blob: 6b7a598d084538f2eddedaa8f6d6ba5dc9fa6294 [file] [log] [blame]
Austin Schuha36c8902019-12-30 18:07:15 -08001#include "aos/events/logging/logfile_utils.h"
2
3#include <fcntl.h>
Austin Schuha36c8902019-12-30 18:07:15 -08004#include <sys/stat.h>
5#include <sys/types.h>
6#include <sys/uio.h>
7
Brian Silvermanf51499a2020-09-21 12:49:08 -07008#include <algorithm>
9#include <climits>
Austin Schuha36c8902019-12-30 18:07:15 -080010
Austin Schuhe4fca832020-03-07 16:58:53 -080011#include "absl/strings/escaping.h"
Austin Schuh05b70472020-01-01 17:11:17 -080012#include "aos/configuration.h"
Austin Schuhfa895892020-01-07 20:07:41 -080013#include "aos/flatbuffer_merge.h"
Austin Schuh6f3babe2020-01-26 20:34:50 -080014#include "aos/util/file.h"
Austin Schuha36c8902019-12-30 18:07:15 -080015#include "flatbuffers/flatbuffers.h"
Austin Schuh05b70472020-01-01 17:11:17 -080016#include "gflags/gflags.h"
17#include "glog/logging.h"
Austin Schuha36c8902019-12-30 18:07:15 -080018
Brian Silvermanf59fe3f2020-09-22 21:04:09 -070019#if defined(__x86_64__)
20#define ENABLE_LZMA 1
21#elif defined(__aarch64__)
22#define ENABLE_LZMA 1
23#else
24#define ENABLE_LZMA 0
25#endif
26
27#if ENABLE_LZMA
28#include "aos/events/logging/lzma_encoder.h"
29#endif
30
Austin Schuh7fbf5a72020-09-21 16:28:13 -070031DEFINE_int32(flush_size, 128000,
Austin Schuha36c8902019-12-30 18:07:15 -080032 "Number of outstanding bytes to allow before flushing to disk.");
33
Brian Silvermanf51499a2020-09-21 12:49:08 -070034namespace aos::logger {
Austin Schuha36c8902019-12-30 18:07:15 -080035
Austin Schuh05b70472020-01-01 17:11:17 -080036namespace chrono = std::chrono;
37
Brian Silvermanf51499a2020-09-21 12:49:08 -070038DetachedBufferWriter::DetachedBufferWriter(
39 std::string_view filename, std::unique_ptr<DetachedBufferEncoder> encoder)
40 : filename_(filename), encoder_(std::move(encoder)) {
Brian Silvermana9f2ec92020-10-06 18:00:53 -070041 if (!util::MkdirPIfSpace(filename, 0777)) {
42 ran_out_of_space_ = true;
43 } else {
44 fd_ = open(std::string(filename).c_str(),
45 O_RDWR | O_CLOEXEC | O_CREAT | O_EXCL, 0774);
46 if (fd_ == -1 && errno == ENOSPC) {
47 ran_out_of_space_ = true;
48 } else {
49 PCHECK(fd_ != -1) << ": Failed to open " << filename << " for writing";
50 VLOG(1) << "Opened " << filename << " for writing";
51 }
52 }
Austin Schuha36c8902019-12-30 18:07:15 -080053}
54
55DetachedBufferWriter::~DetachedBufferWriter() {
Brian Silverman0465fcf2020-09-24 00:29:18 -070056 Close();
57 if (ran_out_of_space_) {
58 CHECK(acknowledge_ran_out_of_space_)
59 << ": Unacknowledged out of disk space, log file was not completed";
Brian Silvermanf51499a2020-09-21 12:49:08 -070060 }
Austin Schuh2f8fd752020-09-01 22:38:28 -070061}
62
Brian Silvermand90905f2020-09-23 14:42:56 -070063DetachedBufferWriter::DetachedBufferWriter(DetachedBufferWriter &&other) {
Austin Schuh2f8fd752020-09-01 22:38:28 -070064 *this = std::move(other);
65}
66
Brian Silverman87ac0402020-09-17 14:47:01 -070067// When other is destroyed "soon" (which it should be because we're getting an
68// rvalue reference to it), it will flush etc all the data we have queued up
69// (because that data will then be its data).
Austin Schuh2f8fd752020-09-01 22:38:28 -070070DetachedBufferWriter &DetachedBufferWriter::operator=(
71 DetachedBufferWriter &&other) {
Austin Schuh2f8fd752020-09-01 22:38:28 -070072 std::swap(filename_, other.filename_);
Brian Silvermanf51499a2020-09-21 12:49:08 -070073 std::swap(encoder_, other.encoder_);
Austin Schuh2f8fd752020-09-01 22:38:28 -070074 std::swap(fd_, other.fd_);
Brian Silverman0465fcf2020-09-24 00:29:18 -070075 std::swap(ran_out_of_space_, other.ran_out_of_space_);
76 std::swap(acknowledge_ran_out_of_space_, other.acknowledge_ran_out_of_space_);
Austin Schuh2f8fd752020-09-01 22:38:28 -070077 std::swap(iovec_, other.iovec_);
Brian Silvermanf51499a2020-09-21 12:49:08 -070078 std::swap(max_write_time_, other.max_write_time_);
79 std::swap(max_write_time_bytes_, other.max_write_time_bytes_);
80 std::swap(max_write_time_messages_, other.max_write_time_messages_);
81 std::swap(total_write_time_, other.total_write_time_);
82 std::swap(total_write_count_, other.total_write_count_);
83 std::swap(total_write_messages_, other.total_write_messages_);
84 std::swap(total_write_bytes_, other.total_write_bytes_);
Austin Schuh2f8fd752020-09-01 22:38:28 -070085 return *this;
Austin Schuha36c8902019-12-30 18:07:15 -080086}
87
Brian Silvermanf51499a2020-09-21 12:49:08 -070088void DetachedBufferWriter::QueueSpan(absl::Span<const uint8_t> span) {
Brian Silvermana9f2ec92020-10-06 18:00:53 -070089 if (ran_out_of_space_) {
90 // We don't want any later data to be written after space becomes
91 // available, so refuse to write anything more once we've dropped data
92 // because we ran out of space.
