blob: 6e92568a968c3c28c331545588337a5ebc4a27f6 [file] [log] [blame]
Austin Schuhb06f03b2021-02-17 22:00:37 -08001#include "aos/events/logging/log_reader.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>
Brian Silverman8ff74aa2021-02-05 16:37:15 -08008
Austin Schuhe309d2a2019-11-29 13:25:21 -08009#include <vector>
10
Austin Schuh2f8fd752020-09-01 22:38:28 -070011#include "absl/strings/escaping.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080012#include "absl/types/span.h"
13#include "aos/events/event_loop.h"
Austin Schuhf6f9bf32020-10-11 14:37:43 -070014#include "aos/events/logging/logfile_sorting.h"
James Kuszmaul38735e82019-12-07 16:42:06 -080015#include "aos/events/logging/logger_generated.h"
Austin Schuh64fab802020-09-09 22:47:47 -070016#include "aos/events/logging/uuid.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080017#include "aos/flatbuffer_merge.h"
Austin Schuh0ca1fd32020-12-18 22:53:05 -080018#include "aos/network/multinode_timestamp_filter.h"
Austin Schuh0de30f32020-12-06 12:44:28 -080019#include "aos/network/remote_message_generated.h"
20#include "aos/network/remote_message_schema.h"
Austin Schuh288479d2019-12-18 19:47:52 -080021#include "aos/network/team_number.h"
Austin Schuh61e973f2021-02-21 21:43:56 -080022#include "aos/network/timestamp_channel.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080023#include "aos/time/time.h"
Brian Silvermanae7c0332020-09-30 16:58:23 -070024#include "aos/util/file.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080025#include "flatbuffers/flatbuffers.h"
Austin Schuh8c399962020-12-25 21:51:45 -080026#include "openssl/sha.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080027
Austin Schuh15649d62019-12-28 16:36:38 -080028DEFINE_bool(skip_missing_forwarding_entries, false,
29 "If true, drop any forwarding entries with missing data. If "
30 "false, CHECK.");
Austin Schuhe309d2a2019-11-29 13:25:21 -080031
Austin Schuh0ca1fd32020-12-18 22:53:05 -080032DECLARE_bool(timestamps_to_csv);
Austin Schuh8bd96322020-02-13 21:18:22 -080033
Austin Schuh2f8fd752020-09-01 22:38:28 -070034DEFINE_bool(skip_order_validation, false,
35 "If true, ignore any out of orderness in replay");
36
Austin Schuhf0688662020-12-19 15:37:45 -080037DEFINE_double(
38 time_estimation_buffer_seconds, 2.0,
39 "The time to buffer ahead in the log file to accurately reconstruct time.");
40
Austin Schuhe309d2a2019-11-29 13:25:21 -080041namespace aos {
42namespace logger {
Austin Schuh0afc4d12020-10-19 11:42:04 -070043namespace {
Austin Schuh8c399962020-12-25 21:51:45 -080044
Austin Schuh315b96b2020-12-11 21:21:12 -080045std::string LogFileVectorToString(std::vector<LogFile> log_files) {
46 std::stringstream ss;
Austin Schuh297d2352021-01-21 19:02:17 -080047 for (const auto &f : log_files) {
Austin Schuh315b96b2020-12-11 21:21:12 -080048 ss << f << "\n";
49 }
50 return ss.str();
51}
52
Austin Schuh0de30f32020-12-06 12:44:28 -080053// Copies the channel, removing the schema as we go. If new_name is provided,
54// it is used instead of the name inside the channel. If new_type is provided,
55// it is used instead of the type in the channel.
56flatbuffers::Offset<Channel> CopyChannel(const Channel *c,
57 std::string_view new_name,
58 std::string_view new_type,
59 flatbuffers::FlatBufferBuilder *fbb) {
60 flatbuffers::Offset<flatbuffers::String> name_offset =
61 fbb->CreateSharedString(new_name.empty() ? c->name()->string_view()
62 : new_name);
63 flatbuffers::Offset<flatbuffers::String> type_offset =
64 fbb->CreateSharedString(new_type.empty() ? c->type()->str() : new_type);
65 flatbuffers::Offset<flatbuffers::String> source_node_offset =
66 c->has_source_node() ? fbb->CreateSharedString(c->source_node()->str())
67 : 0;
68
69 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Connection>>>
70 destination_nodes_offset =
71 aos::RecursiveCopyVectorTable(c->destination_nodes(), fbb);
72
73 flatbuffers::Offset<
74 flatbuffers::Vector<flatbuffers::Offset<flatbuffers::String>>>
75 logger_nodes_offset = aos::CopyVectorSharedString(c->logger_nodes(), fbb);
76
77 Channel::Builder channel_builder(*fbb);
78 channel_builder.add_name(name_offset);
79 channel_builder.add_type(type_offset);
80 if (c->has_frequency()) {
81 channel_builder.add_frequency(c->frequency());
82 }
83 if (c->has_max_size()) {
84 channel_builder.add_max_size(c->max_size());
85 }
86 if (c->has_num_senders()) {
87 channel_builder.add_num_senders(c->num_senders());
88 }
89 if (c->has_num_watchers()) {
90 channel_builder.add_num_watchers(c->num_watchers());
91 }
92 if (!source_node_offset.IsNull()) {
93 channel_builder.add_source_node(source_node_offset);
94 }
95 if (!destination_nodes_offset.IsNull()) {
96 channel_builder.add_destination_nodes(destination_nodes_offset);
97 }
98 if (c->has_logger()) {
99 channel_builder.add_logger(c->logger());
100 }
101 if (!logger_nodes_offset.IsNull()) {
102 channel_builder.add_logger_nodes(logger_nodes_offset);
103 }
104 if (c->has_read_method()) {
105 channel_builder.add_read_method(c->read_method());
106 }
107 if (c->has_num_readers()) {
108 channel_builder.add_num_readers(c->num_readers());
109 }
110 return channel_builder.Finish();
111}
112
Austin Schuhe309d2a2019-11-29 13:25:21 -0800113namespace chrono = std::chrono;
Austin Schuh0de30f32020-12-06 12:44:28 -0800114using message_bridge::RemoteMessage;
Austin Schuh0afc4d12020-10-19 11:42:04 -0700115} // namespace
Austin Schuhe309d2a2019-11-29 13:25:21 -0800116
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800117LogReader::LogReader(std::string_view filename,
118 const Configuration *replay_configuration)
Austin Schuh287d43d2020-12-04 20:19:33 -0800119 : LogReader(SortParts({std::string(filename)}), replay_configuration) {}
Austin Schuhfa895892020-01-07 20:07:41 -0800120
Austin Schuh287d43d2020-12-04 20:19:33 -0800121LogReader::LogReader(std::vector<LogFile> log_files,
Austin Schuhfa895892020-01-07 20:07:41 -0800122 const Configuration *replay_configuration)
Austin Schuh287d43d2020-12-04 20:19:33 -0800123 : log_files_(std::move(log_files)),
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800124 replay_configuration_(replay_configuration) {
Austin Schuh0ca51f32020-12-25 21:51:45 -0800125 CHECK_GT(log_files_.size(), 0u);
126 {
127 // Validate that we have the same config everwhere. This will be true if
128 // all the parts were sorted together and the configs match.
129 const Configuration *config = nullptr;
Austin Schuh297d2352021-01-21 19:02:17 -0800130 for (const LogFile &log_file : log_files_) {
131 if (log_file.config.get() == nullptr) {
132 LOG(FATAL) << "Couldn't find a config in " << log_file;
133 }
Austin Schuh0ca51f32020-12-25 21:51:45 -0800134 if (config == nullptr) {
135 config = log_file.config.get();
136 } else {
137 CHECK_EQ(config, log_file.config.get());
138 }
139 }
140 }
Austin Schuhdda74ec2021-01-03 19:30:37 -0800141
Austin Schuh6331ef92020-01-07 18:28:09 -0800142 MakeRemappedConfig();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800143
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700144 // Remap all existing remote timestamp channels. They will be recreated, and
145 // the data logged isn't relevant anymore.
Austin Schuh3c5dae52020-10-06 18:55:18 -0700146 for (const Node *node : configuration::GetNodes(logged_configuration())) {
Austin Schuh61e973f2021-02-21 21:43:56 -0800147 message_bridge::ChannelTimestampFinder finder(logged_configuration(),
148 "log_reader", node);
149
150 absl::btree_set<std::string_view> remote_nodes;
151
152 for (const Channel *channel : *logged_configuration()->channels()) {
153 if (!configuration::ChannelIsSendableOnNode(channel, node)) {
154 continue;
155 }
156 if (!channel->has_destination_nodes()) {
157 continue;
158 }
159 for (const Connection *connection : *channel->destination_nodes()) {
160 if (configuration::ConnectionDeliveryTimeIsLoggedOnNode(connection,
161 node)) {
162 // Start by seeing if the split timestamp channels are being used for
163 // this message. If so, remap them.
164 const Channel *timestamp_channel = configuration::GetChannel(
165 logged_configuration(),
166 finder.SplitChannelName(channel, connection),
167 RemoteMessage::GetFullyQualifiedName(), "", node, true);
168
169 if (timestamp_channel != nullptr) {
170 if (timestamp_channel->logger() != LoggerConfig::NOT_LOGGED) {
171 RemapLoggedChannel<RemoteMessage>(
172 timestamp_channel->name()->string_view(), node);
173 }
174 continue;
175 }
176
177 // Otherwise collect this one up as a node to look for a combined
178 // channel from. It is more efficient to compare nodes than channels.
179 remote_nodes.insert(connection->name()->string_view());
180 }
181 }
182 }
183
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700184 std::vector<const Node *> timestamp_logger_nodes =
185 configuration::TimestampNodes(logged_configuration(), node);
Austin Schuh61e973f2021-02-21 21:43:56 -0800186 for (const std::string_view remote_node : remote_nodes) {
187 const std::string channel = finder.CombinedChannelName(remote_node);
188
Austin Schuh0de30f32020-12-06 12:44:28 -0800189 // See if the log file is an old log with MessageHeader channels in it, or
190 // a newer log with RemoteMessage. If we find an older log, rename the
191 // type too along with the name.
