blob: 098b78ce837bc5b38afafd2e6558608e65211316 [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 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 Schuh61e973f2021-02-21 21:43:56 -080021#include "aos/network/timestamp_channel.h"
Austin Schuhe309d2a2019-11-29 13:25:21 -080022#include "aos/time/time.h"
Brian Silvermanae7c0332020-09-30 16:58:23 -070023#include "aos/util/file.h"
Austin Schuh4385b142021-03-14 21:31:13 -070024#include "aos/uuid.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)
Austin Schuh8902fa52021-03-14 22:39:24 -0700318 << ": Found parts from different boots for node "
319 << node->name()->string_view() << " "
Austin Schuh315b96b2020-12-11 21:21:12 -0800320 << LogFileVectorToString(log_files_);
321 }
James Kuszmaul4f106fb2021-01-05 20:53:02 -0800322 if (!filtered_parts[0].source_boot_uuid.empty()) {
323 event_loop_factory_->GetNodeEventLoopFactory(node)->set_boot_uuid(
324 filtered_parts[0].source_boot_uuid);
325 }
Austin Schuh315b96b2020-12-11 21:21:12 -0800326 }
327
Austin Schuh287d43d2020-12-04 20:19:33 -0800328 states_[node_index] = std::make_unique<State>(
329 filtered_parts.size() == 0u
330 ? nullptr
331 : std::make_unique<TimestampMapper>(std::move(filtered_parts)));
Austin Schuh8bd96322020-02-13 21:18:22 -0800332 State *state = states_[node_index].get();
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700333 state->set_event_loop(state->SetNodeEventLoopFactory(
Austin Schuh858c9f32020-08-31 16:56:12 -0700334 event_loop_factory_->GetNodeEventLoopFactory(node)));
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700335
336 state->SetChannelCount(logged_configuration()->channels()->size());
Austin Schuhe639ea12021-01-25 13:00:22 -0800337 timestamp_mappers.emplace_back(state->timestamp_mapper());
Austin Schuhcde938c2020-02-02 17:30:07 -0800338 }
Austin Schuhe639ea12021-01-25 13:00:22 -0800339 filters_->SetTimestampMappers(std::move(timestamp_mappers));
340
341 // Note: this needs to be set before any times are pulled, or we won't observe
342 // the timestamps.
Austin Schuh87dd3832021-01-01 23:07:31 -0800343 event_loop_factory_->SetTimeConverter(filters_.get());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700344
Austin Schuh287d43d2020-12-04 20:19:33 -0800345 for (const Node *node : configuration::GetNodes(configuration())) {
346 const size_t node_index =
347 configuration::GetNodeIndex(configuration(), node);
348 State *state = states_[node_index].get();
349 for (const Node *other_node : configuration::GetNodes(configuration())) {
350 const size_t other_node_index =
351 configuration::GetNodeIndex(configuration(), other_node);
352 State *other_state = states_[other_node_index].get();
353 if (other_state != state) {
354 state->AddPeer(other_state);
355 }
356 }
357 }
358
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700359 // Register after making all the State objects so we can build references
360 // between them.
361 for (const Node *node : configuration::GetNodes(configuration())) {
362 const size_t node_index =
363 configuration::GetNodeIndex(configuration(), node);
364 State *state = states_[node_index].get();
365
366 Register(state->event_loop());
367 }
368
James Kuszmaul46d82582020-05-09 19:50:09 -0700369 if (live_nodes_ == 0) {
370 LOG(FATAL)
371 << "Don't have logs from any of the nodes in the replay config--are "
372 "you sure that the replay config matches the original config?";
373 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800374
Austin Schuh87dd3832021-01-01 23:07:31 -0800375 filters_->CheckGraph();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800376
Austin Schuh858c9f32020-08-31 16:56:12 -0700377 for (std::unique_ptr<State> &state : states_) {
378 state->SeedSortedMessages();
379 }
380
Austin Schuh2f8fd752020-09-01 22:38:28 -0700381 // We want to start the log file at the last start time of the log files
382 // from all the nodes. Compute how long each node's simulation needs to run
383 // to move time to this point.
Austin Schuh8bd96322020-02-13 21:18:22 -0800384 distributed_clock::time_point start_time = distributed_clock::min_time;
Austin Schuhcde938c2020-02-02 17:30:07 -0800385
Austin Schuh2f8fd752020-09-01 22:38:28 -0700386 // TODO(austin): We want an "OnStart" callback for each node rather than
387 // running until the last node.
388
Austin Schuh8bd96322020-02-13 21:18:22 -0800389 for (std::unique_ptr<State> &state : states_) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700390 VLOG(1) << "Start time is " << state->monotonic_start_time() << " for node "
391 << MaybeNodeName(state->event_loop()->node()) << "now "
392 << state->monotonic_now();
Austin Schuh287d43d2020-12-04 20:19:33 -0800393 if (state->monotonic_start_time() == monotonic_clock::min_time) {
394 continue;
395 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700396 // And start computing the start time on the distributed clock now that
397 // that works.
Austin Schuh858c9f32020-08-31 16:56:12 -0700398 start_time = std::max(
399 start_time, state->ToDistributedClock(state->monotonic_start_time()));
Austin Schuhcde938c2020-02-02 17:30:07 -0800400 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700401
Austin Schuh87dd3832021-01-01 23:07:31 -0800402 // TODO(austin): If a node doesn't have a start time, we might not queue
403 // enough. If this happens, we'll explode with a frozen error eventually.
404
Austin Schuh2f8fd752020-09-01 22:38:28 -0700405 CHECK_GE(start_time, distributed_clock::epoch())
406 << ": Hmm, we have a node starting before the start of time. Offset "
407 "everything.";
Austin Schuhcde938c2020-02-02 17:30:07 -0800408
Austin Schuh6f3babe2020-01-26 20:34:50 -0800409 // Forwarding is tracked per channel. If it is enabled, we want to turn it
410 // off. Otherwise messages replayed will get forwarded across to the other
Austin Schuh2f8fd752020-09-01 22:38:28 -0700411 // nodes, and also replayed on the other nodes. This may not satisfy all
412 // our users, but it'll start the discussion.
Austin Schuh6f3babe2020-01-26 20:34:50 -0800413 if (configuration::MultiNode(event_loop_factory_->configuration())) {
414 for (size_t i = 0; i < logged_configuration()->channels()->size(); ++i) {
415 const Channel *channel = logged_configuration()->channels()->Get(i);
416 const Node *node = configuration::GetNode(
417 configuration(), channel->source_node()->string_view());
418
Austin Schuh8bd96322020-02-13 21:18:22 -0800419 State *state =
420 states_[configuration::GetNodeIndex(configuration(), node)].get();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800421
422 const Channel *remapped_channel =
Austin Schuh858c9f32020-08-31 16:56:12 -0700423 RemapChannel(state->event_loop(), channel);
Austin Schuh6f3babe2020-01-26 20:34:50 -0800424
425 event_loop_factory_->DisableForwarding(remapped_channel);
426 }
Austin Schuh4c3b9702020-08-30 11:34:55 -0700427
428 // If we are replaying a log, we don't want a bunch of redundant messages
429 // from both the real message bridge and simulated message bridge.
430 event_loop_factory_->DisableStatistics();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800431 }
432
Austin Schuhcde938c2020-02-02 17:30:07 -0800433 // While we are starting the system up, we might be relying on matching data
434 // to timestamps on log files where the timestamp log file starts before the
435 // data. In this case, it is reasonable to expect missing data.
Austin Schuhdda74ec2021-01-03 19:30:37 -0800436 {
437 const bool prior_ignore_missing_data = ignore_missing_data_;
438 ignore_missing_data_ = true;
439 VLOG(1) << "Running until " << start_time << " in Register";
440 event_loop_factory_->RunFor(start_time.time_since_epoch());
441 VLOG(1) << "At start time";
442 // Now that we are running for real, missing data means that the log file is
443 // corrupted or went wrong.
444 ignore_missing_data_ = prior_ignore_missing_data;
445 }
Austin Schuh92547522019-12-28 14:33:43 -0800446
Austin Schuh8bd96322020-02-13 21:18:22 -0800447 for (std::unique_ptr<State> &state : states_) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700448 // Make the RT clock be correct before handing it to the user.