93 VLOG(1) << "Ignoring span: " << span.size();
94 return;
95 }
96
Brian Silvermanf51499a2020-09-21 12:49:08 -070097 if (encoder_->may_bypass() && span.size() > 4096u) {
98 // Over this threshold, we'll assume it's cheaper to add an extra
99 // syscall to write the data immediately instead of copying it to
100 // enqueue.
Austin Schuha36c8902019-12-30 18:07:15 -0800101
Brian Silvermanf51499a2020-09-21 12:49:08 -0700102 // First, flush everything.
103 while (encoder_->queue_size() > 0u) {
104 Flush();
105 }
Austin Schuhde031b72020-01-10 19:34:41 -0800106
Brian Silvermanf51499a2020-09-21 12:49:08 -0700107 // Then, write it directly.
108 const auto start = aos::monotonic_clock::now();
109 const ssize_t written = write(fd_, span.data(), span.size());
110 const auto end = aos::monotonic_clock::now();
Brian Silverman0465fcf2020-09-24 00:29:18 -0700111 HandleWriteReturn(written, span.size());
Brian Silvermanf51499a2020-09-21 12:49:08 -0700112 UpdateStatsForWrite(end - start, written, 1);
113 } else {
114 encoder_->Encode(CopySpanAsDetachedBuffer(span));
Austin Schuha36c8902019-12-30 18:07:15 -0800115 }
Brian Silvermanf51499a2020-09-21 12:49:08 -0700116
117 FlushAtThreshold();
Austin Schuha36c8902019-12-30 18:07:15 -0800118}
119
Brian Silverman0465fcf2020-09-24 00:29:18 -0700120void DetachedBufferWriter::Close() {
121 if (fd_ == -1) {
122 return;
123 }
124 encoder_->Finish();
125 while (encoder_->queue_size() > 0) {
126 Flush();
127 }
128 if (close(fd_) == -1) {
129 if (errno == ENOSPC) {
130 ran_out_of_space_ = true;
131 } else {
132 PLOG(ERROR) << "Closing log file failed";
133 }
134 }
135 fd_ = -1;
136 VLOG(1) << "Closed " << filename_;
137}
138
Austin Schuha36c8902019-12-30 18:07:15 -0800139void DetachedBufferWriter::Flush() {
Brian Silvermanf51499a2020-09-21 12:49:08 -0700140 const auto queue = encoder_->queue();
141 if (queue.empty()) {
Austin Schuha36c8902019-12-30 18:07:15 -0800142 return;
143 }
Brian Silverman0465fcf2020-09-24 00:29:18 -0700144 if (ran_out_of_space_) {
145 // We don't want any later data to be written after space becomes available,
146 // so refuse to write anything more once we've dropped data because we ran
147 // out of space.
148 VLOG(1) << "Ignoring queue: " << queue.size();
149 encoder_->Clear(queue.size());
150 return;
151 }
Brian Silvermanf51499a2020-09-21 12:49:08 -0700152
Austin Schuha36c8902019-12-30 18:07:15 -0800153 iovec_.clear();
Brian Silvermanf51499a2020-09-21 12:49:08 -0700154 const size_t iovec_size = std::min<size_t>(queue.size(), IOV_MAX);
155 iovec_.resize(iovec_size);
Austin Schuha36c8902019-12-30 18:07:15 -0800156 size_t counted_size = 0;
Brian Silvermanf51499a2020-09-21 12:49:08 -0700157 for (size_t i = 0; i < iovec_size; ++i) {
158 iovec_[i].iov_base = const_cast<uint8_t *>(queue[i].data());
159 iovec_[i].iov_len = queue[i].size();
160 counted_size += iovec_[i].iov_len;
Austin Schuha36c8902019-12-30 18:07:15 -0800161 }
Brian Silvermanf51499a2020-09-21 12:49:08 -0700162
163 const auto start = aos::monotonic_clock::now();
Austin Schuha36c8902019-12-30 18:07:15 -0800164 const ssize_t written = writev(fd_, iovec_.data(), iovec_.size());
Brian Silvermanf51499a2020-09-21 12:49:08 -0700165 const auto end = aos::monotonic_clock::now();
Brian Silverman0465fcf2020-09-24 00:29:18 -0700166 HandleWriteReturn(written, counted_size);
Brian Silvermanf51499a2020-09-21 12:49:08 -0700167
168 encoder_->Clear(iovec_size);
169
170 UpdateStatsForWrite(end - start, written, iovec_size);
171}
172
Brian Silverman0465fcf2020-09-24 00:29:18 -0700173void DetachedBufferWriter::HandleWriteReturn(ssize_t write_return,
174 size_t write_size) {
175 if (write_return == -1 && errno == ENOSPC) {
176 ran_out_of_space_ = true;
177 return;
178 }
179 PCHECK(write_return >= 0) << ": write failed";
180 if (write_return < static_cast<ssize_t>(write_size)) {
181 // Sometimes this happens instead of ENOSPC. On a real filesystem, this
182 // never seems to happen in any other case. If we ever want to log to a
183 // socket, this will happen more often. However, until we get there, we'll
184 // just assume it means we ran out of space.