192 if (HasChannel<MessageHeader>(channel, node)) {
193 CHECK(!HasChannel<RemoteMessage>(channel, node))
194 << ": Can't have both a MessageHeader and RemoteMessage remote "
195 "timestamp channel.";
James Kuszmaul4f106fb2021-01-05 20:53:02 -0800196 // In theory, we should check NOT_LOGGED like RemoteMessage and be more
197 // careful about updating the config, but there are fewer and fewer logs
198 // with MessageHeader remote messages, so it isn't worth the effort.
Austin Schuh0de30f32020-12-06 12:44:28 -0800199 RemapLoggedChannel<MessageHeader>(channel, node, "/original",
200 "aos.message_bridge.RemoteMessage");
201 } else {
202 CHECK(HasChannel<RemoteMessage>(channel, node))
203 << ": Failed to find {\"name\": \"" << channel << "\", \"type\": \""
204 << RemoteMessage::GetFullyQualifiedName() << "\"} for node "
205 << node->name()->string_view();
James Kuszmaul4f106fb2021-01-05 20:53:02 -0800206 // Only bother to remap if there's something on the channel. We can
207 // tell if the channel was marked NOT_LOGGED or not. This makes the
208 // config not change un-necesarily when we replay a log with NOT_LOGGED
209 // messages.
210 if (HasLoggedChannel<RemoteMessage>(channel, node)) {
211 RemapLoggedChannel<RemoteMessage>(channel, node);
212 }
Austin Schuh0de30f32020-12-06 12:44:28 -0800213 }
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700214 }
215 }
216
Austin Schuh6aa77be2020-02-22 21:06:40 -0800217 if (replay_configuration) {
218 CHECK_EQ(configuration::MultiNode(configuration()),
219 configuration::MultiNode(replay_configuration))
Austin Schuh2f8fd752020-09-01 22:38:28 -0700220 << ": Log file and replay config need to both be multi or single "
221 "node.";
Austin Schuh6aa77be2020-02-22 21:06:40 -0800222 }
223
Austin Schuh6f3babe2020-01-26 20:34:50 -0800224 if (!configuration::MultiNode(configuration())) {
Austin Schuh287d43d2020-12-04 20:19:33 -0800225 states_.emplace_back(std::make_unique<State>(
226 std::make_unique<TimestampMapper>(FilterPartsForNode(log_files_, ""))));
Austin Schuh8bd96322020-02-13 21:18:22 -0800227 } else {
Austin Schuh6aa77be2020-02-22 21:06:40 -0800228 if (replay_configuration) {
James Kuszmaul46d82582020-05-09 19:50:09 -0700229 CHECK_EQ(logged_configuration()->nodes()->size(),
Austin Schuh6aa77be2020-02-22 21:06:40 -0800230 replay_configuration->nodes()->size())
Austin Schuh2f8fd752020-09-01 22:38:28 -0700231 << ": Log file and replay config need to have matching nodes "
232 "lists.";
James Kuszmaul46d82582020-05-09 19:50:09 -0700233 for (const Node *node : *logged_configuration()->nodes()) {
234 if (configuration::GetNode(replay_configuration, node) == nullptr) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700235 LOG(FATAL) << "Found node " << FlatbufferToJson(node)
236 << " in logged config that is not present in the replay "
237 "config.";
James Kuszmaul46d82582020-05-09 19:50:09 -0700238 }
239 }
Austin Schuh6aa77be2020-02-22 21:06:40 -0800240 }
Austin Schuh8bd96322020-02-13 21:18:22 -0800241 states_.resize(configuration()->nodes()->size());
Austin Schuh6f3babe2020-01-26 20:34:50 -0800242 }
Austin Schuhe309d2a2019-11-29 13:25:21 -0800243}
244
Austin Schuh6aa77be2020-02-22 21:06:40 -0800245LogReader::~LogReader() {
Austin Schuh39580f12020-08-01 14:44:08 -0700246 if (event_loop_factory_unique_ptr_) {
247 Deregister();
248 } else if (event_loop_factory_ != nullptr) {
249 LOG(FATAL) << "Must call Deregister before the SimulatedEventLoopFactory "
250 "is destroyed";
251 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700252 // Zero out some buffers. It's easy to do use-after-frees on these, so make
253 // it more obvious.
Austin Schuh39580f12020-08-01 14:44:08 -0700254 if (remapped_configuration_buffer_) {
255 remapped_configuration_buffer_->Wipe();
256 }
Austin Schuh8bd96322020-02-13 21:18:22 -0800257}
Austin Schuhe309d2a2019-11-29 13:25:21 -0800258
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800259const Configuration *LogReader::logged_configuration() const {
Austin Schuh0ca51f32020-12-25 21:51:45 -0800260 return log_files_[0].config.get();
Austin Schuhe309d2a2019-11-29 13:25:21 -0800261}
262
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800263const Configuration *LogReader::configuration() const {
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800264 return remapped_configuration_;
265}
266
Austin Schuh07676622021-01-21 18:59:17 -0800267std::vector<const Node *> LogReader::LoggedNodes() const {
268 return configuration::GetNodes(logged_configuration());
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800269}
Austin Schuh15649d62019-12-28 16:36:38 -0800270
Austin Schuh11d43732020-09-21 17:28:30 -0700271monotonic_clock::time_point LogReader::monotonic_start_time(
272 const Node *node) const {
Austin Schuh8bd96322020-02-13 21:18:22 -0800273 State *state =
274 states_[configuration::GetNodeIndex(configuration(), node)].get();
275 CHECK(state != nullptr) << ": Unknown node " << FlatbufferToJson(node);
276
Austin Schuh858c9f32020-08-31 16:56:12 -0700277 return state->monotonic_start_time();
Austin Schuhe309d2a2019-11-29 13:25:21 -0800278}
279
Austin Schuh11d43732020-09-21 17:28:30 -0700280realtime_clock::time_point LogReader::realtime_start_time(
281 const Node *node) const {
Austin Schuh8bd96322020-02-13 21:18:22 -0800282 State *state =
283 states_[configuration::GetNodeIndex(configuration(), node)].get();
284 CHECK(state != nullptr) << ": Unknown node " << FlatbufferToJson(node);
285
Austin Schuh858c9f32020-08-31 16:56:12 -0700286 return state->realtime_start_time();
Austin Schuhe309d2a2019-11-29 13:25:21 -0800287}
288
James Kuszmaul84ff3e52020-01-03 19:48:53 -0800289void LogReader::Register() {
290 event_loop_factory_unique_ptr_ =
Austin Schuhac0771c2020-01-07 18:36:30 -0800291 std::make_unique<SimulatedEventLoopFactory>(configuration());
James Kuszmaul84ff3e52020-01-03 19:48:53 -0800292 Register(event_loop_factory_unique_ptr_.get());
293}
294
Austin Schuh92547522019-12-28 14:33:43 -0800295void LogReader::Register(SimulatedEventLoopFactory *event_loop_factory) {
Austin Schuh92547522019-12-28 14:33:43 -0800296 event_loop_factory_ = event_loop_factory;
Austin Schuhe5bbd9e2020-09-21 17:29:20 -0700297 remapped_configuration_ = event_loop_factory_->configuration();
Austin Schuh0ca1fd32020-12-18 22:53:05 -0800298 filters_ =
299 std::make_unique<message_bridge::MultiNodeNoncausalOffsetEstimator>(
Austin Schuhba20ea72021-01-21 16:47:01 -0800300 event_loop_factory_->configuration(), logged_configuration(),
Austin Schuhfe3fb342021-01-16 18:50:37 -0800301 FLAGS_skip_order_validation,
302 chrono::duration_cast<chrono::nanoseconds>(
303 chrono::duration<double>(FLAGS_time_estimation_buffer_seconds)));
Austin Schuh92547522019-12-28 14:33:43 -0800304
Austin Schuhe639ea12021-01-25 13:00:22 -0800305 std::vector<TimestampMapper *> timestamp_mappers;
Brian Silvermand90905f2020-09-23 14:42:56 -0700306 for (const Node *node : configuration::GetNodes(configuration())) {
Austin Schuh8bd96322020-02-13 21:18:22 -0800307 const size_t node_index =
308 configuration::GetNodeIndex(configuration(), node);
Austin Schuh287d43d2020-12-04 20:19:33 -0800309 std::vector<LogParts> filtered_parts = FilterPartsForNode(
310 log_files_, node != nullptr ? node->name()->string_view() : "");
Austin Schuh315b96b2020-12-11 21:21:12 -0800311
312 // Confirm that all the parts are from the same boot if there are enough
313 // parts to not be from the same boot.
314 if (filtered_parts.size() > 1u) {
315 for (size_t i = 1; i < filtered_parts.size(); ++i) {
316 CHECK_EQ(filtered_parts[i].source_boot_uuid,
317 filtered_parts[0].source_boot_uuid)
318 << ": Found parts from different boots "
319 << LogFileVectorToString(log_files_);
320 }
James Kuszmaul4f106fb2021-01-05 20:53:02 -0800321 if (!filtered_parts[0].source_boot_uuid.empty()) {
322 event_loop_factory_->GetNodeEventLoopFactory(node)->set_boot_uuid(
323 filtered_parts[0].source_boot_uuid);
324 }
Austin Schuh315b96b2020-12-11 21:21:12 -0800325 }
326
Austin Schuh287d43d2020-12-04 20:19:33 -0800327 states_[node_index] = std::make_unique<State>(
328 filtered_parts.size() == 0u
329 ? nullptr
330 : std::make_unique<TimestampMapper>(std::move(filtered_parts)));
Austin Schuh8bd96322020-02-13 21:18:22 -0800331 State *state = states_[node_index].get();
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700332 state->set_event_loop(state->SetNodeEventLoopFactory(
Austin Schuh858c9f32020-08-31 16:56:12 -0700333 event_loop_factory_->GetNodeEventLoopFactory(node)));
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700334
335 state->SetChannelCount(logged_configuration()->channels()->size());
Austin Schuhe639ea12021-01-25 13:00:22 -0800336 timestamp_mappers.emplace_back(state->timestamp_mapper());
Austin Schuhcde938c2020-02-02 17:30:07 -0800337 }
Austin Schuhe639ea12021-01-25 13:00:22 -0800338 filters_->SetTimestampMappers(std::move(timestamp_mappers));
339
340 // Note: this needs to be set before any times are pulled, or we won't observe
341 // the timestamps.