449 if (state->realtime_start_time() != realtime_clock::min_time) {
450 state->SetRealtimeOffset(state->monotonic_start_time(),
451 state->realtime_start_time());
452 }
453 VLOG(1) << "Start time is " << state->monotonic_start_time() << " for node "
454 << MaybeNodeName(state->event_loop()->node()) << "now "
455 << state->monotonic_now();
456 }
457
458 if (FLAGS_timestamps_to_csv) {
Austin Schuh0ca1fd32020-12-18 22:53:05 -0800459 filters_->Start(event_loop_factory);
Austin Schuh8bd96322020-02-13 21:18:22 -0800460 }
461}
462
Austin Schuh2f8fd752020-09-01 22:38:28 -0700463message_bridge::NoncausalOffsetEstimator *LogReader::GetFilter(
Austin Schuh8bd96322020-02-13 21:18:22 -0800464 const Node *node_a, const Node *node_b) {
Austin Schuh0ca1fd32020-12-18 22:53:05 -0800465 if (filters_) {
466 return filters_->GetFilter(node_a, node_b);
Austin Schuh8bd96322020-02-13 21:18:22 -0800467 }
Austin Schuh0ca1fd32020-12-18 22:53:05 -0800468 return nullptr;
Austin Schuh8bd96322020-02-13 21:18:22 -0800469}
470
Austin Schuhe309d2a2019-11-29 13:25:21 -0800471void LogReader::Register(EventLoop *event_loop) {
Austin Schuh8bd96322020-02-13 21:18:22 -0800472 State *state =
473 states_[configuration::GetNodeIndex(configuration(), event_loop->node())]
474 .get();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800475
Austin Schuh858c9f32020-08-31 16:56:12 -0700476 state->set_event_loop(event_loop);
Austin Schuhe309d2a2019-11-29 13:25:21 -0800477
Tyler Chatow67ddb032020-01-12 14:30:04 -0800478 // We don't run timing reports when trying to print out logged data, because
479 // otherwise we would end up printing out the timing reports themselves...
480 // This is only really relevant when we are replaying into a simulation.
Austin Schuh6f3babe2020-01-26 20:34:50 -0800481 event_loop->SkipTimingReport();
482 event_loop->SkipAosLog();
Austin Schuh39788ff2019-12-01 18:22:57 -0800483
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700484 for (size_t logged_channel_index = 0;
485 logged_channel_index < logged_configuration()->channels()->size();
486 ++logged_channel_index) {
487 const Channel *channel = RemapChannel(
488 event_loop,
489 logged_configuration()->channels()->Get(logged_channel_index));
Austin Schuh8bd96322020-02-13 21:18:22 -0800490
Austin Schuh532656d2021-01-11 10:17:18 -0800491 if (channel->logger() == LoggerConfig::NOT_LOGGED) {
492 continue;
493 }
494
Austin Schuh2f8fd752020-09-01 22:38:28 -0700495 message_bridge::NoncausalOffsetEstimator *filter = nullptr;
Austin Schuh969cd602021-01-03 00:09:45 -0800496 RemoteMessageSender *remote_timestamp_sender = nullptr;
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700497
498 State *source_state = nullptr;
Austin Schuh8bd96322020-02-13 21:18:22 -0800499
500 if (!configuration::ChannelIsSendableOnNode(channel, event_loop->node()) &&
501 configuration::ChannelIsReadableOnNode(channel, event_loop->node())) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700502 // We've got a message which is being forwarded to this node.
503 const Node *source_node = configuration::GetNode(
Austin Schuh8bd96322020-02-13 21:18:22 -0800504 event_loop->configuration(), channel->source_node()->string_view());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700505 filter = GetFilter(event_loop->node(), source_node);
Austin Schuh8bd96322020-02-13 21:18:22 -0800506
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700507 // Delivery timestamps are supposed to be logged back on the source node.
508 // Configure remote timestamps to be sent.
Austin Schuh61e973f2021-02-21 21:43:56 -0800509 const Connection *connection =
510 configuration::ConnectionToNode(channel, event_loop->node());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700511 const bool delivery_time_is_logged =
Austin Schuh61e973f2021-02-21 21:43:56 -0800512 configuration::ConnectionDeliveryTimeIsLoggedOnNode(connection,
513 source_node);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700514
515 source_state =
516 states_[configuration::GetNodeIndex(configuration(), source_node)]
517 .get();
518
519 if (delivery_time_is_logged) {
520 remote_timestamp_sender =
Austin Schuh61e973f2021-02-21 21:43:56 -0800521 source_state->RemoteTimestampSender(channel, connection);
Austin Schuh8bd96322020-02-13 21:18:22 -0800522 }
523 }
Austin Schuh858c9f32020-08-31 16:56:12 -0700524
Austin Schuh8d7e0bb2020-10-02 17:57:00 -0700525 state->SetChannel(
526 logged_channel_index,
527 configuration::ChannelIndex(event_loop->configuration(), channel),
528 event_loop->MakeRawSender(channel), filter, remote_timestamp_sender,
529 source_state);
Austin Schuhe309d2a2019-11-29 13:25:21 -0800530 }
531
Austin Schuh6aa77be2020-02-22 21:06:40 -0800532 // If we didn't find any log files with data in them, we won't ever get a
533 // callback or be live. So skip the rest of the setup.
Austin Schuh287d43d2020-12-04 20:19:33 -0800534 if (state->OldestMessageTime() == monotonic_clock::max_time) {
Austin Schuh6aa77be2020-02-22 21:06:40 -0800535 return;
536 }
537
Austin Schuh858c9f32020-08-31 16:56:12 -0700538 state->set_timer_handler(event_loop->AddTimer([this, state]() {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700539 VLOG(1) << "Starting sending " << MaybeNodeName(state->event_loop()->node())
540 << "at " << state->event_loop()->context().monotonic_event_time
541 << " now " << state->monotonic_now();
Austin Schuh858c9f32020-08-31 16:56:12 -0700542 if (state->OldestMessageTime() == monotonic_clock::max_time) {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800543 --live_nodes_;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700544 VLOG(1) << MaybeNodeName(state->event_loop()->node()) << "Node down!";
James Kuszmaul71a81932020-12-15 21:08:01 -0800545 if (exit_on_finish_ && live_nodes_ == 0) {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800546 event_loop_factory_->Exit();
547 }
James Kuszmaul314f1672020-01-03 20:02:08 -0800548 return;
549 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700550
Austin Schuhdda74ec2021-01-03 19:30:37 -0800551 TimestampedMessage timestamped_message = state->PopOldest();
Austin Schuh05b70472020-01-01 17:11:17 -0800552
Austin Schuhe309d2a2019-11-29 13:25:21 -0800553 const monotonic_clock::time_point monotonic_now =
Austin Schuh858c9f32020-08-31 16:56:12 -0700554 state->event_loop()->context().monotonic_event_time;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700555 if (!FLAGS_skip_order_validation) {
Austin Schuh287d43d2020-12-04 20:19:33 -0800556 CHECK(monotonic_now == timestamped_message.monotonic_event_time)
Austin Schuh2f8fd752020-09-01 22:38:28 -0700557 << ": " << FlatbufferToJson(state->event_loop()->node()) << " Now "
558 << monotonic_now << " trying to send "
Austin Schuh287d43d2020-12-04 20:19:33 -0800559 << timestamped_message.monotonic_event_time << " failure "
Austin Schuh2f8fd752020-09-01 22:38:28 -0700560 << state->DebugString();
Austin Schuh287d43d2020-12-04 20:19:33 -0800561 } else if (monotonic_now != timestamped_message.monotonic_event_time) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700562 LOG(WARNING) << "Check failed: monotonic_now == "
Austin Schuh287d43d2020-12-04 20:19:33 -0800563 "timestamped_message.monotonic_event_time) ("
Austin Schuh2f8fd752020-09-01 22:38:28 -0700564 << monotonic_now << " vs. "
Austin Schuh287d43d2020-12-04 20:19:33 -0800565 << timestamped_message.monotonic_event_time
Austin Schuh2f8fd752020-09-01 22:38:28 -0700566 << "): " << FlatbufferToJson(state->event_loop()->node())
567 << " Now " << monotonic_now << " trying to send "
Austin Schuh287d43d2020-12-04 20:19:33 -0800568 << timestamped_message.monotonic_event_time << " failure "
Austin Schuh2f8fd752020-09-01 22:38:28 -0700569 << state->DebugString();
570 }
Austin Schuhe309d2a2019-11-29 13:25:21 -0800571
Austin Schuh287d43d2020-12-04 20:19:33 -0800572 if (timestamped_message.monotonic_event_time >
Austin Schuh858c9f32020-08-31 16:56:12 -0700573 state->monotonic_start_time() ||
Austin Schuh15649d62019-12-28 16:36:38 -0800574 event_loop_factory_ != nullptr) {
Austin Schuhdda74ec2021-01-03 19:30:37 -0800575 if (timestamped_message.data.span().size() != 0u) {
576 if (timestamped_message.monotonic_remote_time !=
577 monotonic_clock::min_time) {
Austin Schuh8bd96322020-02-13 21:18:22 -0800578 // Confirm that the message was sent on the sending node before the
579 // destination node (this node). As a proxy, do this by making sure
580 // that time on the source node is past when the message was sent.