185 ran_out_of_space_ = true;
186 return;
187 }
188}
189
Brian Silvermanf51499a2020-09-21 12:49:08 -0700190void DetachedBufferWriter::UpdateStatsForWrite(
191 aos::monotonic_clock::duration duration, ssize_t written, int iovec_size) {
192 if (duration > max_write_time_) {
193 max_write_time_ = duration;
194 max_write_time_bytes_ = written;
195 max_write_time_messages_ = iovec_size;
196 }
197 total_write_time_ += duration;
198 ++total_write_count_;
199 total_write_messages_ += iovec_size;
200 total_write_bytes_ += written;
201}
202
203void DetachedBufferWriter::FlushAtThreshold() {
204 // Flush if we are at the max number of iovs per writev, because there's no
205 // point queueing up any more data in memory. Also flush once we have enough
206 // data queued up.
207 while (encoder_->queued_bytes() > static_cast<size_t>(FLAGS_flush_size) ||
208 encoder_->queue_size() >= IOV_MAX) {
209 Flush();
210 }
Austin Schuha36c8902019-12-30 18:07:15 -0800211}
212
213flatbuffers::Offset<MessageHeader> PackMessage(
214 flatbuffers::FlatBufferBuilder *fbb, const Context &context,
215 int channel_index, LogType log_type) {
216 flatbuffers::Offset<flatbuffers::Vector<uint8_t>> data_offset;
217
218 switch (log_type) {
219 case LogType::kLogMessage:
220 case LogType::kLogMessageAndDeliveryTime:
Austin Schuh6f3babe2020-01-26 20:34:50 -0800221 case LogType::kLogRemoteMessage:
Brian Silvermaneaa41d62020-07-08 19:47:35 -0700222 data_offset = fbb->CreateVector(
223 static_cast<const uint8_t *>(context.data), context.size);
Austin Schuha36c8902019-12-30 18:07:15 -0800224 break;
225
226 case LogType::kLogDeliveryTimeOnly:
227 break;
228 }
229
230 MessageHeader::Builder message_header_builder(*fbb);
231 message_header_builder.add_channel_index(channel_index);
Austin Schuh6f3babe2020-01-26 20:34:50 -0800232
233 switch (log_type) {
234 case LogType::kLogRemoteMessage:
235 message_header_builder.add_queue_index(context.remote_queue_index);
236 message_header_builder.add_monotonic_sent_time(
237 context.monotonic_remote_time.time_since_epoch().count());
238 message_header_builder.add_realtime_sent_time(
239 context.realtime_remote_time.time_since_epoch().count());
240 break;
241
242 case LogType::kLogMessage:
243 case LogType::kLogMessageAndDeliveryTime:
244 case LogType::kLogDeliveryTimeOnly:
245 message_header_builder.add_queue_index(context.queue_index);
246 message_header_builder.add_monotonic_sent_time(
247 context.monotonic_event_time.time_since_epoch().count());
248 message_header_builder.add_realtime_sent_time(
249 context.realtime_event_time.time_since_epoch().count());
250 break;
251 }
Austin Schuha36c8902019-12-30 18:07:15 -0800252
253 switch (log_type) {
254 case LogType::kLogMessage:
Austin Schuh6f3babe2020-01-26 20:34:50 -0800255 case LogType::kLogRemoteMessage:
Austin Schuha36c8902019-12-30 18:07:15 -0800256 message_header_builder.add_data(data_offset);
257 break;
258
259 case LogType::kLogMessageAndDeliveryTime:
260 message_header_builder.add_data(data_offset);
261 [[fallthrough]];
262
263 case LogType::kLogDeliveryTimeOnly:
264 message_header_builder.add_monotonic_remote_time(
265 context.monotonic_remote_time.time_since_epoch().count());
266 message_header_builder.add_realtime_remote_time(
267 context.realtime_remote_time.time_since_epoch().count());
268 message_header_builder.add_remote_queue_index(context.remote_queue_index);
269 break;
270 }
271
272 return message_header_builder.Finish();
273}
274
Brian Silvermanf51499a2020-09-21 12:49:08 -0700275SpanReader::SpanReader(std::string_view filename) : filename_(filename) {
Brian Silvermanf59fe3f2020-09-22 21:04:09 -0700276 static const std::string_view kXz = ".xz";
277 if (filename.substr(filename.size() - kXz.size()) == kXz) {
278#if ENABLE_LZMA
279 decoder_ = std::make_unique<LzmaDecoder>(filename);
280#else
281 LOG(FATAL) << "Reading xz-compressed files not supported on this platform";
282#endif
283 } else {
284 decoder_ = std::make_unique<DummyDecoder>(filename);
285 }
Austin Schuh05b70472020-01-01 17:11:17 -0800286}
287
288absl::Span<const uint8_t> SpanReader::ReadMessage() {
289 // Make sure we have enough for the size.
290 if (data_.size() - consumed_data_ < sizeof(flatbuffers::uoffset_t)) {
291 if (!ReadBlock()) {
292 return absl::Span<const uint8_t>();
293 }
294 }
295
296 // Now make sure we have enough for the message.
297 const size_t data_size =
298 flatbuffers::GetPrefixedSize(data_.data() + consumed_data_) +
299 sizeof(flatbuffers::uoffset_t);
Austin Schuhe4fca832020-03-07 16:58:53 -0800300 if (data_size == sizeof(flatbuffers::uoffset_t)) {
301 LOG(ERROR) << "Size of data is zero. Log file end is corrupted, skipping.";
302 LOG(ERROR) << " Rest of log file is "
303 << absl::BytesToHexString(std::string_view(
304 reinterpret_cast<const char *>(data_.data() +
305 consumed_data_),
306 data_.size() - consumed_data_));
307 return absl::Span<const uint8_t>();
308 }
Austin Schuh05b70472020-01-01 17:11:17 -0800309 while (data_.size() < consumed_data_ + data_size) {
310 if (!ReadBlock()) {
311 return absl::Span<const uint8_t>();
312 }
313 }
314
315 // And return it, consuming the data.