Austin Schuh87dd3832021-01-01 23:07:31 -0800342 event_loop_factory_->SetTimeConverter(filters_.get());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700343
Austin Schuh287d43d2020-12-04 20:19:33 -0800344 for (const Node *node : configuration::GetNodes(configuration())) {
345 const size_t node_index =
346 configuration::GetNodeIndex(configuration(), node);
347 State *state = states_[node_index].get();
348 for (const Node *other_node : configuration::GetNodes(configuration())) {
349 const size_t other_node_index =
350 configuration::GetNodeIndex(configuration(), other_node);
351 State *other_state = states_[other_node_index].get();
352 if (other_state != state) {
353 state->AddPeer(other_state);
354 }
355 }
356 }
357
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700358 // Register after making all the State objects so we can build references
359 // between them.
360 for (const Node *node : configuration::GetNodes(configuration())) {
361 const size_t node_index =
362 configuration::GetNodeIndex(configuration(), node);
363 State *state = states_[node_index].get();
364
365 Register(state->event_loop());
366 }
367
James Kuszmaul46d82582020-05-09 19:50:09 -0700368 if (live_nodes_ == 0) {
369 LOG(FATAL)
370 << "Don't have logs from any of the nodes in the replay config--are "
371 "you sure that the replay config matches the original config?";
372 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800373
Austin Schuh87dd3832021-01-01 23:07:31 -0800374 filters_->CheckGraph();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800375
Austin Schuh858c9f32020-08-31 16:56:12 -0700376 for (std::unique_ptr<State> &state : states_) {
377 state->SeedSortedMessages();
378 }
379
Austin Schuh2f8fd752020-09-01 22:38:28 -0700380 // We want to start the log file at the last start time of the log files
381 // from all the nodes. Compute how long each node's simulation needs to run
382 // to move time to this point.
Austin Schuh8bd96322020-02-13 21:18:22 -0800383 distributed_clock::time_point start_time = distributed_clock::min_time;
Austin Schuhcde938c2020-02-02 17:30:07 -0800384
Austin Schuh2f8fd752020-09-01 22:38:28 -0700385 // TODO(austin): We want an "OnStart" callback for each node rather than
386 // running until the last node.
387
Austin Schuh8bd96322020-02-13 21:18:22 -0800388 for (std::unique_ptr<State> &state : states_) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700389 VLOG(1) << "Start time is " << state->monotonic_start_time() << " for node "
390 << MaybeNodeName(state->event_loop()->node()) << "now "
391 << state->monotonic_now();
Austin Schuh287d43d2020-12-04 20:19:33 -0800392 if (state->monotonic_start_time() == monotonic_clock::min_time) {
393 continue;
394 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700395 // And start computing the start time on the distributed clock now that
396 // that works.
Austin Schuh858c9f32020-08-31 16:56:12 -0700397 start_time = std::max(
398 start_time, state->ToDistributedClock(state->monotonic_start_time()));
Austin Schuhcde938c2020-02-02 17:30:07 -0800399 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700400
Austin Schuh87dd3832021-01-01 23:07:31 -0800401 // TODO(austin): If a node doesn't have a start time, we might not queue
402 // enough. If this happens, we'll explode with a frozen error eventually.
403
Austin Schuh2f8fd752020-09-01 22:38:28 -0700404 CHECK_GE(start_time, distributed_clock::epoch())
405 << ": Hmm, we have a node starting before the start of time. Offset "
406 "everything.";
Austin Schuhcde938c2020-02-02 17:30:07 -0800407
Austin Schuh6f3babe2020-01-26 20:34:50 -0800408 // Forwarding is tracked per channel. If it is enabled, we want to turn it
409 // off. Otherwise messages replayed will get forwarded across to the other
Austin Schuh2f8fd752020-09-01 22:38:28 -0700410 // nodes, and also replayed on the other nodes. This may not satisfy all
411 // our users, but it'll start the discussion.
Austin Schuh6f3babe2020-01-26 20:34:50 -0800412 if (configuration::MultiNode(event_loop_factory_->configuration())) {
413 for (size_t i = 0; i < logged_configuration()->channels()->size(); ++i) {
414 const Channel *channel = logged_configuration()->channels()->Get(i);
415 const Node *node = configuration::GetNode(
416 configuration(), channel->source_node()->string_view());
417
Austin Schuh8bd96322020-02-13 21:18:22 -0800418 State *state =
419 states_[configuration::GetNodeIndex(configuration(), node)].get();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800420
421 const Channel *remapped_channel =
Austin Schuh858c9f32020-08-31 16:56:12 -0700422 RemapChannel(state->event_loop(), channel);
Austin Schuh6f3babe2020-01-26 20:34:50 -0800423
424 event_loop_factory_->DisableForwarding(remapped_channel);
425 }
Austin Schuh4c3b9702020-08-30 11:34:55 -0700426
427 // If we are replaying a log, we don't want a bunch of redundant messages
428 // from both the real message bridge and simulated message bridge.
429 event_loop_factory_->DisableStatistics();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800430 }
431
Austin Schuhcde938c2020-02-02 17:30:07 -0800432 // While we are starting the system up, we might be relying on matching data
433 // to timestamps on log files where the timestamp log file starts before the
434 // data. In this case, it is reasonable to expect missing data.
Austin Schuhdda74ec2021-01-03 19:30:37 -0800435 {
436 const bool prior_ignore_missing_data = ignore_missing_data_;
437 ignore_missing_data_ = true;
438 VLOG(1) << "Running until " << start_time << " in Register";
439 event_loop_factory_->RunFor(start_time.time_since_epoch());
440 VLOG(1) << "At start time";
441 // Now that we are running for real, missing data means that the log file is
442 // corrupted or went wrong.
443 ignore_missing_data_ = prior_ignore_missing_data;
444 }
Austin Schuh92547522019-12-28 14:33:43 -0800445
Austin Schuh8bd96322020-02-13 21:18:22 -0800446 for (std::unique_ptr<State> &state : states_) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700447 // Make the RT clock be correct before handing it to the user.
448 if (state->realtime_start_time() != realtime_clock::min_time) {
449 state->SetRealtimeOffset(state->monotonic_start_time(),
450 state->realtime_start_time());
451 }
452 VLOG(1) << "Start time is " << state->monotonic_start_time() << " for node "
453 << MaybeNodeName(state->event_loop()->node()) << "now "
454 << state->monotonic_now();
455 }
456
457 if (FLAGS_timestamps_to_csv) {
Austin Schuh0ca1fd32020-12-18 22:53:05 -0800458 filters_->Start(event_loop_factory);
Austin Schuh8bd96322020-02-13 21:18:22 -0800459 }
460}
461
Austin Schuh2f8fd752020-09-01 22:38:28 -0700462message_bridge::NoncausalOffsetEstimator *LogReader::GetFilter(
Austin Schuh8bd96322020-02-13 21:18:22 -0800463 const Node *node_a, const Node *node_b) {
Austin Schuh0ca1fd32020-12-18 22:53:05 -0800464 if (filters_) {
465 return filters_->GetFilter(node_a, node_b);
Austin Schuh8bd96322020-02-13 21:18:22 -0800466 }
Austin Schuh0ca1fd32020-12-18 22:53:05 -0800467 return nullptr;
Austin Schuh8bd96322020-02-13 21:18:22 -0800468}
469
Austin Schuhe309d2a2019-11-29 13:25:21 -0800470void LogReader::Register(EventLoop *event_loop) {
Austin Schuh8bd96322020-02-13 21:18:22 -0800471 State *state =
472 states_[configuration::GetNodeIndex(configuration(), event_loop->node())]
473 .get();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800474
Austin Schuh858c9f32020-08-31 16:56:12 -0700475 state->set_event_loop(event_loop);
Austin Schuhe309d2a2019-11-29 13:25:21 -0800476
Tyler Chatow67ddb032020-01-12 14:30:04 -0800477 // We don't run timing reports when trying to print out logged data, because
478 // otherwise we would end up printing out the timing reports themselves...
479 // This is only really relevant when we are replaying into a simulation.
Austin Schuh6f3babe2020-01-26 20:34:50 -0800480 event_loop->SkipTimingReport();
481 event_loop->SkipAosLog();
Austin Schuh39788ff2019-12-01 18:22:57 -0800482
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700483 for (size_t logged_channel_index = 0;
484 logged_channel_index < logged_configuration()->channels()->size();
485 ++logged_channel_index) {
486 const Channel *channel = RemapChannel(
487 event_loop,
488 logged_configuration()->channels()->Get(logged_channel_index));
Austin Schuh8bd96322020-02-13 21:18:22 -0800489
Austin Schuh532656d2021-01-11 10:17:18 -0800490 if (channel->logger() == LoggerConfig::NOT_LOGGED) {
491 continue;
492 }
493
Austin Schuh2f8fd752020-09-01 22:38:28 -0700494 message_bridge::NoncausalOffsetEstimator *filter = nullptr;
Austin Schuh969cd602021-01-03 00:09:45 -0800495 RemoteMessageSender *remote_timestamp_sender = nullptr;
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700496
497 State *source_state = nullptr;
Austin Schuh8bd96322020-02-13 21:18:22 -0800498
499 if (!configuration::ChannelIsSendableOnNode(channel, event_loop->node()) &&
500 configuration::ChannelIsReadableOnNode(channel, event_loop->node())) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700501 // We've got a message which is being forwarded to this node.
502 const Node *source_node = configuration::GetNode(
Austin Schuh8bd96322020-02-13 21:18:22 -0800503 event_loop->configuration(), channel->source_node()->string_view());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700504 filter = GetFilter(event_loop->node(), source_node);
Austin Schuh8bd96322020-02-13 21:18:22 -0800505
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700506 // Delivery timestamps are supposed to be logged back on the source node.
507 // Configure remote timestamps to be sent.