Austin Schuh87dd3832021-01-01 23:07:31 -0800581 //
582 // TODO(austin): <= means that the cause message (which we know) could
583 // happen after the effect even though we know they are at the same
584 // time. I doubt anyone will notice for a bit, but we should really
585 // fix that.
Austin Schuh2f8fd752020-09-01 22:38:28 -0700586 if (!FLAGS_skip_order_validation) {
Austin Schuh87dd3832021-01-01 23:07:31 -0800587 CHECK_LE(
Austin Schuh287d43d2020-12-04 20:19:33 -0800588 timestamped_message.monotonic_remote_time,
589 state->monotonic_remote_now(timestamped_message.channel_index))
Austin Schuh2f8fd752020-09-01 22:38:28 -0700590 << state->event_loop()->node()->name()->string_view() << " to "
Austin Schuh287d43d2020-12-04 20:19:33 -0800591 << state->remote_node(timestamped_message.channel_index)
592 ->name()
593 ->string_view()
Austin Schuh315b96b2020-12-11 21:21:12 -0800594 << " while trying to send a message on "
595 << configuration::CleanedChannelToString(
596 logged_configuration()->channels()->Get(
597 timestamped_message.channel_index))
Austin Schuh2f8fd752020-09-01 22:38:28 -0700598 << " " << state->DebugString();
Austin Schuh87dd3832021-01-01 23:07:31 -0800599 } else if (timestamped_message.monotonic_remote_time >
Austin Schuh287d43d2020-12-04 20:19:33 -0800600 state->monotonic_remote_now(
601 timestamped_message.channel_index)) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700602 LOG(WARNING)
Austin Schuh287d43d2020-12-04 20:19:33 -0800603 << "Check failed: timestamped_message.monotonic_remote_time < "
604 "state->monotonic_remote_now(timestamped_message.channel_"
605 "index) ("
606 << timestamped_message.monotonic_remote_time << " vs. "
607 << state->monotonic_remote_now(
608 timestamped_message.channel_index)
609 << ") " << state->event_loop()->node()->name()->string_view()
610 << " to "
611 << state->remote_node(timestamped_message.channel_index)
612 ->name()
613 ->string_view()
614 << " currently " << timestamped_message.monotonic_event_time
Austin Schuh2f8fd752020-09-01 22:38:28 -0700615 << " ("
616 << state->ToDistributedClock(
Austin Schuh287d43d2020-12-04 20:19:33 -0800617 timestamped_message.monotonic_event_time)
Austin Schuh2f8fd752020-09-01 22:38:28 -0700618 << ") remote event time "
Austin Schuh287d43d2020-12-04 20:19:33 -0800619 << timestamped_message.monotonic_remote_time << " ("
Austin Schuh2f8fd752020-09-01 22:38:28 -0700620 << state->RemoteToDistributedClock(
Austin Schuh287d43d2020-12-04 20:19:33 -0800621 timestamped_message.channel_index,
622 timestamped_message.monotonic_remote_time)
Austin Schuh2f8fd752020-09-01 22:38:28 -0700623 << ") " << state->DebugString();
624 }
Austin Schuh8bd96322020-02-13 21:18:22 -0800625 }
626
Austin Schuh15649d62019-12-28 16:36:38 -0800627 // If we have access to the factory, use it to fix the realtime time.
Austin Schuh287d43d2020-12-04 20:19:33 -0800628 state->SetRealtimeOffset(timestamped_message.monotonic_event_time,
629 timestamped_message.realtime_event_time);
Austin Schuh15649d62019-12-28 16:36:38 -0800630
Austin Schuh2f8fd752020-09-01 22:38:28 -0700631 VLOG(1) << MaybeNodeName(state->event_loop()->node()) << "Sending "
Austin Schuh287d43d2020-12-04 20:19:33 -0800632 << timestamped_message.monotonic_event_time;
Austin Schuh2f8fd752020-09-01 22:38:28 -0700633 // TODO(austin): std::move channel_data in and make that efficient in
634 // simulation.
Austin Schuh287d43d2020-12-04 20:19:33 -0800635 state->Send(std::move(timestamped_message));
Austin Schuhdda74ec2021-01-03 19:30:37 -0800636 } else if (!ignore_missing_data_ &&
Austin Schuh5ee56872021-01-30 16:53:34 -0800637 // When starting up, we can have data which was sent before the
638 // log starts, but the timestamp was after the log starts. This
639 // is unreasonable to avoid, so ignore the missing data.
640 timestamped_message.monotonic_remote_time >=
641 state->monotonic_remote_start_time(
642 timestamped_message.channel_index) &&
Austin Schuhdda74ec2021-01-03 19:30:37 -0800643 !FLAGS_skip_missing_forwarding_entries) {
Austin Schuh5ee56872021-01-30 16:53:34 -0800644 // We've found a timestamp without data that we expect to have data for.
645 // This likely means that we are at the end of the log file. Record it
646 // and CHECK that in the rest of the log file, we don't find any more
647 // data on that channel. Not all channels will end at the same point in
648 // time since they can be in different files.
Austin Schuhdda74ec2021-01-03 19:30:37 -0800649 VLOG(1) << "Found the last message on channel "
650 << timestamped_message.channel_index;
651
652 // Vector storing if we've seen a nullptr message or not per channel.
653 std::vector<bool> last_message;
654 last_message.resize(logged_configuration()->channels()->size(), false);
655
656 last_message[timestamped_message.channel_index] = true;
657
658 // Now that we found the end of one channel, artificially stop the
659 // rest. It is confusing when part of your data gets replayed but not
Austin Schuh5ee56872021-01-30 16:53:34 -0800660 // all. Read the rest of the messages and drop them on the floor while
661 // doing some basic validation.
Austin Schuh858c9f32020-08-31 16:56:12 -0700662 while (state->OldestMessageTime() != monotonic_clock::max_time) {
Austin Schuhdda74ec2021-01-03 19:30:37 -0800663 TimestampedMessage next = state->PopOldest();
664 // Make sure that once we have seen the last message on a channel,
665 // data doesn't start back up again. If the user wants to play
666 // through events like this, they can set
667 // --skip_missing_forwarding_entries or ignore_missing_data_.