316 const uint8_t *data_ptr = data_.data() + consumed_data_;
317
318 consumed_data_ += data_size;
319
320 return absl::Span<const uint8_t>(data_ptr, data_size);
321}
322
Austin Schuh05b70472020-01-01 17:11:17 -0800323bool SpanReader::ReadBlock() {
Brian Silvermanf51499a2020-09-21 12:49:08 -0700324 // This is the amount of data we grab at a time. Doing larger chunks minimizes
325 // syscalls and helps decompressors batch things more efficiently.
Austin Schuh05b70472020-01-01 17:11:17 -0800326 constexpr size_t kReadSize = 256 * 1024;
327
328 // Strip off any unused data at the front.
329 if (consumed_data_ != 0) {
Brian Silvermanf51499a2020-09-21 12:49:08 -0700330 data_.erase_front(consumed_data_);
Austin Schuh05b70472020-01-01 17:11:17 -0800331 consumed_data_ = 0;
332 }
333
334 const size_t starting_size = data_.size();
335
336 // This should automatically grow the backing store. It won't shrink if we
337 // get a small chunk later. This reduces allocations when we want to append
338 // more data.
Brian Silvermanf51499a2020-09-21 12:49:08 -0700339 data_.resize(starting_size + kReadSize);
Austin Schuh05b70472020-01-01 17:11:17 -0800340
Brian Silvermanf51499a2020-09-21 12:49:08 -0700341 const size_t count =
342 decoder_->Read(data_.begin() + starting_size, data_.end());
343 data_.resize(starting_size + count);
Austin Schuh05b70472020-01-01 17:11:17 -0800344 if (count == 0) {
Austin Schuh05b70472020-01-01 17:11:17 -0800345 return false;
346 }
Austin Schuh05b70472020-01-01 17:11:17 -0800347
348 return true;
349}
350
Austin Schuh6f3babe2020-01-26 20:34:50 -0800351FlatbufferVector<LogFileHeader> ReadHeader(std::string_view filename) {
352 SpanReader span_reader(filename);
Austin Schuh6f3babe2020-01-26 20:34:50 -0800353 absl::Span<const uint8_t> config_data = span_reader.ReadMessage();
354
355 // Make sure something was read.
Austin Schuh97789fc2020-08-01 14:42:45 -0700356 CHECK(config_data != absl::Span<const uint8_t>())
357 << ": Failed to read header from: " << filename;
Austin Schuh6f3babe2020-01-26 20:34:50 -0800358
Austin Schuh5212cad2020-09-09 23:12:09 -0700359 // And copy the config so we have it forever, removing the size prefix.
Brian Silverman354697a2020-09-22 21:06:32 -0700360 ResizeableBuffer data;
361 data.resize(config_data.size() - sizeof(flatbuffers::uoffset_t));
362 memcpy(data.data(), config_data.begin() + sizeof(flatbuffers::uoffset_t),
363 data.size());
Austin Schuh6f3babe2020-01-26 20:34:50 -0800364 return FlatbufferVector<LogFileHeader>(std::move(data));
365}
366
Austin Schuh5212cad2020-09-09 23:12:09 -0700367FlatbufferVector<MessageHeader> ReadNthMessage(std::string_view filename,
368 size_t n) {
369 SpanReader span_reader(filename);
370 absl::Span<const uint8_t> data_span = span_reader.ReadMessage();
371 for (size_t i = 0; i < n + 1; ++i) {
372 data_span = span_reader.ReadMessage();
373
374 // Make sure something was read.
375 CHECK(data_span != absl::Span<const uint8_t>())
376 << ": Failed to read data from: " << filename;
377 }
378
Brian Silverman354697a2020-09-22 21:06:32 -0700379 // And copy the config so we have it forever, removing the size prefix.
380 ResizeableBuffer data;
381 data.resize(data_span.size() - sizeof(flatbuffers::uoffset_t));
382 memcpy(data.data(), data_span.begin() + sizeof(flatbuffers::uoffset_t),
383 data.size());
Austin Schuh5212cad2020-09-09 23:12:09 -0700384 return FlatbufferVector<MessageHeader>(std::move(data));
385}
386
Austin Schuh05b70472020-01-01 17:11:17 -0800387MessageReader::MessageReader(std::string_view filename)
Austin Schuh97789fc2020-08-01 14:42:45 -0700388 : span_reader_(filename),
389 raw_log_file_header_(FlatbufferVector<LogFileHeader>::Empty()) {
Austin Schuh05b70472020-01-01 17:11:17 -0800390 // Make sure we have enough to read the size.
Austin Schuh97789fc2020-08-01 14:42:45 -0700391 absl::Span<const uint8_t> header_data = span_reader_.ReadMessage();
Austin Schuh05b70472020-01-01 17:11:17 -0800392
393 // Make sure something was read.
Austin Schuh97789fc2020-08-01 14:42:45 -0700394 CHECK(header_data != absl::Span<const uint8_t>())
395 << ": Failed to read header from: " << filename;
Austin Schuh05b70472020-01-01 17:11:17 -0800396
Austin Schuh97789fc2020-08-01 14:42:45 -0700397 // And copy the header data so we have it forever.