Austin Schuh61e973f2021-02-21 21:43:56 -0800508 const Connection *connection =
509 configuration::ConnectionToNode(channel, event_loop->node());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700510 const bool delivery_time_is_logged =
Austin Schuh61e973f2021-02-21 21:43:56 -0800511 configuration::ConnectionDeliveryTimeIsLoggedOnNode(connection,
512 source_node);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700513
514 source_state =
515 states_[configuration::GetNodeIndex(configuration(), source_node)]
516 .get();
517
518 if (delivery_time_is_logged) {
519 remote_timestamp_sender =
Austin Schuh61e973f2021-02-21 21:43:56 -0800520 source_state->RemoteTimestampSender(channel, connection);
Austin Schuh8bd96322020-02-13 21:18:22 -0800521 }
522 }
Austin Schuh858c9f32020-08-31 16:56:12 -0700523
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700524 state->SetChannel(
525 logged_channel_index,
526 configuration::ChannelIndex(event_loop->configuration(), channel),
527 event_loop->MakeRawSender(channel), filter, remote_timestamp_sender,
528 source_state);
Austin Schuhe309d2a2019-11-29 13:25:21 -0800529 }
530
Austin Schuh6aa77be2020-02-22 21:06:40 -0800531 // If we didn't find any log files with data in them, we won't ever get a
532 // callback or be live. So skip the rest of the setup.
Austin Schuh287d43d2020-12-04 20:19:33 -0800533 if (state->OldestMessageTime() == monotonic_clock::max_time) {
Austin Schuh6aa77be2020-02-22 21:06:40 -0800534 return;
535 }
536
Austin Schuh858c9f32020-08-31 16:56:12 -0700537 state->set_timer_handler(event_loop->AddTimer([this, state]() {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700538 VLOG(1) << "Starting sending " << MaybeNodeName(state->event_loop()->node())
539 << "at " << state->event_loop()->context().monotonic_event_time
540 << " now " << state->monotonic_now();
Austin Schuh858c9f32020-08-31 16:56:12 -0700541 if (state->OldestMessageTime() == monotonic_clock::max_time) {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800542 --live_nodes_;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700543 VLOG(1) << MaybeNodeName(state->event_loop()->node()) << "Node down!";
James Kuszmaul71a81932020-12-15 21:08:01 -0800544 if (exit_on_finish_ && live_nodes_ == 0) {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800545 event_loop_factory_->Exit();
546 }
James Kuszmaul314f1672020-01-03 20:02:08 -0800547 return;
548 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700549
Austin Schuhdda74ec2021-01-03 19:30:37 -0800550 TimestampedMessage timestamped_message = state->PopOldest();
Austin Schuh05b70472020-01-01 17:11:17 -0800551
Austin Schuhe309d2a2019-11-29 13:25:21 -0800552 const monotonic_clock::time_point monotonic_now =
Austin Schuh858c9f32020-08-31 16:56:12 -0700553 state->event_loop()->context().monotonic_event_time;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700554 if (!FLAGS_skip_order_validation) {
Austin Schuh287d43d2020-12-04 20:19:33 -0800555 CHECK(monotonic_now == timestamped_message.monotonic_event_time)
Austin Schuh2f8fd752020-09-01 22:38:28 -0700556 << ": " << FlatbufferToJson(state->event_loop()->node()) << " Now "
557 << monotonic_now << " trying to send "
Austin Schuh287d43d2020-12-04 20:19:33 -0800558 << timestamped_message.monotonic_event_time << " failure "
Austin Schuh2f8fd752020-09-01 22:38:28 -0700559 << state->DebugString();
Austin Schuh287d43d2020-12-04 20:19:33 -0800560 } else if (monotonic_now != timestamped_message.monotonic_event_time) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700561 LOG(WARNING) << "Check failed: monotonic_now == "
Austin Schuh287d43d2020-12-04 20:19:33 -0800562 "timestamped_message.monotonic_event_time) ("
Austin Schuh2f8fd752020-09-01 22:38:28 -0700563 << monotonic_now << " vs. "
Austin Schuh287d43d2020-12-04 20:19:33 -0800564 << timestamped_message.monotonic_event_time
Austin Schuh2f8fd752020-09-01 22:38:28 -0700565 << "): " << FlatbufferToJson(state->event_loop()->node())
566 << " Now " << monotonic_now << " trying to send "
Austin Schuh287d43d2020-12-04 20:19:33 -0800567 << timestamped_message.monotonic_event_time << " failure "
Austin Schuh2f8fd752020-09-01 22:38:28 -0700568 << state->DebugString();
569 }
Austin Schuhe309d2a2019-11-29 13:25:21 -0800570
Austin Schuh287d43d2020-12-04 20:19:33 -0800571 if (timestamped_message.monotonic_event_time >
Austin Schuh858c9f32020-08-31 16:56:12 -0700572 state->monotonic_start_time() ||
Austin Schuh15649d62019-12-28 16:36:38 -0800573 event_loop_factory_ != nullptr) {
Austin Schuhdda74ec2021-01-03 19:30:37 -0800574 if (timestamped_message.data.span().size() != 0u) {
575 if (timestamped_message.monotonic_remote_time !=
576 monotonic_clock::min_time) {
Austin Schuh8bd96322020-02-13 21:18:22 -0800577 // Confirm that the message was sent on the sending node before the
578 // destination node (this node). As a proxy, do this by making sure
579 // that time on the source node is past when the message was sent.
Austin Schuh87dd3832021-01-01 23:07:31 -0800580 //
581 // TODO(austin): <= means that the cause message (which we know) could
582 // happen after the effect even though we know they are at the same
583 // time. I doubt anyone will notice for a bit, but we should really
584 // fix that.
Austin Schuh2f8fd752020-09-01 22:38:28 -0700585 if (!FLAGS_skip_order_validation) {
Austin Schuh87dd3832021-01-01 23:07:31 -0800586 CHECK_LE(
Austin Schuh287d43d2020-12-04 20:19:33 -0800587 timestamped_message.monotonic_remote_time,
588 state->monotonic_remote_now(timestamped_message.channel_index))
Austin Schuh2f8fd752020-09-01 22:38:28 -0700589 << state->event_loop()->node()->name()->string_view() << " to "
Austin Schuh287d43d2020-12-04 20:19:33 -0800590 << state->remote_node(timestamped_message.channel_index)
591 ->name()
592 ->string_view()
Austin Schuh315b96b2020-12-11 21:21:12 -0800593 << " while trying to send a message on "
594 << configuration::CleanedChannelToString(
595 logged_configuration()->channels()->Get(
596 timestamped_message.channel_index))
Austin Schuh2f8fd752020-09-01 22:38:28 -0700597 << " " << state->DebugString();
Austin Schuh87dd3832021-01-01 23:07:31 -0800598 } else if (timestamped_message.monotonic_remote_time >
Austin Schuh287d43d2020-12-04 20:19:33 -0800599 state->monotonic_remote_now(
600 timestamped_message.channel_index)) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700601 LOG(WARNING)
Austin Schuh287d43d2020-12-04 20:19:33 -0800602 << "Check failed: timestamped_message.monotonic_remote_time < "
603 "state->monotonic_remote_now(timestamped_message.channel_"
604 "index) ("
605 << timestamped_message.monotonic_remote_time << " vs. "
606 << state->monotonic_remote_now(
607 timestamped_message.channel_index)
608 << ") " << state->event_loop()->node()->name()->string_view()
609 << " to "
610 << state->remote_node(timestamped_message.channel_index)
611 ->name()
612 ->string_view()
613 << " currently " << timestamped_message.monotonic_event_time
Austin Schuh2f8fd752020-09-01 22:38:28 -0700614 << " ("
615 << state->ToDistributedClock(
Austin Schuh287d43d2020-12-04 20:19:33 -0800616 timestamped_message.monotonic_event_time)
Austin Schuh2f8fd752020-09-01 22:38:28 -0700617 << ") remote event time "
Austin Schuh287d43d2020-12-04 20:19:33 -0800618 << timestamped_message.monotonic_remote_time << " ("
Austin Schuh2f8fd752020-09-01 22:38:28 -0700619 << state->RemoteToDistributedClock(
Austin Schuh287d43d2020-12-04 20:19:33 -0800620 timestamped_message.channel_index,
621 timestamped_message.monotonic_remote_time)
Austin Schuh2f8fd752020-09-01 22:38:28 -0700622 << ") " << state->DebugString();
623 }
Austin Schuh8bd96322020-02-13 21:18:22 -0800624 }
625
Austin Schuh15649d62019-12-28 16:36:38 -0800626 // If we have access to the factory, use it to fix the realtime time.
Austin Schuh287d43d2020-12-04 20:19:33 -0800627 state->SetRealtimeOffset(timestamped_message.monotonic_event_time,
628 timestamped_message.realtime_event_time);
Austin Schuh15649d62019-12-28 16:36:38 -0800629
Austin Schuh2f8fd752020-09-01 22:38:28 -0700630 VLOG(1) << MaybeNodeName(state->event_loop()->node()) << "Sending "
Austin Schuh287d43d2020-12-04 20:19:33 -0800631 << timestamped_message.monotonic_event_time;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700632 // TODO(austin): std::move channel_data in and make that efficient in
633 // simulation.
Austin Schuh287d43d2020-12-04 20:19:33 -0800634 state->Send(std::move(timestamped_message));
Austin Schuhdda74ec2021-01-03 19:30:37 -0800635 } else if (!ignore_missing_data_ &&
Austin Schuh5ee56872021-01-30 16:53:34 -0800636 // When starting up, we can have data which was sent before the
637 // log starts, but the timestamp was after the log starts. This
638 // is unreasonable to avoid, so ignore the missing data.
639 timestamped_message.monotonic_remote_time >=
640 state->monotonic_remote_start_time(
641 timestamped_message.channel_index) &&
Austin Schuhdda74ec2021-01-03 19:30:37 -0800642 !FLAGS_skip_missing_forwarding_entries) {
Austin Schuh5ee56872021-01-30 16:53:34 -0800643 // We've found a timestamp without data that we expect to have data for.
644 // This likely means that we are at the end of the log file. Record it
645 // and CHECK that in the rest of the log file, we don't find any more
646 // data on that channel. Not all channels will end at the same point in
647 // time since they can be in different files.
Austin Schuhdda74ec2021-01-03 19:30:37 -0800648 VLOG(1) << "Found the last message on channel "
649 << timestamped_message.channel_index;
650
651 // Vector storing if we've seen a nullptr message or not per channel.
652 std::vector<bool> last_message;
653 last_message.resize(logged_configuration()->channels()->size(), false);
654
655 last_message[timestamped_message.channel_index] = true;
656
657 // Now that we found the end of one channel, artificially stop the
658 // rest. It is confusing when part of your data gets replayed but not
Austin Schuh5ee56872021-01-30 16:53:34 -0800659 // all. Read the rest of the messages and drop them on the floor while
660 // doing some basic validation.