668 CHECK_LT(next.channel_index, last_message.size());
669 if (next.data.span().size() == 0u) {
670 last_message[next.channel_index] = true;
671 } else {
672 if (last_message[next.channel_index]) {
673 LOG(FATAL)
674 << "Found missing data in the middle of the log file on "
675 "channel "
676 << next.channel_index << " Last "
677 << last_message[next.channel_index] << state->DebugString();
678 }
679 }
Austin Schuh8bd96322020-02-13 21:18:22 -0800680 }
Austin Schuh92547522019-12-28 14:33:43 -0800681 }
Austin Schuhe309d2a2019-11-29 13:25:21 -0800682 } else {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800683 LOG(WARNING)
684 << "Not sending data from before the start of the log file. "
Austin Schuh287d43d2020-12-04 20:19:33 -0800685 << timestamped_message.monotonic_event_time.time_since_epoch().count()
Austin Schuh6f3babe2020-01-26 20:34:50 -0800686 << " start " << monotonic_start_time().time_since_epoch().count()
Austin Schuhd85baf82020-10-19 11:50:12 -0700687 << " "
Austin Schuh287d43d2020-12-04 20:19:33 -0800688 << FlatbufferToJson(timestamped_message.data,
Austin Schuhd85baf82020-10-19 11:50:12 -0700689 {.multi_line = false, .max_vector_size = 100});
Austin Schuhe309d2a2019-11-29 13:25:21 -0800690 }
691
Austin Schuh858c9f32020-08-31 16:56:12 -0700692 const monotonic_clock::time_point next_time = state->OldestMessageTime();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800693 if (next_time != monotonic_clock::max_time) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700694 VLOG(1) << "Scheduling " << MaybeNodeName(state->event_loop()->node())
695 << "wakeup for " << next_time << "("
696 << state->ToDistributedClock(next_time)
697 << " distributed), now is " << state->monotonic_now();
Austin Schuh858c9f32020-08-31 16:56:12 -0700698 state->Setup(next_time);
James Kuszmaul314f1672020-01-03 20:02:08 -0800699 } else {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700700 VLOG(1) << MaybeNodeName(state->event_loop()->node())
701 << "No next message, scheduling shutdown";
702 // Set a timer up immediately after now to die. If we don't do this,
703 // then the senders waiting on the message we just read will never get
704 // called.
Austin Schuheecb9282020-01-08 17:43:30 -0800705 if (event_loop_factory_ != nullptr) {
Austin Schuh858c9f32020-08-31 16:56:12 -0700706 state->Setup(monotonic_now + event_loop_factory_->send_delay() +
707 std::chrono::nanoseconds(1));
Austin Schuheecb9282020-01-08 17:43:30 -0800708 }
Austin Schuhe309d2a2019-11-29 13:25:21 -0800709 }
Austin Schuh8bd96322020-02-13 21:18:22 -0800710
Austin Schuh2f8fd752020-09-01 22:38:28 -0700711 VLOG(1) << MaybeNodeName(state->event_loop()->node()) << "Done sending at "
712 << state->event_loop()->context().monotonic_event_time << " now "
713 << state->monotonic_now();
Austin Schuh858c9f32020-08-31 16:56:12 -0700714 }));
Austin Schuhe309d2a2019-11-29 13:25:21 -0800715
Austin Schuh6f3babe2020-01-26 20:34:50 -0800716 ++live_nodes_;
717
Austin Schuh858c9f32020-08-31 16:56:12 -0700718 if (state->OldestMessageTime() != monotonic_clock::max_time) {
719 event_loop->OnRun([state]() { state->Setup(state->OldestMessageTime()); });
Austin Schuhe309d2a2019-11-29 13:25:21 -0800720 }
721}
722
723void LogReader::Deregister() {
James Kuszmaul84ff3e52020-01-03 19:48:53 -0800724 // Make sure that things get destroyed in the correct order, rather than
725 // relying on getting the order correct in the class definition.
Austin Schuh8bd96322020-02-13 21:18:22 -0800726 for (std::unique_ptr<State> &state : states_) {
Austin Schuh858c9f32020-08-31 16:56:12 -0700727 state->Deregister();
Austin Schuhe309d2a2019-11-29 13:25:21 -0800728 }
Austin Schuh92547522019-12-28 14:33:43 -0800729
James Kuszmaul84ff3e52020-01-03 19:48:53 -0800730 event_loop_factory_unique_ptr_.reset();
731 event_loop_factory_ = nullptr;
Austin Schuhe309d2a2019-11-29 13:25:21 -0800732}
733
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800734void LogReader::RemapLoggedChannel(std::string_view name, std::string_view type,
Austin Schuh0de30f32020-12-06 12:44:28 -0800735 std::string_view add_prefix,
736 std::string_view new_type) {
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800737 for (size_t ii = 0; ii < logged_configuration()->channels()->size(); ++ii) {
738 const Channel *const channel = logged_configuration()->channels()->Get(ii);
739 if (channel->name()->str() == name &&
740 channel->type()->string_view() == type) {
741 CHECK_EQ(0u, remapped_channels_.count(ii))
742 << "Already remapped channel "
743 << configuration::CleanedChannelToString(channel);
Austin Schuh0de30f32020-12-06 12:44:28 -0800744 RemappedChannel remapped_channel;
745 remapped_channel.remapped_name =
746 std::string(add_prefix) + std::string(name);
747 remapped_channel.new_type = new_type;
748 remapped_channels_[ii] = std::move(remapped_channel);
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800749 VLOG(1) << "Remapping channel "
750 << configuration::CleanedChannelToString(channel)
Austin Schuh0de30f32020-12-06 12:44:28 -0800751 << " to have name " << remapped_channels_[ii].remapped_name;
Austin Schuh6331ef92020-01-07 18:28:09 -0800752 MakeRemappedConfig();
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800753 return;
754 }
755 }
756 LOG(FATAL) << "Unabled to locate channel with name " << name << " and type "
757 << type;
758}
759
Austin Schuh01b4c352020-09-21 23:09:39 -0700760void LogReader::RemapLoggedChannel(std::string_view name, std::string_view type,
761 const Node *node,
Austin Schuh0de30f32020-12-06 12:44:28 -0800762 std::string_view add_prefix,
763 std::string_view new_type) {
Austin Schuh01b4c352020-09-21 23:09:39 -0700764 VLOG(1) << "Node is " << aos::FlatbufferToJson(node);
765 const Channel *remapped_channel =
766 configuration::GetChannel(logged_configuration(), name, type, "", node);
767 CHECK(remapped_channel != nullptr) << ": Failed to find {\"name\": \"" << name
768 << "\", \"type\": \"" << type << "\"}";
769 VLOG(1) << "Original {\"name\": \"" << name << "\", \"type\": \"" << type
770 << "\"}";
771 VLOG(1) << "Remapped "
772 << aos::configuration::StrippedChannelToString(remapped_channel);
773
774 // We want to make /spray on node 0 go to /0/spray by snooping the maps. And
775 // we want it to degrade if the heuristics fail to just work.
776 //
777 // The easiest way to do this is going to be incredibly specific and verbose.
778 // Look up /spray, to /0/spray. Then, prefix the result with /original to get
779 // /original/0/spray. Then, create a map from /original/spray to
780 // /original/0/spray for just the type we were asked for.
781 if (name != remapped_channel->name()->string_view()) {
782 MapT new_map;
783 new_map.match = std::make_unique<ChannelT>();
784 new_map.match->name = absl::StrCat(add_prefix, name);
785 new_map.match->type = type;
786 if (node != nullptr) {
787 new_map.match->source_node = node->name()->str();
788 }
789 new_map.rename = std::make_unique<ChannelT>();
790 new_map.rename->name =
791 absl::StrCat(add_prefix, remapped_channel->name()->string_view());
792 maps_.emplace_back(std::move(new_map));
793 }
794
795 const size_t channel_index =
796 configuration::ChannelIndex(logged_configuration(), remapped_channel);
797 CHECK_EQ(0u, remapped_channels_.count(channel_index))
798 << "Already remapped channel "
799 << configuration::CleanedChannelToString(remapped_channel);
Austin Schuh0de30f32020-12-06 12:44:28 -0800800
801 RemappedChannel remapped_channel_struct;
802 remapped_channel_struct.remapped_name =
803 std::string(add_prefix) +
804 std::string(remapped_channel->name()->string_view());
805 remapped_channel_struct.new_type = new_type;
806 remapped_channels_[channel_index] = std::move(remapped_channel_struct);
Austin Schuh01b4c352020-09-21 23:09:39 -0700807 MakeRemappedConfig();
808}
809
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800810void LogReader::MakeRemappedConfig() {
Austin Schuh8bd96322020-02-13 21:18:22 -0800811 for (std::unique_ptr<State> &state : states_) {
Austin Schuh6aa77be2020-02-22 21:06:40 -0800812 if (state) {
Austin Schuh858c9f32020-08-31 16:56:12 -0700813 CHECK(!state->event_loop())
Austin Schuh6aa77be2020-02-22 21:06:40 -0800814 << ": Can't change the mapping after the events are scheduled.";
815 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800816 }
Austin Schuhac0771c2020-01-07 18:36:30 -0800817
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800818 // If no remapping occurred and we are using the original config, then there
819 // is nothing interesting to do here.