Brian Silverman354697a2020-09-22 21:06:32 -0700398 ResizeableBuffer header_data_copy;
399 header_data_copy.resize(header_data.size() - sizeof(flatbuffers::uoffset_t));
400 memcpy(header_data_copy.data(),
401 header_data.begin() + sizeof(flatbuffers::uoffset_t),
402 header_data_copy.size());
Austin Schuh97789fc2020-08-01 14:42:45 -0700403 raw_log_file_header_ =
404 FlatbufferVector<LogFileHeader>(std::move(header_data_copy));
Austin Schuh05b70472020-01-01 17:11:17 -0800405
Austin Schuhcde938c2020-02-02 17:30:07 -0800406 max_out_of_order_duration_ =
Austin Schuh2f8fd752020-09-01 22:38:28 -0700407 chrono::nanoseconds(log_file_header()->max_out_of_order_duration());
Austin Schuhcde938c2020-02-02 17:30:07 -0800408
409 VLOG(1) << "Opened " << filename << " as node "
410 << FlatbufferToJson(log_file_header()->node());
Austin Schuh05b70472020-01-01 17:11:17 -0800411}
412
413std::optional<FlatbufferVector<MessageHeader>> MessageReader::ReadMessage() {
414 absl::Span<const uint8_t> msg_data = span_reader_.ReadMessage();
415 if (msg_data == absl::Span<const uint8_t>()) {
416 return std::nullopt;
417 }
418
Brian Silverman354697a2020-09-22 21:06:32 -0700419 ResizeableBuffer result_buffer;
420 result_buffer.resize(msg_data.size() - sizeof(flatbuffers::uoffset_t));
421 memcpy(result_buffer.data(),
422 msg_data.begin() + sizeof(flatbuffers::uoffset_t),
423 result_buffer.size());
424 FlatbufferVector<MessageHeader> result(std::move(result_buffer));
Austin Schuh05b70472020-01-01 17:11:17 -0800425
426 const monotonic_clock::time_point timestamp = monotonic_clock::time_point(
427 chrono::nanoseconds(result.message().monotonic_sent_time()));
428
429 newest_timestamp_ = std::max(newest_timestamp_, timestamp);
Austin Schuh8bd96322020-02-13 21:18:22 -0800430 VLOG(2) << "Read from " << filename() << " data " << FlatbufferToJson(result);
Austin Schuh6f3babe2020-01-26 20:34:50 -0800431 return std::move(result);
Austin Schuh05b70472020-01-01 17:11:17 -0800432}
433
Austin Schuhc41603c2020-10-11 16:17:37 -0700434PartsMessageReader::PartsMessageReader(LogParts log_parts)
435 : parts_(std::move(log_parts)), message_reader_(parts_.parts[0]) {}
436
437std::optional<FlatbufferVector<MessageHeader>>
438PartsMessageReader::ReadMessage() {
439 while (!done_) {
440 std::optional<FlatbufferVector<MessageHeader>> message =
441 message_reader_.ReadMessage();
442 if (message) {
443 newest_timestamp_ = message_reader_.newest_timestamp();
444 return message;
445 }
446 NextLog();
447 }
448 return std::nullopt;
449}
450
451void PartsMessageReader::NextLog() {
452 if (next_part_index_ == parts_.parts.size()) {
453 done_ = true;
454 return;
455 }
456 message_reader_ = MessageReader(parts_.parts[next_part_index_]);
457 ++next_part_index_;
458}
459
Austin Schuh6f3babe2020-01-26 20:34:50 -0800460SplitMessageReader::SplitMessageReader(
Austin Schuhfa895892020-01-07 20:07:41 -0800461 const std::vector<std::string> &filenames)
462 : filenames_(filenames),
Austin Schuh97789fc2020-08-01 14:42:45 -0700463 log_file_header_(FlatbufferVector<LogFileHeader>::Empty()) {
Austin Schuhfa895892020-01-07 20:07:41 -0800464 CHECK(NextLogFile()) << ": filenames is empty. Need files to read.";
465
Austin Schuh6f3babe2020-01-26 20:34:50 -0800466 // Grab any log file header. They should all match (and we will check as we
467 // open more of them).
Austin Schuh97789fc2020-08-01 14:42:45 -0700468 log_file_header_ = message_reader_->raw_log_file_header();
Austin Schuhfa895892020-01-07 20:07:41 -0800469
Austin Schuh2f8fd752020-09-01 22:38:28 -0700470 for (size_t i = 1; i < filenames_.size(); ++i) {
471 MessageReader message_reader(filenames_[i]);
472
473 const monotonic_clock::time_point new_monotonic_start_time(
474 chrono::nanoseconds(
475 message_reader.log_file_header()->monotonic_start_time()));
476 const realtime_clock::time_point new_realtime_start_time(
477 chrono::nanoseconds(
478 message_reader.log_file_header()->realtime_start_time()));
479
480 // There are 2 types of part files. Part files from before time estimation
481 // has started, and part files after. We don't declare a log file "started"
482 // until time estimation is up. And once a log file starts, it should never
483 // stop again, and should remain constant.
484 // To compare both types of headers, we mutate our saved copy of the header
485 // to match the next chunk by updating time if we detect a stopped ->
486 // started transition.
487 if (monotonic_start_time() == monotonic_clock::min_time) {
488 CHECK_EQ(realtime_start_time(), realtime_clock::min_time);
489 // We should only be missing the monotonic start time when logging data
Brian Silverman87ac0402020-09-17 14:47:01 -0700490 // for remote nodes. We don't have a good way to determine the remote
Austin Schuh2f8fd752020-09-01 22:38:28 -0700491 // realtime offset, so it shouldn't be filled out.
492 // TODO(austin): If we have a good way, feel free to fill it out. It
493 // probably won't be better than we could do in post though with the same
494 // data.
495 CHECK(!log_file_header_.mutable_message()->has_realtime_start_time());
496 if (new_monotonic_start_time != monotonic_clock::min_time) {
497 // If we finally found our start time, update the header. Do this once
498 // because it should never change again.