Austin Schuh858c9f32020-08-31 16:56:12 -0700661 while (state->OldestMessageTime() != monotonic_clock::max_time) {
Austin Schuhdda74ec2021-01-03 19:30:37 -0800662 TimestampedMessage next = state->PopOldest();
663 // Make sure that once we have seen the last message on a channel,
664 // data doesn't start back up again. If the user wants to play
665 // through events like this, they can set
666 // --skip_missing_forwarding_entries or ignore_missing_data_.
667 CHECK_LT(next.channel_index, last_message.size());
668 if (next.data.span().size() == 0u) {
669 last_message[next.channel_index] = true;
670 } else {
671 if (last_message[next.channel_index]) {
672 LOG(FATAL)
673 << "Found missing data in the middle of the log file on "
674 "channel "
675 << next.channel_index << " Last "
676 << last_message[next.channel_index] << state->DebugString();
677 }
678 }
Austin Schuh8bd96322020-02-13 21:18:22 -0800679 }
Austin Schuh92547522019-12-28 14:33:43 -0800680 }
Austin Schuhe309d2a2019-11-29 13:25:21 -0800681 } else {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800682 LOG(WARNING)
683 << "Not sending data from before the start of the log file. "
Austin Schuh287d43d2020-12-04 20:19:33 -0800684 << timestamped_message.monotonic_event_time.time_since_epoch().count()
Austin Schuh6f3babe2020-01-26 20:34:50 -0800685 << " start " << monotonic_start_time().time_since_epoch().count()
Austin Schuhd85baf82020-10-19 11:50:12 -0700686 << " "
Austin Schuh287d43d2020-12-04 20:19:33 -0800687 << FlatbufferToJson(timestamped_message.data,
Austin Schuhd85baf82020-10-19 11:50:12 -0700688 {.multi_line = false, .max_vector_size = 100});
Austin Schuhe309d2a2019-11-29 13:25:21 -0800689 }
690
Austin Schuh858c9f32020-08-31 16:56:12 -0700691 const monotonic_clock::time_point next_time = state->OldestMessageTime();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800692 if (next_time != monotonic_clock::max_time) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700693 VLOG(1) << "Scheduling " << MaybeNodeName(state->event_loop()->node())
694 << "wakeup for " << next_time << "("
695 << state->ToDistributedClock(next_time)
696 << " distributed), now is " << state->monotonic_now();
Austin Schuh858c9f32020-08-31 16:56:12 -0700697 state->Setup(next_time);
James Kuszmaul314f1672020-01-03 20:02:08 -0800698 } else {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700699 VLOG(1) << MaybeNodeName(state->event_loop()->node())
700 << "No next message, scheduling shutdown";
701 // Set a timer up immediately after now to die. If we don't do this,
702 // then the senders waiting on the message we just read will never get
703 // called.
Austin Schuheecb9282020-01-08 17:43:30 -0800704 if (event_loop_factory_ != nullptr) {
Austin Schuh858c9f32020-08-31 16:56:12 -0700705 state->Setup(monotonic_now + event_loop_factory_->send_delay() +
706 std::chrono::nanoseconds(1));
Austin Schuheecb9282020-01-08 17:43:30 -0800707 }
Austin Schuhe309d2a2019-11-29 13:25:21 -0800708 }
Austin Schuh8bd96322020-02-13 21:18:22 -0800709
Austin Schuh2f8fd752020-09-01 22:38:28 -0700710 VLOG(1) << MaybeNodeName(state->event_loop()->node()) << "Done sending at "
711 << state->event_loop()->context().monotonic_event_time << " now "
712 << state->monotonic_now();
Austin Schuh858c9f32020-08-31 16:56:12 -0700713 }));
Austin Schuhe309d2a2019-11-29 13:25:21 -0800714
Austin Schuh6f3babe2020-01-26 20:34:50 -0800715 ++live_nodes_;
716
Austin Schuh858c9f32020-08-31 16:56:12 -0700717 if (state->OldestMessageTime() != monotonic_clock::max_time) {
718 event_loop->OnRun([state]() { state->Setup(state->OldestMessageTime()); });
Austin Schuhe309d2a2019-11-29 13:25:21 -0800719 }
720}
721
722void LogReader::Deregister() {
James Kuszmaul84ff3e52020-01-03 19:48:53 -0800723 // Make sure that things get destroyed in the correct order, rather than
724 // relying on getting the order correct in the class definition.
Austin Schuh8bd96322020-02-13 21:18:22 -0800725 for (std::unique_ptr<State> &state : states_) {
Austin Schuh858c9f32020-08-31 16:56:12 -0700726 state->Deregister();
Austin Schuhe309d2a2019-11-29 13:25:21 -0800727 }
Austin Schuh92547522019-12-28 14:33:43 -0800728
James Kuszmaul84ff3e52020-01-03 19:48:53 -0800729 event_loop_factory_unique_ptr_.reset();
730 event_loop_factory_ = nullptr;
Austin Schuhe309d2a2019-11-29 13:25:21 -0800731}
732
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800733void LogReader::RemapLoggedChannel(std::string_view name, std::string_view type,
Austin Schuh0de30f32020-12-06 12:44:28 -0800734 std::string_view add_prefix,
735 std::string_view new_type) {
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800736 for (size_t ii = 0; ii < logged_configuration()->channels()->size(); ++ii) {
737 const Channel *const channel = logged_configuration()->channels()->Get(ii);
738 if (channel->name()->str() == name &&
739 channel->type()->string_view() == type) {
740 CHECK_EQ(0u, remapped_channels_.count(ii))
741 << "Already remapped channel "
742 << configuration::CleanedChannelToString(channel);
Austin Schuh0de30f32020-12-06 12:44:28 -0800743 RemappedChannel remapped_channel;
744 remapped_channel.remapped_name =
745 std::string(add_prefix) + std::string(name);
746 remapped_channel.new_type = new_type;
747 remapped_channels_[ii] = std::move(remapped_channel);
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800748 VLOG(1) << "Remapping channel "
749 << configuration::CleanedChannelToString(channel)
Austin Schuh0de30f32020-12-06 12:44:28 -0800750 << " to have name " << remapped_channels_[ii].remapped_name;
Austin Schuh6331ef92020-01-07 18:28:09 -0800751 MakeRemappedConfig();
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800752 return;
753 }
754 }
755 LOG(FATAL) << "Unabled to locate channel with name " << name << " and type "
756 << type;
757}
758
Austin Schuh01b4c352020-09-21 23:09:39 -0700759void LogReader::RemapLoggedChannel(std::string_view name, std::string_view type,
760 const Node *node,
Austin Schuh0de30f32020-12-06 12:44:28 -0800761 std::string_view add_prefix,
762 std::string_view new_type) {
Austin Schuh01b4c352020-09-21 23:09:39 -0700763 VLOG(1) << "Node is " << aos::FlatbufferToJson(node);
764 const Channel *remapped_channel =
765 configuration::GetChannel(logged_configuration(), name, type, "", node);
766 CHECK(remapped_channel != nullptr) << ": Failed to find {\"name\": \"" << name
767 << "\", \"type\": \"" << type << "\"}";
768 VLOG(1) << "Original {\"name\": \"" << name << "\", \"type\": \"" << type
769 << "\"}";
770 VLOG(1) << "Remapped "
771 << aos::configuration::StrippedChannelToString(remapped_channel);
772
773 // We want to make /spray on node 0 go to /0/spray by snooping the maps. And
774 // we want it to degrade if the heuristics fail to just work.
775 //
776 // The easiest way to do this is going to be incredibly specific and verbose.
777 // Look up /spray, to /0/spray. Then, prefix the result with /original to get
778 // /original/0/spray. Then, create a map from /original/spray to
779 // /original/0/spray for just the type we were asked for.
780 if (name != remapped_channel->name()->string_view()) {
781 MapT new_map;
782 new_map.match = std::make_unique<ChannelT>();
783 new_map.match->name = absl::StrCat(add_prefix, name);
784 new_map.match->type = type;
785 if (node != nullptr) {
786 new_map.match->source_node = node->name()->str();
787 }
788 new_map.rename = std::make_unique<ChannelT>();
789 new_map.rename->name =
790 absl::StrCat(add_prefix, remapped_channel->name()->string_view());
791 maps_.emplace_back(std::move(new_map));
792 }
793
794 const size_t channel_index =
795 configuration::ChannelIndex(logged_configuration(), remapped_channel);
796 CHECK_EQ(0u, remapped_channels_.count(channel_index))
797 << "Already remapped channel "
798 << configuration::CleanedChannelToString(remapped_channel);
Austin Schuh0de30f32020-12-06 12:44:28 -0800799
800 RemappedChannel remapped_channel_struct;
801 remapped_channel_struct.remapped_name =
802 std::string(add_prefix) +
803 std::string(remapped_channel->name()->string_view());
804 remapped_channel_struct.new_type = new_type;
805 remapped_channels_[channel_index] = std::move(remapped_channel_struct);
Austin Schuh01b4c352020-09-21 23:09:39 -0700806 MakeRemappedConfig();
807}
808
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800809void LogReader::MakeRemappedConfig() {
Austin Schuh8bd96322020-02-13 21:18:22 -0800810 for (std::unique_ptr<State> &state : states_) {
Austin Schuh6aa77be2020-02-22 21:06:40 -0800811 if (state) {
Austin Schuh858c9f32020-08-31 16:56:12 -0700812 CHECK(!state->event_loop())
Austin Schuh6aa77be2020-02-22 21:06:40 -0800813 << ": Can't change the mapping after the events are scheduled.";
814 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800815 }
Austin Schuhac0771c2020-01-07 18:36:30 -0800816
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800817 // If no remapping occurred and we are using the original config, then there
818 // is nothing interesting to do here.
819 if (remapped_channels_.empty() && replay_configuration_ == nullptr) {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800820 remapped_configuration_ = logged_configuration();
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800821 return;
822 }
823 // Config to copy Channel definitions from. Use the specified
824 // replay_configuration_ if it has been provided.
825 const Configuration *const base_config = replay_configuration_ == nullptr
826 ? logged_configuration()
827 : replay_configuration_;
Austin Schuh0de30f32020-12-06 12:44:28 -0800828
829 // Create a config with all the channels, but un-sorted/merged. Collect up
830 // the schemas while we do this. Call MergeConfiguration to sort everything,
831 // and then merge it all in together.
