08-31 16:56:12 -07002094 << std::get<0>(sorted_messages_.front()).monotonic_event_time;
2095 return std::get<0>(sorted_messages_.front()).monotonic_event_time;
2096 }
2097
Austin Schuh287d43d2020-12-04 20:19:33 -08002098 TimestampedMessage *m =
2099 timestamp_mapper_ ? timestamp_mapper_->Front() : nullptr;
2100 if (m == nullptr) {
2101 return monotonic_clock::max_time;
2102 }
2103 return m->monotonic_event_time;
Austin Schuh858c9f32020-08-31 16:56:12 -07002104}
2105
2106void LogReader::State::SeedSortedMessages() {
Austin Schuh287d43d2020-12-04 20:19:33 -08002107 if (!timestamp_mapper_) return;
Austin Schuh858c9f32020-08-31 16:56:12 -07002108 const aos::monotonic_clock::time_point end_queue_time =
2109 (sorted_messages_.size() > 0
2110 ? std::get<0>(sorted_messages_.front()).monotonic_event_time
Austin Schuh287d43d2020-12-04 20:19:33 -08002111 : timestamp_mapper_->monotonic_start_time()) +
Austin Schuhf0688662020-12-19 15:37:45 -08002112 chrono::duration_cast<chrono::seconds>(
2113 chrono::duration<double>(FLAGS_time_estimation_buffer_seconds));
Austin Schuh858c9f32020-08-31 16:56:12 -07002114
2115 while (true) {
Austin Schuh287d43d2020-12-04 20:19:33 -08002116 TimestampedMessage *m = timestamp_mapper_->Front();
2117 if (m == nullptr) {
Austin Schuh858c9f32020-08-31 16:56:12 -07002118 return;
2119 }
2120 if (sorted_messages_.size() > 0) {
Austin Schuhf0688662020-12-19 15:37:45 -08002121 // Stop placing sorted messages on the list once we have
2122 // --time_estimation_buffer_seconds seconds queued up (but queue at least
2123 // until the log starts.
Austin Schuh858c9f32020-08-31 16:56:12 -07002124 if (end_queue_time <
2125 std::get<0>(sorted_messages_.back()).monotonic_event_time) {
2126 return;
2127 }
2128 }
2129
Austin Schuh2f8fd752020-09-01 22:38:28 -07002130 message_bridge::NoncausalOffsetEstimator *filter = nullptr;
2131
Austin Schuh287d43d2020-12-04 20:19:33 -08002132 TimestampedMessage timestamped_message = std::move(*m);
2133 timestamp_mapper_->PopFront();
Austin Schuh858c9f32020-08-31 16:56:12 -07002134
Austin Schuh2f8fd752020-09-01 22:38:28 -07002135 // Skip any messages without forwarding information.
Austin Schuh0de30f32020-12-06 12:44:28 -08002136 if (timestamped_message.monotonic_remote_time !=
2137 monotonic_clock::min_time) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07002138 // Got a forwarding timestamp!
Austin Schuh287d43d2020-12-04 20:19:33 -08002139 filter = filters_[timestamped_message.channel_index];
Austin Schuh2f8fd752020-09-01 22:38:28 -07002140
2141 CHECK(filter != nullptr);
2142
2143 // Call the correct method depending on if we are the forward or
2144 // reverse direction here.
2145 filter->Sample(event_loop_->node(),
Austin Schuh287d43d2020-12-04 20:19:33 -08002146 timestamped_message.monotonic_event_time,
2147 timestamped_message.monotonic_remote_time);
Austin Schuh816e5d62021-01-05 23:42:20 -08002148
2149 if (timestamped_message.monotonic_timestamp_time !=
2150 monotonic_clock::min_time) {
2151 // TODO(austin): This assumes that this timestamp is only logged on the
2152 // node which sent the data. That is correct for now, but should be
2153 // explicitly checked somewhere.
2154 filter->ReverseSample(event_loop_->node(),
2155 timestamped_message.monotonic_event_time,
2156 timestamped_message.monotonic_timestamp_time);
2157 }
Austin Schuh2f8fd752020-09-01 22:38:28 -07002158 }
Austin Schuh287d43d2020-12-04 20:19:33 -08002159 sorted_messages_.emplace_back(std::move(timestamped_message), filter);
Austin Schuh858c9f32020-08-31 16:56:12 -07002160 }
2161}
2162
2163void LogReader::State::Deregister() {
2164 for (size_t i = 0; i < channels_.size(); ++i) {
2165 channels_[i].reset();
2166 }
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07002167 remote_timestamp_senders_map_.clear();
Austin Schuh858c9f32020-08-31 16:56:12 -07002168 event_loop_unique_ptr_.reset();
2169 event_loop_ = nullptr;
2170 timer_handler_ = nullptr;
2171 node_event_loop_factory_ = nullptr;
2172}
2173
Austin Schuhe309d2a2019-11-29 13:25:21 -08002174} // namespace logger
2175} // namespace aos