499 log_file_header_.mutable_message()->mutate_monotonic_start_time(
500 new_monotonic_start_time.time_since_epoch().count());
501 log_file_header_.mutable_message()->mutate_realtime_start_time(
502 new_realtime_start_time.time_since_epoch().count());
503 }
504 }
505
Austin Schuh64fab802020-09-09 22:47:47 -0700506 // We don't have a good way to set the realtime start time on remote nodes.
507 // Confirm it remains consistent.
508 CHECK_EQ(log_file_header_.mutable_message()->has_realtime_start_time(),
509 message_reader.log_file_header()->has_realtime_start_time());
510
511 // Parts index will *not* match unless we set them to match. We only want
512 // to accept the start time and parts mismatching, so set them.
513 log_file_header_.mutable_message()->mutate_parts_index(
514 message_reader.log_file_header()->parts_index());
515
Austin Schuh2f8fd752020-09-01 22:38:28 -0700516 // Now compare that the headers match.
Austin Schuh64fab802020-09-09 22:47:47 -0700517 if (!CompareFlatBuffer(message_reader.raw_log_file_header(),
518 log_file_header_)) {
Brian Silvermanae7c0332020-09-30 16:58:23 -0700519 if (message_reader.log_file_header()->has_log_event_uuid() &&
520 log_file_header_.message().has_log_event_uuid() &&
521 message_reader.log_file_header()->log_event_uuid()->string_view() !=
522 log_file_header_.message().log_event_uuid()->string_view()) {
Austin Schuh64fab802020-09-09 22:47:47 -0700523 LOG(FATAL) << "Logger UUIDs don't match between log file chunks "
524 << filenames_[0] << " and " << filenames_[i]
525 << ", this is not supported.";
526 }
527 if (message_reader.log_file_header()->has_parts_uuid() &&
528 log_file_header_.message().has_parts_uuid() &&
529 message_reader.log_file_header()->parts_uuid()->string_view() !=
530 log_file_header_.message().parts_uuid()->string_view()) {
531 LOG(FATAL) << "Parts UUIDs don't match between log file chunks "
532 << filenames_[0] << " and " << filenames_[i]
533 << ", this is not supported.";
534 }
535
536 LOG(FATAL) << "Header is different between log file chunks "
537 << filenames_[0] << " and " << filenames_[i]
538 << ", this is not supported.";
539 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700540 }
Austin Schuh64fab802020-09-09 22:47:47 -0700541 // Put the parts index back to the first log file chunk.
542 log_file_header_.mutable_message()->mutate_parts_index(
543 message_reader_->log_file_header()->parts_index());
Austin Schuh2f8fd752020-09-01 22:38:28 -0700544
Austin Schuh6f3babe2020-01-26 20:34:50 -0800545 // Setup per channel state.
Austin Schuh05b70472020-01-01 17:11:17 -0800546 channels_.resize(configuration()->channels()->size());
Austin Schuh6f3babe2020-01-26 20:34:50 -0800547 for (ChannelData &channel_data : channels_) {
548 channel_data.data.split_reader = this;
549 // Build up the timestamp list.
550 if (configuration::MultiNode(configuration())) {
551 channel_data.timestamps.resize(configuration()->nodes()->size());
552 for (MessageHeaderQueue &queue : channel_data.timestamps) {
553 queue.timestamps = true;
554 queue.split_reader = this;
555 }
556 }
557 }
Austin Schuh05b70472020-01-01 17:11:17 -0800558
Austin Schuh6f3babe2020-01-26 20:34:50 -0800559 // Build up channels_to_write_ as an optimization to make it fast to figure
560 // out which datastructure to place any new data from a channel on.
561 for (const Channel *channel : *configuration()->channels()) {
562 // This is the main case. We will only see data on this node.
563 if (configuration::ChannelIsSendableOnNode(channel, node())) {
564 channels_to_write_.emplace_back(
565 &channels_[channels_to_write_.size()].data);
566 } else
567 // If we can't send, but can receive, we should be able to see
568 // timestamps here.
569 if (configuration::ChannelIsReadableOnNode(channel, node())) {
570 channels_to_write_.emplace_back(
571 &(channels_[channels_to_write_.size()]
572 .timestamps[configuration::GetNodeIndex(configuration(),
573 node())]));
574 } else {
575 channels_to_write_.emplace_back(nullptr);
576 }
577 }
Austin Schuh05b70472020-01-01 17:11:17 -0800578}
579
Austin Schuh6f3babe2020-01-26 20:34:50 -0800580bool SplitMessageReader::NextLogFile() {
Austin Schuhfa895892020-01-07 20:07:41 -0800581 if (next_filename_index_ == filenames_.size()) {
582 return false;
583 }
584 message_reader_ =
585 std::make_unique<MessageReader>(filenames_[next_filename_index_]);
586
587 // We can't support the config diverging between two log file headers. See if
588 // they are the same.
589 if (next_filename_index_ != 0) {
Austin Schuh64fab802020-09-09 22:47:47 -0700590 // In order for the headers to identically compare, they need to have the
591 // same parts_index. Rewrite the saved header with the new parts_index,
592 // compare, and then restore.
593 const int32_t original_parts_index =
594 log_file_header_.message().parts_index();
595 log_file_header_.mutable_message()->mutate_parts_index(
596 message_reader_->log_file_header()->parts_index());
597
Austin Schuh97789fc2020-08-01 14:42:45 -0700598 CHECK(CompareFlatBuffer(message_reader_->raw_log_file_header(),
599 log_file_header_))
Austin Schuhfa895892020-01-07 20:07:41 -0800600 << ": Header is different between log file chunks "
601 << filenames_[next_filename_index_] << " and "
602 << filenames_[next_filename_index_ - 1] << ", this is not supported.";
Austin Schuh64fab802020-09-09 22:47:47 -0700603
604 log_file_header_.mutable_message()->mutate_parts_index(
605 original_parts_index);
Austin Schuhfa895892020-01-07 20:07:41 -0800606 }
607
608 ++next_filename_index_;
609 return true;
610}
611
Austin Schuh6f3babe2020-01-26 20:34:50 -0800612bool SplitMessageReader::QueueMessages(
Austin Schuhcde938c2020-02-02 17:30:07 -0800613 monotonic_clock::time_point last_dequeued_time) {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800614 // TODO(austin): Once we are happy that everything works, read a 256kb chunk
615 // to reduce the need to re-heap down below.