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800832
833 // This is the builder that we use for the config containing all the new
834 // channels.
Austin Schuh0de30f32020-12-06 12:44:28 -0800835 flatbuffers::FlatBufferBuilder fbb;
836 fbb.ForceDefaults(true);
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800837 std::vector<flatbuffers::Offset<Channel>> channel_offsets;
Austin Schuh0de30f32020-12-06 12:44:28 -0800838
839 CHECK_EQ(Channel::MiniReflectTypeTable()->num_elems, 13u)
840 << ": Merging logic needs to be updated when the number of channel "
841 "fields changes.";
842
843 // List of schemas.
844 std::map<std::string_view, FlatbufferVector<reflection::Schema>> schema_map;
845 // Make sure our new RemoteMessage schema is in there for old logs without it.
846 schema_map.insert(std::make_pair(
847 RemoteMessage::GetFullyQualifiedName(),
848 FlatbufferVector<reflection::Schema>(FlatbufferSpan<reflection::Schema>(
849 message_bridge::RemoteMessageSchema()))));
850
851 // Reconstruct the remapped channels.
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800852 for (auto &pair : remapped_channels_) {
Austin Schuh0de30f32020-12-06 12:44:28 -0800853 const Channel *const c = CHECK_NOTNULL(configuration::GetChannel(
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800854 base_config, logged_configuration()->channels()->Get(pair.first), "",
855 nullptr));
Austin Schuh0de30f32020-12-06 12:44:28 -0800856 channel_offsets.emplace_back(
857 CopyChannel(c, pair.second.remapped_name, "", &fbb));
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800858 }
Austin Schuh01b4c352020-09-21 23:09:39 -0700859
Austin Schuh0de30f32020-12-06 12:44:28 -0800860 // Now reconstruct the original channels, translating types as needed
861 for (const Channel *c : *base_config->channels()) {
862 // Search for a mapping channel.
863 std::string_view new_type = "";
864 for (auto &pair : remapped_channels_) {
865 const Channel *const remapped_channel =
866 logged_configuration()->channels()->Get(pair.first);
867 if (remapped_channel->name()->string_view() == c->name()->string_view() &&
868 remapped_channel->type()->string_view() == c->type()->string_view()) {
869 new_type = pair.second.new_type;
870 break;
871 }
872 }
873
874 // Copy everything over.
875 channel_offsets.emplace_back(CopyChannel(c, "", new_type, &fbb));
876
877 // Add the schema if it doesn't exist.
878 if (schema_map.find(c->type()->string_view()) == schema_map.end()) {
879 CHECK(c->has_schema());
880 schema_map.insert(std::make_pair(c->type()->string_view(),
881 RecursiveCopyFlatBuffer(c->schema())));
882 }
883 }
884
885 // The MergeConfiguration API takes a vector, not a map. Convert.
886 std::vector<FlatbufferVector<reflection::Schema>> schemas;
887 while (!schema_map.empty()) {
888 schemas.emplace_back(std::move(schema_map.begin()->second));
889 schema_map.erase(schema_map.begin());
890 }
891
892 // Create the Configuration containing the new channels that we want to add.
893 const flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Channel>>>
894 channels_offset =
895 channel_offsets.empty() ? 0 : fbb.CreateVector(channel_offsets);
896
897 // Copy over the old maps.
Austin Schuh01b4c352020-09-21 23:09:39 -0700898 std::vector<flatbuffers::Offset<Map>> map_offsets;
Austin Schuh0de30f32020-12-06 12:44:28 -0800899 if (base_config->maps()) {
900 for (const Map *map : *base_config->maps()) {
901 map_offsets.emplace_back(aos::RecursiveCopyFlatBuffer(map, &fbb));
902 }
903 }
904
905 // Now create the new maps. These are second so they take effect first.
Austin Schuh01b4c352020-09-21 23:09:39 -0700906 for (const MapT &map : maps_) {
907 const flatbuffers::Offset<flatbuffers::String> match_name_offset =
Austin Schuh0de30f32020-12-06 12:44:28 -0800908 fbb.CreateString(map.match->name);
Austin Schuh01b4c352020-09-21 23:09:39 -0700909 const flatbuffers::Offset<flatbuffers::String> match_type_offset =
Austin Schuh0de30f32020-12-06 12:44:28 -0800910 fbb.CreateString(map.match->type);
Austin Schuh01b4c352020-09-21 23:09:39 -0700911 const flatbuffers::Offset<flatbuffers::String> rename_name_offset =
Austin Schuh0de30f32020-12-06 12:44:28 -0800912 fbb.CreateString(map.rename->name);
Austin Schuh01b4c352020-09-21 23:09:39 -0700913 flatbuffers::Offset<flatbuffers::String> match_source_node_offset;
914 if (!map.match->source_node.empty()) {
Austin Schuh0de30f32020-12-06 12:44:28 -0800915 match_source_node_offset = fbb.CreateString(map.match->source_node);
Austin Schuh01b4c352020-09-21 23:09:39 -0700916 }
Austin Schuh0de30f32020-12-06 12:44:28 -0800917 Channel::Builder match_builder(fbb);
Austin Schuh01b4c352020-09-21 23:09:39 -0700918 match_builder.add_name(match_name_offset);
919 match_builder.add_type(match_type_offset);
920 if (!map.match->source_node.empty()) {
921 match_builder.add_source_node(match_source_node_offset);
922 }
923 const flatbuffers::Offset<Channel> match_offset = match_builder.Finish();
924
Austin Schuh0de30f32020-12-06 12:44:28 -0800925 Channel::Builder rename_builder(fbb);
Austin Schuh01b4c352020-09-21 23:09:39 -0700926 rename_builder.add_name(rename_name_offset);
927 const flatbuffers::Offset<Channel> rename_offset = rename_builder.Finish();
928
Austin Schuh0de30f32020-12-06 12:44:28 -0800929 Map::Builder map_builder(fbb);
Austin Schuh01b4c352020-09-21 23:09:39 -0700930 map_builder.add_match(match_offset);
931 map_builder.add_rename(rename_offset);
932 map_offsets.emplace_back(map_builder.Finish());
933 }
934
Austin Schuh0de30f32020-12-06 12:44:28 -0800935 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Map>>>
936 maps_offsets = map_offsets.empty() ? 0 : fbb.CreateVector(map_offsets);
Austin Schuh01b4c352020-09-21 23:09:39 -0700937
Austin Schuh0de30f32020-12-06 12:44:28 -0800938 // And copy everything else over.
939 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Node>>>
940 nodes_offset = aos::RecursiveCopyVectorTable(base_config->nodes(), &fbb);
941
942 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Application>>>
943 applications_offset =
944 aos::RecursiveCopyVectorTable(base_config->applications(), &fbb);
945
946 // Now insert everything else in unmodified.
947 ConfigurationBuilder configuration_builder(fbb);
948 if (!channels_offset.IsNull()) {
949 configuration_builder.add_channels(channels_offset);
950 }
951 if (!maps_offsets.IsNull()) {
952 configuration_builder.add_maps(maps_offsets);
953 }
954 if (!nodes_offset.IsNull()) {
955 configuration_builder.add_nodes(nodes_offset);
956 }
957 if (!applications_offset.IsNull()) {
958 configuration_builder.add_applications(applications_offset);
959 }
960
961 if (base_config->has_channel_storage_duration()) {
962 configuration_builder.add_channel_storage_duration(
963 base_config->channel_storage_duration());
964 }
965
966 CHECK_EQ(Configuration::MiniReflectTypeTable()->num_elems, 6u)
967 << ": Merging logic needs to be updated when the number of configuration "
968 "fields changes.";
969
970 fbb.Finish(configuration_builder.Finish());
971
972 // Clean it up and return it! By using MergeConfiguration here, we'll
973 // actually get a deduplicated config for free too.
974 FlatbufferDetachedBuffer<Configuration> new_merged_config =
975 configuration::MergeConfiguration(
976 FlatbufferDetachedBuffer<Configuration>(fbb.Release()));
977
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800978 remapped_configuration_buffer_ =
979 std::make_unique<FlatbufferDetachedBuffer<Configuration>>(
Austin Schuh0de30f32020-12-06 12:44:28 -0800980 configuration::MergeConfiguration(new_merged_config, schemas));
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800981
982 remapped_configuration_ = &remapped_configuration_buffer_->message();
Austin Schuh0de30f32020-12-06 12:44:28 -0800983
984 // TODO(austin): Lazily re-build to save CPU?
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800985}
986
Austin Schuh6f3babe2020-01-26 20:34:50 -0800987const Channel *LogReader::RemapChannel(const EventLoop *event_loop,
988 const Channel *channel) {
989 std::string_view channel_name = channel->name()->string_view();
990 std::string_view channel_type = channel->type()->string_view();
991 const int channel_index =
992 configuration::ChannelIndex(logged_configuration(), channel);
993 // If the channel is remapped, find the correct channel name to use.