820 if (remapped_channels_.empty() && replay_configuration_ == nullptr) {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800821 remapped_configuration_ = logged_configuration();
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800822 return;
823 }
824 // Config to copy Channel definitions from. Use the specified
825 // replay_configuration_ if it has been provided.
826 const Configuration *const base_config = replay_configuration_ == nullptr
827 ? logged_configuration()
828 : replay_configuration_;
Austin Schuh0de30f32020-12-06 12:44:28 -0800829
830 // Create a config with all the channels, but un-sorted/merged. Collect up
831 // the schemas while we do this. Call MergeConfiguration to sort everything,
832 // and then merge it all in together.
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800833
834 // This is the builder that we use for the config containing all the new
835 // channels.
Austin Schuh0de30f32020-12-06 12:44:28 -0800836 flatbuffers::FlatBufferBuilder fbb;
837 fbb.ForceDefaults(true);
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800838 std::vector<flatbuffers::Offset<Channel>> channel_offsets;
Austin Schuh0de30f32020-12-06 12:44:28 -0800839
840 CHECK_EQ(Channel::MiniReflectTypeTable()->num_elems, 13u)
841 << ": Merging logic needs to be updated when the number of channel "
842 "fields changes.";
843
844 // List of schemas.
845 std::map<std::string_view, FlatbufferVector<reflection::Schema>> schema_map;
846 // Make sure our new RemoteMessage schema is in there for old logs without it.
847 schema_map.insert(std::make_pair(
848 RemoteMessage::GetFullyQualifiedName(),
849 FlatbufferVector<reflection::Schema>(FlatbufferSpan<reflection::Schema>(
850 message_bridge::RemoteMessageSchema()))));
851
852 // Reconstruct the remapped channels.
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800853 for (auto &pair : remapped_channels_) {
Austin Schuh0de30f32020-12-06 12:44:28 -0800854 const Channel *const c = CHECK_NOTNULL(configuration::GetChannel(
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800855 base_config, logged_configuration()->channels()->Get(pair.first), "",
856 nullptr));
Austin Schuh0de30f32020-12-06 12:44:28 -0800857 channel_offsets.emplace_back(
858 CopyChannel(c, pair.second.remapped_name, "", &fbb));
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800859 }
Austin Schuh01b4c352020-09-21 23:09:39 -0700860
Austin Schuh0de30f32020-12-06 12:44:28 -0800861 // Now reconstruct the original channels, translating types as needed
862 for (const Channel *c : *base_config->channels()) {
863 // Search for a mapping channel.
864 std::string_view new_type = "";
865 for (auto &pair : remapped_channels_) {
866 const Channel *const remapped_channel =
867 logged_configuration()->channels()->Get(pair.first);
868 if (remapped_channel->name()->string_view() == c->name()->string_view() &&
869 remapped_channel->type()->string_view() == c->type()->string_view()) {
870 new_type = pair.second.new_type;
871 break;
872 }
873 }
874
875 // Copy everything over.
876 channel_offsets.emplace_back(CopyChannel(c, "", new_type, &fbb));
877
878 // Add the schema if it doesn't exist.
879 if (schema_map.find(c->type()->string_view()) == schema_map.end()) {
880 CHECK(c->has_schema());
881 schema_map.insert(std::make_pair(c->type()->string_view(),
882 RecursiveCopyFlatBuffer(c->schema())));
883 }
884 }
885
886 // The MergeConfiguration API takes a vector, not a map. Convert.
887 std::vector<FlatbufferVector<reflection::Schema>> schemas;
888 while (!schema_map.empty()) {
889 schemas.emplace_back(std::move(schema_map.begin()->second));
890 schema_map.erase(schema_map.begin());
891 }
892
893 // Create the Configuration containing the new channels that we want to add.
894 const flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Channel>>>
895 channels_offset =
896 channel_offsets.empty() ? 0 : fbb.CreateVector(channel_offsets);
897
898 // Copy over the old maps.
Austin Schuh01b4c352020-09-21 23:09:39 -0700899 std::vector<flatbuffers::Offset<Map>> map_offsets;
Austin Schuh0de30f32020-12-06 12:44:28 -0800900 if (base_config->maps()) {
901 for (const Map *map : *base_config->maps()) {
902 map_offsets.emplace_back(aos::RecursiveCopyFlatBuffer(map, &fbb));
903 }
904 }
905
906 // Now create the new maps. These are second so they take effect first.
Austin Schuh01b4c352020-09-21 23:09:39 -0700907 for (const MapT &map : maps_) {
908 const flatbuffers::Offset<flatbuffers::String> match_name_offset =
Austin Schuh0de30f32020-12-06 12:44:28 -0800909 fbb.CreateString(map.match->name);
Austin Schuh01b4c352020-09-21 23:09:39 -0700910 const flatbuffers::Offset<flatbuffers::String> match_type_offset =
Austin Schuh0de30f32020-12-06 12:44:28 -0800911 fbb.CreateString(map.match->type);
Austin Schuh01b4c352020-09-21 23:09:39 -0700912 const flatbuffers::Offset<flatbuffers::String> rename_name_offset =
Austin Schuh0de30f32020-12-06 12:44:28 -0800913 fbb.CreateString(map.rename->name);
Austin Schuh01b4c352020-09-21 23:09:39 -0700914 flatbuffers::Offset<flatbuffers::String> match_source_node_offset;
915 if (!map.match->source_node.empty()) {
Austin Schuh0de30f32020-12-06 12:44:28 -0800916 match_source_node_offset = fbb.CreateString(map.match->source_node);
Austin Schuh01b4c352020-09-21 23:09:39 -0700917 }
Austin Schuh0de30f32020-12-06 12:44:28 -0800918 Channel::Builder match_builder(fbb);
Austin Schuh01b4c352020-09-21 23:09:39 -0700919 match_builder.add_name(match_name_offset);
920 match_builder.add_type(match_type_offset);
921 if (!map.match->source_node.empty()) {
922 match_builder.add_source_node(match_source_node_offset);
923 }
924 const flatbuffers::Offset<Channel> match_offset = match_builder.Finish();
925
Austin Schuh0de30f32020-12-06 12:44:28 -0800926 Channel::Builder rename_builder(fbb);
Austin Schuh01b4c352020-09-21 23:09:39 -0700927 rename_builder.add_name(rename_name_offset);
928 const flatbuffers::Offset<Channel> rename_offset = rename_builder.Finish();
929
Austin Schuh0de30f32020-12-06 12:44:28 -0800930 Map::Builder map_builder(fbb);
Austin Schuh01b4c352020-09-21 23:09:39 -0700931 map_builder.add_match(match_offset);
932 map_builder.add_rename(rename_offset);
933 map_offsets.emplace_back(map_builder.Finish());
934 }
935
Austin Schuh0de30f32020-12-06 12:44:28 -0800936 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Map>>>
937 maps_offsets = map_offsets.empty() ? 0 : fbb.CreateVector(map_offsets);
Austin Schuh01b4c352020-09-21 23:09:39 -0700938
Austin Schuh0de30f32020-12-06 12:44:28 -0800939 // And copy everything else over.
940 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Node>>>
941 nodes_offset = aos::RecursiveCopyVectorTable(base_config->nodes(), &fbb);
942
943 flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<Application>>>
944 applications_offset =
945 aos::RecursiveCopyVectorTable(base_config->applications(), &fbb);
946
947 // Now insert everything else in unmodified.