Austin Schuhcde938c2020-02-02 17:30:07 -0800616
617 // Special case no more data. Otherwise we blow up on the CHECK statement
618 // confirming that we have enough data queued.
619 if (at_end_) {
620 return false;
621 }
622
623 // If this isn't the first time around, confirm that we had enough data queued
624 // to follow the contract.
625 if (time_to_queue_ != monotonic_clock::min_time) {
626 CHECK_LE(last_dequeued_time,
627 newest_timestamp() - max_out_of_order_duration())
628 << " node " << FlatbufferToJson(node()) << " on " << this;
629
630 // Bail if there is enough data already queued.
631 if (last_dequeued_time < time_to_queue_) {
Austin Schuhee711052020-08-24 16:06:09 -0700632 VLOG(1) << MaybeNodeName(target_node_) << "All up to date on " << this
633 << ", dequeued " << last_dequeued_time << " queue time "
634 << time_to_queue_;
Austin Schuhcde938c2020-02-02 17:30:07 -0800635 return true;
636 }
637 } else {
638 // Startup takes a special dance. We want to queue up until the start time,
639 // but we then want to find the next message to read. The conservative
640 // answer is to immediately trigger a second requeue to get things moving.
641 time_to_queue_ = monotonic_start_time();
Austin Schuheeba0292020-10-11 16:20:05 -0700642 CHECK_NE(time_to_queue_, monotonic_clock::min_time);
Austin Schuhcde938c2020-02-02 17:30:07 -0800643 QueueMessages(time_to_queue_);
644 }
645
646 // If we are asked to queue, queue for at least max_out_of_order_duration past
647 // the last known time in the log file (ie the newest timestep read). As long
648 // as we requeue exactly when time_to_queue_ is dequeued and go no further, we
649 // are safe. And since we pop in order, that works.
650 //
651 // Special case the start of the log file. There should be at most 1 message
652 // from each channel at the start of the log file. So always force the start
653 // of the log file to just be read.
654 time_to_queue_ = std::max(time_to_queue_, newest_timestamp());
Austin Schuhee711052020-08-24 16:06:09 -0700655 VLOG(1) << MaybeNodeName(target_node_) << "Queueing, going until "
656 << time_to_queue_ << " " << filename();
Austin Schuhcde938c2020-02-02 17:30:07 -0800657
658 bool was_emplaced = false;
Austin Schuh6f3babe2020-01-26 20:34:50 -0800659 while (true) {
Austin Schuhcde938c2020-02-02 17:30:07 -0800660 // Stop if we have enough.
Brian Silverman98360e22020-04-28 16:51:20 -0700661 if (newest_timestamp() > time_to_queue_ + max_out_of_order_duration() &&
Austin Schuhcde938c2020-02-02 17:30:07 -0800662 was_emplaced) {
Austin Schuhee711052020-08-24 16:06:09 -0700663 VLOG(1) << MaybeNodeName(target_node_) << "Done queueing on " << this
664 << ", queued to " << newest_timestamp() << " with requeue time "
665 << time_to_queue_;
Austin Schuh6f3babe2020-01-26 20:34:50 -0800666 return true;
667 }
Austin Schuh05b70472020-01-01 17:11:17 -0800668
Austin Schuh6f3babe2020-01-26 20:34:50 -0800669 if (std::optional<FlatbufferVector<MessageHeader>> msg =
670 message_reader_->ReadMessage()) {
671 const MessageHeader &header = msg.value().message();
672
Austin Schuhcde938c2020-02-02 17:30:07 -0800673 const monotonic_clock::time_point timestamp = monotonic_clock::time_point(
674 chrono::nanoseconds(header.monotonic_sent_time()));
Austin Schuh6f3babe2020-01-26 20:34:50 -0800675
Austin Schuh0b5fd032020-03-28 17:36:49 -0700676 if (VLOG_IS_ON(2)) {
Brian Silvermand90905f2020-09-23 14:42:56 -0700677 LOG(INFO) << MaybeNodeName(target_node_) << "Queued " << this << " "
678 << filename() << " ttq: " << time_to_queue_ << " now "
Austin Schuhee711052020-08-24 16:06:09 -0700679 << newest_timestamp() << " start time "
680 << monotonic_start_time() << " " << FlatbufferToJson(&header);
Austin Schuh0b5fd032020-03-28 17:36:49 -0700681 } else if (VLOG_IS_ON(1)) {
682 FlatbufferVector<MessageHeader> copy = msg.value();
683 copy.mutable_message()->clear_data();
Austin Schuhee711052020-08-24 16:06:09 -0700684 LOG(INFO) << MaybeNodeName(target_node_) << "Queued " << this << " "
685 << filename() << " ttq: " << time_to_queue_ << " now "
686 << newest_timestamp() << " start time "
687 << monotonic_start_time() << " " << FlatbufferToJson(copy);
Austin Schuh0b5fd032020-03-28 17:36:49 -0700688 }
Austin Schuhcde938c2020-02-02 17:30:07 -0800689
690 const int channel_index = header.channel_index();
691 was_emplaced = channels_to_write_[channel_index]->emplace_back(
692 std::move(msg.value()));
693 if (was_emplaced) {
694 newest_timestamp_ = std::max(newest_timestamp_, timestamp);
695 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800696 } else {
697 if (!NextLogFile()) {
Austin Schuhee711052020-08-24 16:06:09 -0700698 VLOG(1) << MaybeNodeName(target_node_) << "No more files, last was "
699 << filenames_.back();
Austin Schuhcde938c2020-02-02 17:30:07 -0800700 at_end_ = true;
Austin Schuh8bd96322020-02-13 21:18:22 -0800701 for (MessageHeaderQueue *queue : channels_to_write_) {
702 if (queue == nullptr || queue->timestamp_merger == nullptr) {
703 continue;
704 }
705 queue->timestamp_merger->NoticeAtEnd();
706 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800707 return false;
708 }
709 }
Austin Schuh05b70472020-01-01 17:11:17 -0800710 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800711}
712
713void SplitMessageReader::SetTimestampMerger(TimestampMerger *timestamp_merger,
714 int channel_index,
715 const Node *target_node) {
716 const Node *reinterpreted_target_node =
717 configuration::GetNodeOrDie(configuration(), target_node);
Austin Schuhee711052020-08-24 16:06:09 -0700718 target_node_ = reinterpreted_target_node;
719
Austin Schuh6f3babe2020-01-26 20:34:50 -0800720 const Channel *const channel =
721 configuration()->channels()->Get(channel_index);
722
Austin Schuhcde938c2020-02-02 17:30:07 -0800723 VLOG(1) << " Configuring merger " << this << " for channel " << channel_index
724 << " "
725 << configuration::CleanedChannelToString(
726 configuration()->channels()->Get(channel_index));
727
Austin Schuh6f3babe2020-01-26 20:34:50 -0800728 MessageHeaderQueue *message_header_queue = nullptr;
729
730 // Figure out if this log file is from our point of view, or the other node's
731 // point of view.