994 if (remapped_channels_.count(channel_index) > 0) {
Austin Schuhee711052020-08-24 16:06:09 -0700995 VLOG(3) << "Got remapped channel on "
Austin Schuh6f3babe2020-01-26 20:34:50 -0800996 << configuration::CleanedChannelToString(channel);
Austin Schuh0de30f32020-12-06 12:44:28 -0800997 channel_name = remapped_channels_[channel_index].remapped_name;
Austin Schuh6f3babe2020-01-26 20:34:50 -0800998 }
999
Austin Schuhee711052020-08-24 16:06:09 -07001000 VLOG(2) << "Going to remap channel " << channel_name << " " << channel_type;
Austin Schuh6f3babe2020-01-26 20:34:50 -08001001 const Channel *remapped_channel = configuration::GetChannel(
1002 event_loop->configuration(), channel_name, channel_type,
1003 event_loop->name(), event_loop->node());
1004
1005 CHECK(remapped_channel != nullptr)
1006 << ": Unable to send {\"name\": \"" << channel_name << "\", \"type\": \""
1007 << channel_type << "\"} because it is not in the provided configuration.";
1008
1009 return remapped_channel;
1010}
1011
Austin Schuh287d43d2020-12-04 20:19:33 -08001012LogReader::State::State(std::unique_ptr<TimestampMapper> timestamp_mapper)
1013 : timestamp_mapper_(std::move(timestamp_mapper)) {}
1014
1015void LogReader::State::AddPeer(State *peer) {
1016 if (timestamp_mapper_ && peer->timestamp_mapper_) {
1017 timestamp_mapper_->AddPeer(peer->timestamp_mapper_.get());
1018 }
1019}
Austin Schuh858c9f32020-08-31 16:56:12 -07001020
1021EventLoop *LogReader::State::SetNodeEventLoopFactory(
1022 NodeEventLoopFactory *node_event_loop_factory) {
1023 node_event_loop_factory_ = node_event_loop_factory;
1024 event_loop_unique_ptr_ =
1025 node_event_loop_factory_->MakeEventLoop("log_reader");
1026 return event_loop_unique_ptr_.get();
1027}
1028
1029void LogReader::State::SetChannelCount(size_t count) {
1030 channels_.resize(count);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001031 remote_timestamp_senders_.resize(count);
Austin Schuh858c9f32020-08-31 16:56:12 -07001032 filters_.resize(count);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001033 channel_source_state_.resize(count);
1034 factory_channel_index_.resize(count);
1035 queue_index_map_.resize(count);
Austin Schuh858c9f32020-08-31 16:56:12 -07001036}
1037
1038void LogReader::State::SetChannel(
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001039 size_t logged_channel_index, size_t factory_channel_index,
1040 std::unique_ptr<RawSender> sender,
Austin Schuh2f8fd752020-09-01 22:38:28 -07001041 message_bridge::NoncausalOffsetEstimator *filter,
Austin Schuh969cd602021-01-03 00:09:45 -08001042 RemoteMessageSender *remote_timestamp_sender, State *source_state) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001043 channels_[logged_channel_index] = std::move(sender);
1044 filters_[logged_channel_index] = filter;
1045 remote_timestamp_senders_[logged_channel_index] = remote_timestamp_sender;
1046
1047 if (source_state) {
1048 channel_source_state_[logged_channel_index] = source_state;
1049
1050 if (remote_timestamp_sender != nullptr) {
1051 source_state->queue_index_map_[logged_channel_index] =
Austin Schuh9942bae2021-01-07 22:06:44 -08001052 std::make_unique<std::vector<State::ContiguousSentTimestamp>>();
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001053 }
1054 }
1055
1056 factory_channel_index_[logged_channel_index] = factory_channel_index;
1057}
1058
Austin Schuh287d43d2020-12-04 20:19:33 -08001059bool LogReader::State::Send(const TimestampedMessage &timestamped_message) {
1060 aos::RawSender *sender = channels_[timestamped_message.channel_index].get();
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001061 uint32_t remote_queue_index = 0xffffffff;
1062
Austin Schuh287d43d2020-12-04 20:19:33 -08001063 if (remote_timestamp_senders_[timestamped_message.channel_index] != nullptr) {
Austin Schuh9942bae2021-01-07 22:06:44 -08001064 std::vector<ContiguousSentTimestamp> *queue_index_map = CHECK_NOTNULL(
Austin Schuh287d43d2020-12-04 20:19:33 -08001065 CHECK_NOTNULL(channel_source_state_[timestamped_message.channel_index])
1066 ->queue_index_map_[timestamped_message.channel_index]
1067 .get());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001068
Austin Schuh9942bae2021-01-07 22:06:44 -08001069 struct SentTimestamp {
1070 monotonic_clock::time_point monotonic_event_time;
1071 uint32_t queue_index;
1072 } search;
1073
Austin Schuh287d43d2020-12-04 20:19:33 -08001074 search.monotonic_event_time = timestamped_message.monotonic_remote_time;
Austin Schuh287d43d2020-12-04 20:19:33 -08001075 search.queue_index = timestamped_message.remote_queue_index;
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001076
1077 // Find the sent time if available.
1078 auto element = std::lower_bound(
1079 queue_index_map->begin(), queue_index_map->end(), search,
Austin Schuh9942bae2021-01-07 22:06:44 -08001080 [](ContiguousSentTimestamp a, SentTimestamp b) {
1081 if (a.ending_monotonic_event_time < b.monotonic_event_time) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001082 return true;
1083 }
Austin Schuh9942bae2021-01-07 22:06:44 -08001084 if (a.starting_monotonic_event_time > b.monotonic_event_time) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001085 return false;
1086 }
Austin Schuh9942bae2021-01-07 22:06:44 -08001087
1088 if (a.ending_queue_index < b.queue_index) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001089 return true;
1090 }
Austin Schuh9942bae2021-01-07 22:06:44 -08001091 if (a.starting_queue_index >= b.queue_index) {
1092 return false;
1093 }
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001094
Austin Schuh9942bae2021-01-07 22:06:44 -08001095 // If it isn't clearly below or above, it is below. Since we return
1096 // the last element <, this will return a match.
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001097 return false;
1098 });
1099
1100 // TODO(austin): Be a bit more principled here, but we will want to do that
1101 // after the logger rewrite. We hit this when one node finishes, but the
1102 // other node isn't done yet. So there is no send time, but there is a
1103 // receive time.
1104 if (element != queue_index_map->end()) {
Austin Schuh9942bae2021-01-07 22:06:44 -08001105 CHECK_GE(timestamped_message.monotonic_remote_time,
1106 element->starting_monotonic_event_time);
1107 CHECK_LE(timestamped_message.monotonic_remote_time,
1108 element->ending_monotonic_event_time);
1109 CHECK_GE(timestamped_message.remote_queue_index,
1110 element->starting_queue_index);
1111 CHECK_LE(timestamped_message.remote_queue_index,
1112 element->ending_queue_index);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001113
Austin Schuh9942bae2021-01-07 22:06:44 -08001114 remote_queue_index = timestamped_message.remote_queue_index +
1115 element->actual_queue_index -
1116 element->starting_queue_index;
1117 } else {
1118 VLOG(1) << "No timestamp match in the map.";
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001119 }
1120 }
1121
1122 // Send! Use the replayed queue index here instead of the logged queue index
1123 // for the remote queue index. This makes re-logging work.
Austin Schuh287d43d2020-12-04 20:19:33 -08001124 const bool sent = sender->Send(
1125 timestamped_message.data.message().data()->Data(),
1126 timestamped_message.data.message().data()->size(),
1127 timestamped_message.monotonic_remote_time,
1128 timestamped_message.realtime_remote_time, remote_queue_index);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001129 if (!sent) return false;
1130
Austin Schuh287d43d2020-12-04 20:19:33 -08001131 if (queue_index_map_[timestamped_message.channel_index]) {
Austin Schuh9942bae2021-01-07 22:06:44 -08001132 if (queue_index_map_[timestamped_message.channel_index]->empty()) {
1133 // Nothing here, start a range with 0 length.
1134 ContiguousSentTimestamp timestamp;
1135 timestamp.starting_monotonic_event_time =
1136 timestamp.ending_monotonic_event_time =
1137 timestamped_message.monotonic_event_time;
1138 timestamp.starting_queue_index = timestamp.ending_queue_index =
1139 timestamped_message.queue_index;
1140 timestamp.actual_queue_index = sender->sent_queue_index();
1141 queue_index_map_[timestamped_message.channel_index]->emplace_back(
1142 timestamp);
1143 } else {
1144 // We've got something. See if the next timestamp is still contiguous. If
1145 // so, grow it.
1146 ContiguousSentTimestamp *back =
1147 &queue_index_map_[timestamped_message.channel_index]->back();
1148 if ((back->starting_queue_index - back->actual_queue_index) ==
1149 (timestamped_message.queue_index - sender->sent_queue_index())) {
1150 back->ending_queue_index = timestamped_message.queue_index;
1151 back->ending_monotonic_event_time =
1152 timestamped_message.monotonic_event_time;
1153 } else {
1154 // Otherwise, make a new one.
1155 ContiguousSentTimestamp timestamp;
1156 timestamp.starting_monotonic_event_time =
1157 timestamp.ending_monotonic_event_time =
1158 timestamped_message.monotonic_event_time;
1159 timestamp.starting_queue_index = timestamp.ending_queue_index =
1160 timestamped_message.queue_index;
1161 timestamp.actual_queue_index = sender->sent_queue_index();
1162 queue_index_map_[timestamped_message.channel_index]->emplace_back(
1163 timestamp);
1164 }
1165 }
1166
1167 // TODO(austin): Should we prune the map? On a many day log, I only saw the
1168 // queue index diverge a couple of elements, which would be a very small
1169 // map.
Austin Schuh287d43d2020-12-04 20:19:33 -08001170 } else if (remote_timestamp_senders_[timestamped_message.channel_index] !=
1171 nullptr) {
Austin Schuh969cd602021-01-03 00:09:45 -08001172 flatbuffers::FlatBufferBuilder fbb;
1173 fbb.ForceDefaults(true);
Austin Schuh315b96b2020-12-11 21:21:12 -08001174 flatbuffers::Offset<flatbuffers::String> boot_uuid_offset =
Austin Schuh969cd602021-01-03 00:09:45 -08001175 fbb.CreateString(event_loop_->boot_uuid().string_view());
Austin Schuh315b96b2020-12-11 21:21:12 -08001176
Austin Schuh969cd602021-01-03 00:09:45 -08001177 RemoteMessage::Builder message_header_builder(fbb);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001178
1179 message_header_builder.add_channel_index(
Austin Schuh287d43d2020-12-04 20:19:33 -08001180 factory_channel_index_[timestamped_message.channel_index]);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001181
1182 // Swap the remote and sent metrics. They are from the sender's
1183 // perspective, not the receiver's perspective.