948 ConfigurationBuilder configuration_builder(fbb);
949 if (!channels_offset.IsNull()) {
950 configuration_builder.add_channels(channels_offset);
951 }
952 if (!maps_offsets.IsNull()) {
953 configuration_builder.add_maps(maps_offsets);
954 }
955 if (!nodes_offset.IsNull()) {
956 configuration_builder.add_nodes(nodes_offset);
957 }
958 if (!applications_offset.IsNull()) {
959 configuration_builder.add_applications(applications_offset);
960 }
961
962 if (base_config->has_channel_storage_duration()) {
963 configuration_builder.add_channel_storage_duration(
964 base_config->channel_storage_duration());
965 }
966
967 CHECK_EQ(Configuration::MiniReflectTypeTable()->num_elems, 6u)
968 << ": Merging logic needs to be updated when the number of configuration "
969 "fields changes.";
970
971 fbb.Finish(configuration_builder.Finish());
972
973 // Clean it up and return it! By using MergeConfiguration here, we'll
974 // actually get a deduplicated config for free too.
975 FlatbufferDetachedBuffer<Configuration> new_merged_config =
976 configuration::MergeConfiguration(
977 FlatbufferDetachedBuffer<Configuration>(fbb.Release()));
978
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800979 remapped_configuration_buffer_ =
980 std::make_unique<FlatbufferDetachedBuffer<Configuration>>(
Austin Schuh0de30f32020-12-06 12:44:28 -0800981 configuration::MergeConfiguration(new_merged_config, schemas));
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800982
983 remapped_configuration_ = &remapped_configuration_buffer_->message();
Austin Schuh0de30f32020-12-06 12:44:28 -0800984
985 // TODO(austin): Lazily re-build to save CPU?
James Kuszmaulc7bbb3e2020-01-03 20:01:00 -0800986}
987
Austin Schuh6f3babe2020-01-26 20:34:50 -0800988const Channel *LogReader::RemapChannel(const EventLoop *event_loop,
989 const Channel *channel) {
990 std::string_view channel_name = channel->name()->string_view();
991 std::string_view channel_type = channel->type()->string_view();
992 const int channel_index =
993 configuration::ChannelIndex(logged_configuration(), channel);
994 // If the channel is remapped, find the correct channel name to use.
995 if (remapped_channels_.count(channel_index) > 0) {
Austin Schuhee711052020-08-24 16:06:09 -0700996 VLOG(3) << "Got remapped channel on "
Austin Schuh6f3babe2020-01-26 20:34:50 -0800997 << configuration::CleanedChannelToString(channel);
Austin Schuh0de30f32020-12-06 12:44:28 -0800998 channel_name = remapped_channels_[channel_index].remapped_name;
Austin Schuh6f3babe2020-01-26 20:34:50 -0800999 }
1000
Austin Schuhee711052020-08-24 16:06:09 -07001001 VLOG(2) << "Going to remap channel " << channel_name << " " << channel_type;
Austin Schuh6f3babe2020-01-26 20:34:50 -08001002 const Channel *remapped_channel = configuration::GetChannel(
1003 event_loop->configuration(), channel_name, channel_type,
1004 event_loop->name(), event_loop->node());
1005
1006 CHECK(remapped_channel != nullptr)
1007 << ": Unable to send {\"name\": \"" << channel_name << "\", \"type\": \""
1008 << channel_type << "\"} because it is not in the provided configuration.";
1009
1010 return remapped_channel;
1011}
1012
Austin Schuh287d43d2020-12-04 20:19:33 -08001013LogReader::State::State(std::unique_ptr<TimestampMapper> timestamp_mapper)
1014 : timestamp_mapper_(std::move(timestamp_mapper)) {}
1015
1016void LogReader::State::AddPeer(State *peer) {
1017 if (timestamp_mapper_ && peer->timestamp_mapper_) {
1018 timestamp_mapper_->AddPeer(peer->timestamp_mapper_.get());
1019 }
1020}
Austin Schuh858c9f32020-08-31 16:56:12 -07001021
1022EventLoop *LogReader::State::SetNodeEventLoopFactory(
1023 NodeEventLoopFactory *node_event_loop_factory) {
1024 node_event_loop_factory_ = node_event_loop_factory;
1025 event_loop_unique_ptr_ =
1026 node_event_loop_factory_->MakeEventLoop("log_reader");
1027 return event_loop_unique_ptr_.get();
1028}
1029
1030void LogReader::State::SetChannelCount(size_t count) {
1031 channels_.resize(count);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001032 remote_timestamp_senders_.resize(count);
Austin Schuh858c9f32020-08-31 16:56:12 -07001033 filters_.resize(count);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001034 channel_source_state_.resize(count);
1035 factory_channel_index_.resize(count);
1036 queue_index_map_.resize(count);
Austin Schuh858c9f32020-08-31 16:56:12 -07001037}
1038
1039void LogReader::State::SetChannel(
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001040 size_t logged_channel_index, size_t factory_channel_index,
1041 std::unique_ptr<RawSender> sender,
Austin Schuh2f8fd752020-09-01 22:38:28 -07001042 message_bridge::NoncausalOffsetEstimator *filter,
Austin Schuh969cd602021-01-03 00:09:45 -08001043 RemoteMessageSender *remote_timestamp_sender, State *source_state) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001044 channels_[logged_channel_index] = std::move(sender);
1045 filters_[logged_channel_index] = filter;
1046 remote_timestamp_senders_[logged_channel_index] = remote_timestamp_sender;
1047
1048 if (source_state) {
1049 channel_source_state_[logged_channel_index] = source_state;
1050
1051 if (remote_timestamp_sender != nullptr) {
1052 source_state->queue_index_map_[logged_channel_index] =
Austin Schuh9942bae2021-01-07 22:06:44 -08001053 std::make_unique<std::vector<State::ContiguousSentTimestamp>>();
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001054 }
1055 }
1056
1057 factory_channel_index_[logged_channel_index] = factory_channel_index;
1058}
1059
Austin Schuh287d43d2020-12-04 20:19:33 -08001060bool LogReader::State::Send(const TimestampedMessage &timestamped_message) {
1061 aos::RawSender *sender = channels_[timestamped_message.channel_index].get();
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001062 uint32_t remote_queue_index = 0xffffffff;
1063
Austin Schuh287d43d2020-12-04 20:19:33 -08001064 if (remote_timestamp_senders_[timestamped_message.channel_index] != nullptr) {
Austin Schuh9942bae2021-01-07 22:06:44 -08001065 std::vector<ContiguousSentTimestamp> *queue_index_map = CHECK_NOTNULL(
Austin Schuh287d43d2020-12-04 20:19:33 -08001066 CHECK_NOTNULL(channel_source_state_[timestamped_message.channel_index])
1067 ->queue_index_map_[timestamped_message.channel_index]
1068 .get());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001069
Austin Schuh9942bae2021-01-07 22:06:44 -08001070 struct SentTimestamp {
1071 monotonic_clock::time_point monotonic_event_time;
1072 uint32_t queue_index;
1073 } search;
1074
Austin Schuh287d43d2020-12-04 20:19:33 -08001075 search.monotonic_event_time = timestamped_message.monotonic_remote_time;
Austin Schuh287d43d2020-12-04 20:19:33 -08001076 search.queue_index = timestamped_message.remote_queue_index;
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001077
1078 // Find the sent time if available.
1079 auto element = std::lower_bound(
1080 queue_index_map->begin(), queue_index_map->end(), search,
Austin Schuh9942bae2021-01-07 22:06:44 -08001081 [](ContiguousSentTimestamp a, SentTimestamp b) {
1082 if (a.ending_monotonic_event_time < b.monotonic_event_time) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001083 return true;
1084 }
Austin Schuh9942bae2021-01-07 22:06:44 -08001085 if (a.starting_monotonic_event_time > b.monotonic_event_time) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001086 return false;
1087 }
Austin Schuh9942bae2021-01-07 22:06:44 -08001088
1089 if (a.ending_queue_index < b.queue_index) {
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001090 return true;
1091 }
Austin Schuh9942bae2021-01-07 22:06:44 -08001092 if (a.starting_queue_index >= b.queue_index) {
1093 return false;
1094 }
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001095
Austin Schuh9942bae2021-01-07 22:06:44 -08001096 // If it isn't clearly below or above, it is below. Since we return
1097 // the last element <, this will return a match.
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001098 return false;
1099 });
1100
1101 // TODO(austin): Be a bit more principled here, but we will want to do that
1102 // after the logger rewrite. We hit this when one node finishes, but the
1103 // other node isn't done yet. So there is no send time, but there is a
1104 // receive time.
1105 if (element != queue_index_map->end()) {
Austin Schuh9942bae2021-01-07 22:06:44 -08001106 CHECK_GE(timestamped_message.monotonic_remote_time,
1107 element->starting_monotonic_event_time);
1108 CHECK_LE(timestamped_message.monotonic_remote_time,
1109 element->ending_monotonic_event_time);
1110 CHECK_GE(timestamped_message.remote_queue_index,
1111 element->starting_queue_index);
1112 CHECK_LE(timestamped_message.remote_queue_index,
1113 element->ending_queue_index);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001114
Austin Schuh9942bae2021-01-07 22:06:44 -08001115 remote_queue_index = timestamped_message.remote_queue_index +
1116 element->actual_queue_index -
1117 element->starting_queue_index;
1118 } else {
1119 VLOG(1) << "No timestamp match in the map.";
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001120 }
1121 }
1122
1123 // Send! Use the replayed queue index here instead of the logged queue index
1124 // for the remote queue index. This makes re-logging work.
Austin Schuh287d43d2020-12-04 20:19:33 -08001125 const bool sent = sender->Send(
1126 timestamped_message.data.message().data()->Data(),
1127 timestamped_message.data.message().data()->size(),
1128 timestamped_message.monotonic_remote_time,
Austin Schuh8902fa52021-03-14 22:39:24 -07001129 timestamped_message.realtime_remote_time, remote_queue_index,
1130 (channel_source_state_[timestamped_message.channel_index] != nullptr
1131 ? CHECK_NOTNULL(
1132 channel_source_state_[timestamped_message.channel_index])
1133 ->event_loop_->boot_uuid()
1134 : event_loop_->boot_uuid()));
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001135 if (!sent) return false;
1136
Austin Schuh287d43d2020-12-04 20:19:33 -08001137 if (queue_index_map_[timestamped_message.channel_index]) {
Austin Schuh9942bae2021-01-07 22:06:44 -08001138 if (queue_index_map_[timestamped_message.channel_index]->empty()) {
1139 // Nothing here, start a range with 0 length.
1140 ContiguousSentTimestamp timestamp;
1141 timestamp.starting_monotonic_event_time =
1142 timestamp.ending_monotonic_event_time =
1143 timestamped_message.monotonic_event_time;
1144 timestamp.starting_queue_index = timestamp.ending_queue_index =
1145 timestamped_message.queue_index;
1146 timestamp.actual_queue_index = sender->sent_queue_index();
1147 queue_index_map_[timestamped_message.channel_index]->emplace_back(
1148 timestamp);
1149 } else {
1150 // We've got something. See if the next timestamp is still contiguous. If
1151 // so, grow it.
1152 ContiguousSentTimestamp *back =
1153 &queue_index_map_[timestamped_message.channel_index]->back();
1154 if ((back->starting_queue_index - back->actual_queue_index) ==
1155 (timestamped_message.queue_index - sender->sent_queue_index())) {
1156 back->ending_queue_index = timestamped_message.queue_index;
1157 back->ending_monotonic_event_time =
1158 timestamped_message.monotonic_event_time;
1159 } else {
1160 // Otherwise, make a new one.
1161 ContiguousSentTimestamp timestamp;
1162 timestamp.starting_monotonic_event_time =
1163 timestamp.ending_monotonic_event_time =
1164 timestamped_message.monotonic_event_time;
1165 timestamp.starting_queue_index = timestamp.ending_queue_index =
1166 timestamped_message.queue_index;
1167 timestamp.actual_queue_index = sender->sent_queue_index();
1168 queue_index_map_[timestamped_message.channel_index]->emplace_back(
1169 timestamp);
1170 }
1171 }
1172
1173 // TODO(austin): Should we prune the map? On a many day log, I only saw the
1174 // queue index diverge a couple of elements, which would be a very small
1175 // map.
Austin Schuh287d43d2020-12-04 20:19:33 -08001176 } else if (remote_timestamp_senders_[timestamped_message.channel_index] !=
1177 nullptr) {
Austin Schuh969cd602021-01-03 00:09:45 -08001178 flatbuffers::FlatBufferBuilder fbb;
1179 fbb.ForceDefaults(true);
Austin Schuh315b96b2020-12-11 21:21:12 -08001180 flatbuffers::Offset<flatbuffers::String> boot_uuid_offset =
Austin Schuh5e2bfb82021-03-13 22:46:55 -08001181 event_loop_->boot_uuid().PackString(&fbb);
Austin Schuh315b96b2020-12-11 21:21:12 -08001182
Austin Schuh969cd602021-01-03 00:09:45 -08001183 RemoteMessage::Builder message_header_builder(fbb);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001184
1185 message_header_builder.add_channel_index(
Austin Schuh287d43d2020-12-04 20:19:33 -08001186 factory_channel_index_[timestamped_message.channel_index]);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001187
1188 // Swap the remote and sent metrics. They are from the sender's
1189 // perspective, not the receiver's perspective.
1190 message_header_builder.add_monotonic_sent_time(
1191 sender->monotonic_sent_time().time_since_epoch().count());
1192 message_header_builder.add_realtime_sent_time(
1193 sender->realtime_sent_time().time_since_epoch().count());
1194 message_header_builder.add_queue_index(sender->sent_queue_index());
1195
1196 message_header_builder.add_monotonic_remote_time(
Austin Schuh287d43d2020-12-04 20:19:33 -08001197 timestamped_message.monotonic_remote_time.time_since_epoch().count());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001198 message_header_builder.add_realtime_remote_time(
Austin Schuh287d43d2020-12-04 20:19:33 -08001199 timestamped_message.realtime_remote_time.time_since_epoch().count());
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001200
1201 message_header_builder.add_remote_queue_index(remote_queue_index);
Austin Schuh315b96b2020-12-11 21:21:12 -08001202 message_header_builder.add_boot_uuid(boot_uuid_offset);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001203
Austin Schuh969cd602021-01-03 00:09:45 -08001204 fbb.Finish(message_header_builder.Finish());
1205
1206 remote_timestamp_senders_[timestamped_message.channel_index]->Send(
1207 FlatbufferDetachedBuffer<RemoteMessage>(fbb.Release()),
1208 timestamped_message.monotonic_timestamp_time);
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001209 }
1210
1211 return true;
1212}
1213
Austin Schuh969cd602021-01-03 00:09:45 -08001214LogReader::RemoteMessageSender::RemoteMessageSender(
1215 aos::Sender<message_bridge::RemoteMessage> sender, EventLoop *event_loop)
1216 : event_loop_(event_loop),
1217 sender_(std::move(sender)),
1218 timer_(event_loop->AddTimer([this]() { SendTimestamp(); })) {}
1219
1220void LogReader::RemoteMessageSender::ScheduleTimestamp() {
1221 if (remote_timestamps_.empty()) {
1222 CHECK_NOTNULL(timer_);
1223 timer_->Disable();
1224 scheduled_time_ = monotonic_clock::min_time;
1225 return;
1226 }
1227
1228 if (scheduled_time_ != remote_timestamps_.front().monotonic_timestamp_time) {
1229 CHECK_NOTNULL(timer_);
Austin Schuh816e5d62021-01-05 23:42:20 -08001230 timer_->Setup(remote_timestamps_.front().monotonic_timestamp_time);
Austin Schuh969cd602021-01-03 00:09:45 -08001231 scheduled_time_ = remote_timestamps_.front().monotonic_timestamp_time;
Austin Schuh3d94be02021-02-12 23:15:20 -08001232 CHECK_GE(scheduled_time_, event_loop_->monotonic_now())
1233 << event_loop_->node()->name()->string_view();
Austin Schuh969cd602021-01-03 00:09:45 -08001234 }
1235}
1236
1237void LogReader::RemoteMessageSender::Send(
1238 FlatbufferDetachedBuffer<RemoteMessage> remote_message,
1239 monotonic_clock::time_point monotonic_timestamp_time) {
1240 // There are 2 cases. Either we have a monotonic_timestamp_time and need to
1241 // resend the timestamp at the correct time, or we don't and can send it
1242 // immediately.