1184 message_header_builder.add_monotonic_sent_time(
1185 sender->monotonic_sent_time().time_since_epoch().count());
1186 message_header_builder.add_realtime_sent_time(
1187 sender->realtime_sent_time().time_since_epoch().count());
1188 message_header_builder.add_queue_index(sender->sent_queue_index());
1189
1190 message_header_builder.add_monotonic_remote_time(
Austin Schuh287d43d2020-12-04 20:19:33 -08001191 timestamped_message.monotonic_remote_time.time_since_epoch().count());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001192 message_header_builder.add_realtime_remote_time(
Austin Schuh287d43d2020-12-04 20:19:33 -08001193 timestamped_message.realtime_remote_time.time_since_epoch().count());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001194
1195 message_header_builder.add_remote_queue_index(remote_queue_index);
Austin Schuh315b96b2020-12-11 21:21:12 -08001196 message_header_builder.add_boot_uuid(boot_uuid_offset);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001197
Austin Schuh969cd602021-01-03 00:09:45 -08001198 fbb.Finish(message_header_builder.Finish());
1199
1200 remote_timestamp_senders_[timestamped_message.channel_index]->Send(
1201 FlatbufferDetachedBuffer<RemoteMessage>(fbb.Release()),
1202 timestamped_message.monotonic_timestamp_time);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001203 }
1204
1205 return true;
1206}
1207
Austin Schuh969cd602021-01-03 00:09:45 -08001208LogReader::RemoteMessageSender::RemoteMessageSender(
1209 aos::Sender<message_bridge::RemoteMessage> sender, EventLoop *event_loop)
1210 : event_loop_(event_loop),
1211 sender_(std::move(sender)),
1212 timer_(event_loop->AddTimer([this]() { SendTimestamp(); })) {}
1213
1214void LogReader::RemoteMessageSender::ScheduleTimestamp() {
1215 if (remote_timestamps_.empty()) {
1216 CHECK_NOTNULL(timer_);
1217 timer_->Disable();
1218 scheduled_time_ = monotonic_clock::min_time;
1219 return;
1220 }
1221
1222 if (scheduled_time_ != remote_timestamps_.front().monotonic_timestamp_time) {
1223 CHECK_NOTNULL(timer_);
Austin Schuh816e5d62021-01-05 23:42:20 -08001224 timer_->Setup(remote_timestamps_.front().monotonic_timestamp_time);
Austin Schuh969cd602021-01-03 00:09:45 -08001225 scheduled_time_ = remote_timestamps_.front().monotonic_timestamp_time;
Austin Schuh3d94be02021-02-12 23:15:20 -08001226 CHECK_GE(scheduled_time_, event_loop_->monotonic_now())
1227 << event_loop_->node()->name()->string_view();
Austin Schuh969cd602021-01-03 00:09:45 -08001228 }
1229}
1230
1231void LogReader::RemoteMessageSender::Send(
1232 FlatbufferDetachedBuffer<RemoteMessage> remote_message,
1233 monotonic_clock::time_point monotonic_timestamp_time) {
1234 // There are 2 cases. Either we have a monotonic_timestamp_time and need to
1235 // resend the timestamp at the correct time, or we don't and can send it
1236 // immediately.
1237 if (monotonic_timestamp_time == monotonic_clock::min_time) {
1238 CHECK(remote_timestamps_.empty())
1239 << ": Unsupported mix of timestamps and no timestamps.";
1240 sender_.Send(std::move(remote_message));
1241 } else {
Austin Schuhb22ae422021-01-31 17:57:06 -08001242 remote_timestamps_.emplace(
1243 std::upper_bound(
1244 remote_timestamps_.begin(), remote_timestamps_.end(),
1245 monotonic_timestamp_time,
1246 [](const aos::monotonic_clock::time_point monotonic_timestamp_time,
1247 const Timestamp &timestamp) {
1248 return monotonic_timestamp_time <
1249 timestamp.monotonic_timestamp_time;
1250 }),
1251 std::move(remote_message), monotonic_timestamp_time);
Austin Schuh969cd602021-01-03 00:09:45 -08001252 ScheduleTimestamp();
1253 }
1254}
1255
1256void LogReader::RemoteMessageSender::SendTimestamp() {
Austin Schuh3d94be02021-02-12 23:15:20 -08001257 CHECK_EQ(event_loop_->context().monotonic_event_time, scheduled_time_)
1258 << event_loop_->node()->name()->string_view();
Austin Schuh969cd602021-01-03 00:09:45 -08001259 CHECK(!remote_timestamps_.empty());
1260
1261 // Send out all timestamps at the currently scheduled time.
1262 while (remote_timestamps_.front().monotonic_timestamp_time ==
1263 scheduled_time_) {
1264 sender_.Send(std::move(remote_timestamps_.front().remote_message));
1265 remote_timestamps_.pop_front();
1266 if (remote_timestamps_.empty()) {
1267 break;
1268 }
1269 }
1270 scheduled_time_ = monotonic_clock::min_time;
1271
1272 ScheduleTimestamp();
1273}
1274
1275LogReader::RemoteMessageSender *LogReader::State::RemoteTimestampSender(
Austin Schuh61e973f2021-02-21 21:43:56 -08001276 const Channel *channel, const Connection *connection) {
1277 message_bridge::ChannelTimestampFinder finder(event_loop_);
1278 // Look at any pre-created channel/connection pairs.
1279 {
1280 auto it =
1281 channel_timestamp_loggers_.find(std::make_pair(channel, connection));
1282 if (it != channel_timestamp_loggers_.end()) {
1283 return it->second.get();
1284 }
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001285 }
1286
Austin Schuh61e973f2021-02-21 21:43:56 -08001287 // That failed, so resolve the RemoteMessage channel timestamps will be logged
1288 // to.
1289 const Channel *timestamp_channel = finder.ForChannel(channel, connection);
1290
1291 {
1292 // See if that has been created before. If so, cache it in
1293 // channel_timestamp_loggers_ and return.
1294 auto it = timestamp_loggers_.find(timestamp_channel);
1295 if (it != timestamp_loggers_.end()) {
1296 CHECK(channel_timestamp_loggers_
1297 .try_emplace(std::make_pair(channel, connection), it->second)
1298 .second);
1299 return it->second.get();
1300 }
1301 }
1302
1303 // Otherwise, make a sender, save it, and cache it.
1304 auto result = channel_timestamp_loggers_.try_emplace(
1305 std::make_pair(channel, connection),
1306 std::make_shared<RemoteMessageSender>(
1307 event_loop()->MakeSender<RemoteMessage>(
1308 timestamp_channel->name()->string_view()),
1309 event_loop()));
1310
1311 CHECK(timestamp_loggers_.try_emplace(timestamp_channel, result.first->second)
1312 .second);
1313 return result.first->second.get();
Austin Schuh858c9f32020-08-31 16:56:12 -07001314}
1315
Austin Schuhdda74ec2021-01-03 19:30:37 -08001316TimestampedMessage LogReader::State::PopOldest() {
Austin Schuhe639ea12021-01-25 13:00:22 -08001317 CHECK(timestamp_mapper_ != nullptr);
1318 TimestampedMessage *result_ptr = timestamp_mapper_->Front();
1319 CHECK(result_ptr != nullptr);
Austin Schuh858c9f32020-08-31 16:56:12 -07001320
Austin Schuhe639ea12021-01-25 13:00:22 -08001321 TimestampedMessage result = std::move(*result_ptr);
1322
Austin Schuh2f8fd752020-09-01 22:38:28 -07001323 VLOG(2) << MaybeNodeName(event_loop_->node()) << "PopOldest Popping "
Austin Schuhe639ea12021-01-25 13:00:22 -08001324 << result.monotonic_event_time;
1325 timestamp_mapper_->PopFront();
Austin Schuh858c9f32020-08-31 16:56:12 -07001326 SeedSortedMessages();
1327
Austin Schuhe639ea12021-01-25 13:00:22 -08001328 if (result.monotonic_remote_time != monotonic_clock::min_time) {
1329 message_bridge::NoncausalOffsetEstimator *filter =
1330 filters_[result.channel_index];
1331 CHECK(filter != nullptr);
1332
1333 // TODO(austin): We probably want to push this down into the timestamp
1334 // mapper directly.
Austin Schuh3d94be02021-02-12 23:15:20 -08001335 filter->Pop(event_loop_->node(), event_loop_->monotonic_now());
Austin Schuh2f8fd752020-09-01 22:38:28 -07001336 }
Austin Schuh5ee56872021-01-30 16:53:34 -08001337 VLOG(1) << "Popped " << result
1338 << configuration::CleanedChannelToString(
1339 event_loop_->configuration()->channels()->Get(
1340 factory_channel_index_[result.channel_index]));
Austin Schuhe639ea12021-01-25 13:00:22 -08001341 return result;
Austin Schuh858c9f32020-08-31 16:56:12 -07001342}
1343
1344monotonic_clock::time_point LogReader::State::OldestMessageTime() const {
Austin Schuhe639ea12021-01-25 13:00:22 -08001345 if (timestamp_mapper_ == nullptr) {
Austin Schuh287d43d2020-12-04 20:19:33 -08001346 return monotonic_clock::max_time;
1347 }
Austin Schuhe639ea12021-01-25 13:00:22 -08001348 TimestampedMessage *result_ptr = timestamp_mapper_->Front();
1349 if (result_ptr == nullptr) {
1350 return monotonic_clock::max_time;
1351 }
1352 VLOG(2) << MaybeNodeName(event_loop_->node()) << "oldest message at "
1353 << result_ptr->monotonic_event_time;
1354 return result_ptr->monotonic_event_time;
Austin Schuh858c9f32020-08-31 16:56:12 -07001355}
1356
1357void LogReader::State::SeedSortedMessages() {
Austin Schuh287d43d2020-12-04 20:19:33 -08001358 if (!timestamp_mapper_) return;
Austin Schuh858c9f32020-08-31 16:56:12 -07001359
Austin Schuhe639ea12021-01-25 13:00:22 -08001360 timestamp_mapper_->QueueFor(chrono::duration_cast<chrono::seconds>(
1361 chrono::duration<double>(FLAGS_time_estimation_buffer_seconds)));
Austin Schuh858c9f32020-08-31 16:56:12 -07001362}
1363
1364void LogReader::State::Deregister() {
1365 for (size_t i = 0; i < channels_.size(); ++i) {
1366 channels_[i].reset();
1367 }
Austin Schuh61e973f2021-02-21 21:43:56 -08001368 channel_timestamp_loggers_.clear();
1369 timestamp_loggers_.clear();
Austin Schuh858c9f32020-08-31 16:56:12 -07001370 event_loop_unique_ptr_.reset();
1371 event_loop_ = nullptr;
1372 timer_handler_ = nullptr;
1373 node_event_loop_factory_ = nullptr;
1374}
1375
Austin Schuhe309d2a2019-11-29 13:25:21 -08001376} // namespace logger
1377} // namespace aos