1243 if (monotonic_timestamp_time == monotonic_clock::min_time) {
1244 CHECK(remote_timestamps_.empty())
1245 << ": Unsupported mix of timestamps and no timestamps.";
1246 sender_.Send(std::move(remote_message));
1247 } else {
Austin Schuhb22ae422021-01-31 17:57:06 -08001248 remote_timestamps_.emplace(
1249 std::upper_bound(
1250 remote_timestamps_.begin(), remote_timestamps_.end(),
1251 monotonic_timestamp_time,
1252 [](const aos::monotonic_clock::time_point monotonic_timestamp_time,
1253 const Timestamp &timestamp) {
1254 return monotonic_timestamp_time <
1255 timestamp.monotonic_timestamp_time;
1256 }),
1257 std::move(remote_message), monotonic_timestamp_time);
Austin Schuh969cd602021-01-03 00:09:45 -08001258 ScheduleTimestamp();
1259 }
1260}
1261
1262void LogReader::RemoteMessageSender::SendTimestamp() {
Austin Schuh3d94be02021-02-12 23:15:20 -08001263 CHECK_EQ(event_loop_->context().monotonic_event_time, scheduled_time_)
1264 << event_loop_->node()->name()->string_view();
Austin Schuh969cd602021-01-03 00:09:45 -08001265 CHECK(!remote_timestamps_.empty());
1266
1267 // Send out all timestamps at the currently scheduled time.
1268 while (remote_timestamps_.front().monotonic_timestamp_time ==
1269 scheduled_time_) {
1270 sender_.Send(std::move(remote_timestamps_.front().remote_message));
1271 remote_timestamps_.pop_front();
1272 if (remote_timestamps_.empty()) {
1273 break;
1274 }
1275 }
1276 scheduled_time_ = monotonic_clock::min_time;
1277
1278 ScheduleTimestamp();
1279}
1280
1281LogReader::RemoteMessageSender *LogReader::State::RemoteTimestampSender(
Austin Schuh61e973f2021-02-21 21:43:56 -08001282 const Channel *channel, const Connection *connection) {
1283 message_bridge::ChannelTimestampFinder finder(event_loop_);
1284 // Look at any pre-created channel/connection pairs.
1285 {
1286 auto it =
1287 channel_timestamp_loggers_.find(std::make_pair(channel, connection));
1288 if (it != channel_timestamp_loggers_.end()) {
1289 return it->second.get();
1290 }
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001291 }
1292
Austin Schuh61e973f2021-02-21 21:43:56 -08001293 // That failed, so resolve the RemoteMessage channel timestamps will be logged
1294 // to.
1295 const Channel *timestamp_channel = finder.ForChannel(channel, connection);
1296
1297 {
1298 // See if that has been created before. If so, cache it in
1299 // channel_timestamp_loggers_ and return.
1300 auto it = timestamp_loggers_.find(timestamp_channel);
1301 if (it != timestamp_loggers_.end()) {
1302 CHECK(channel_timestamp_loggers_
1303 .try_emplace(std::make_pair(channel, connection), it->second)
1304 .second);
1305 return it->second.get();
1306 }
1307 }
1308
1309 // Otherwise, make a sender, save it, and cache it.
1310 auto result = channel_timestamp_loggers_.try_emplace(
1311 std::make_pair(channel, connection),
1312 std::make_shared<RemoteMessageSender>(
1313 event_loop()->MakeSender<RemoteMessage>(
1314 timestamp_channel->name()->string_view()),
1315 event_loop()));
1316
1317 CHECK(timestamp_loggers_.try_emplace(timestamp_channel, result.first->second)
1318 .second);
1319 return result.first->second.get();
Austin Schuh858c9f32020-08-31 16:56:12 -07001320}
1321
Austin Schuhdda74ec2021-01-03 19:30:37 -08001322TimestampedMessage LogReader::State::PopOldest() {
Austin Schuhe639ea12021-01-25 13:00:22 -08001323 CHECK(timestamp_mapper_ != nullptr);
1324 TimestampedMessage *result_ptr = timestamp_mapper_->Front();
1325 CHECK(result_ptr != nullptr);
Austin Schuh858c9f32020-08-31 16:56:12 -07001326
Austin Schuhe639ea12021-01-25 13:00:22 -08001327 TimestampedMessage result = std::move(*result_ptr);
1328
Austin Schuh2f8fd752020-09-01 22:38:28 -07001329 VLOG(2) << MaybeNodeName(event_loop_->node()) << "PopOldest Popping "
Austin Schuhe639ea12021-01-25 13:00:22 -08001330 << result.monotonic_event_time;
1331 timestamp_mapper_->PopFront();
Austin Schuh858c9f32020-08-31 16:56:12 -07001332 SeedSortedMessages();
1333
Austin Schuhe639ea12021-01-25 13:00:22 -08001334 if (result.monotonic_remote_time != monotonic_clock::min_time) {
1335 message_bridge::NoncausalOffsetEstimator *filter =
1336 filters_[result.channel_index];
1337 CHECK(filter != nullptr);
1338
1339 // TODO(austin): We probably want to push this down into the timestamp
1340 // mapper directly.
Austin Schuh3d94be02021-02-12 23:15:20 -08001341 filter->Pop(event_loop_->node(), event_loop_->monotonic_now());
Austin Schuh2f8fd752020-09-01 22:38:28 -07001342 }
Austin Schuh5ee56872021-01-30 16:53:34 -08001343 VLOG(1) << "Popped " << result
1344 << configuration::CleanedChannelToString(
1345 event_loop_->configuration()->channels()->Get(
1346 factory_channel_index_[result.channel_index]));
Austin Schuhe639ea12021-01-25 13:00:22 -08001347 return result;
Austin Schuh858c9f32020-08-31 16:56:12 -07001348}
1349
1350monotonic_clock::time_point LogReader::State::OldestMessageTime() const {
Austin Schuhe639ea12021-01-25 13:00:22 -08001351 if (timestamp_mapper_ == nullptr) {
Austin Schuh287d43d2020-12-04 20:19:33 -08001352 return monotonic_clock::max_time;
1353 }
Austin Schuhe639ea12021-01-25 13:00:22 -08001354 TimestampedMessage *result_ptr = timestamp_mapper_->Front();
1355 if (result_ptr == nullptr) {
1356 return monotonic_clock::max_time;
1357 }
1358 VLOG(2) << MaybeNodeName(event_loop_->node()) << "oldest message at "
1359 << result_ptr->monotonic_event_time;
1360 return result_ptr->monotonic_event_time;
Austin Schuh858c9f32020-08-31 16:56:12 -07001361}
1362
1363void LogReader::State::SeedSortedMessages() {
Austin Schuh287d43d2020-12-04 20:19:33 -08001364 if (!timestamp_mapper_) return;
Austin Schuh858c9f32020-08-31 16:56:12 -07001365
Austin Schuhe639ea12021-01-25 13:00:22 -08001366 timestamp_mapper_->QueueFor(chrono::duration_cast<chrono::seconds>(
1367 chrono::duration<double>(FLAGS_time_estimation_buffer_seconds)));
Austin Schuh858c9f32020-08-31 16:56:12 -07001368}
1369
1370void LogReader::State::Deregister() {
1371 for (size_t i = 0; i < channels_.size(); ++i) {
1372 channels_[i].reset();
1373 }
Austin Schuh61e973f2021-02-21 21:43:56 -08001374 channel_timestamp_loggers_.clear();
1375 timestamp_loggers_.clear();
Austin Schuh858c9f32020-08-31 16:56:12 -07001376 event_loop_unique_ptr_.reset();
1377 event_loop_ = nullptr;
1378 timer_handler_ = nullptr;
1379 node_event_loop_factory_ = nullptr;
1380}
1381
Austin Schuhe309d2a2019-11-29 13:25:21 -08001382} // namespace logger
1383} // namespace aos