732 if (node() == reinterpreted_target_node) {
Austin Schuhcde938c2020-02-02 17:30:07 -0800733 VLOG(1) << " Replaying as logged node " << filename();
734
735 if (configuration::ChannelIsSendableOnNode(channel, node())) {
736 VLOG(1) << " Data on node";
737 message_header_queue = &(channels_[channel_index].data);
738 } else if (configuration::ChannelIsReadableOnNode(channel, node())) {
739 VLOG(1) << " Timestamps on node";
740 message_header_queue =
741 &(channels_[channel_index].timestamps[configuration::GetNodeIndex(
742 configuration(), node())]);
Austin Schuh6f3babe2020-01-26 20:34:50 -0800743 } else {
Austin Schuhcde938c2020-02-02 17:30:07 -0800744 VLOG(1) << " Dropping";
Austin Schuh6f3babe2020-01-26 20:34:50 -0800745 }
746 } else {
Austin Schuhcde938c2020-02-02 17:30:07 -0800747 VLOG(1) << " Replaying as other node " << filename();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800748 // We are replaying from another node's point of view. The only interesting
Austin Schuhcde938c2020-02-02 17:30:07 -0800749 // data is data that is sent from our node and received on theirs.
750 if (configuration::ChannelIsReadableOnNode(channel,
751 reinterpreted_target_node) &&
752 configuration::ChannelIsSendableOnNode(channel, node())) {
753 VLOG(1) << " Readable on target node";
Austin Schuh6f3babe2020-01-26 20:34:50 -0800754 // Data from another node.
755 message_header_queue = &(channels_[channel_index].data);
756 } else {
Austin Schuhcde938c2020-02-02 17:30:07 -0800757 VLOG(1) << " Dropping";
Austin Schuh6f3babe2020-01-26 20:34:50 -0800758 // This is either not sendable on the other node, or is a timestamp and
759 // therefore not interesting.
760 }
761 }
762
763 // If we found one, write it down. This will be nullptr when there is nothing
764 // relevant on this channel on this node for the target node. In that case,
765 // we want to drop the message instead of queueing it.
766 if (message_header_queue != nullptr) {
767 message_header_queue->timestamp_merger = timestamp_merger;
768 }
769}
770
771std::tuple<monotonic_clock::time_point, uint32_t,
772 FlatbufferVector<MessageHeader>>
773SplitMessageReader::PopOldest(int channel_index) {
774 CHECK_GT(channels_[channel_index].data.size(), 0u);
Austin Schuhcde938c2020-02-02 17:30:07 -0800775 const std::tuple<monotonic_clock::time_point, uint32_t, const MessageHeader *>
776 timestamp = channels_[channel_index].data.front_timestamp();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800777 FlatbufferVector<MessageHeader> front =
778 std::move(channels_[channel_index].data.front());
Austin Schuh2f8fd752020-09-01 22:38:28 -0700779 channels_[channel_index].data.PopFront();
Austin Schuhcde938c2020-02-02 17:30:07 -0800780
Austin Schuh2f8fd752020-09-01 22:38:28 -0700781 VLOG(1) << MaybeNodeName(target_node_) << "Popped Data " << this << " "
782 << std::get<0>(timestamp) << " for "
783 << configuration::StrippedChannelToString(
784 configuration()->channels()->Get(channel_index))
785 << " (" << channel_index << ")";
Austin Schuhcde938c2020-02-02 17:30:07 -0800786
787 QueueMessages(std::get<0>(timestamp));
Austin Schuh6f3babe2020-01-26 20:34:50 -0800788
789 return std::make_tuple(std::get<0>(timestamp), std::get<1>(timestamp),
790 std::move(front));
791}
792
793std::tuple<monotonic_clock::time_point, uint32_t,
794 FlatbufferVector<MessageHeader>>
Austin Schuh2f8fd752020-09-01 22:38:28 -0700795SplitMessageReader::PopOldestTimestamp(int channel, int node_index) {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800796 CHECK_GT(channels_[channel].timestamps[node_index].size(), 0u);
Austin Schuhcde938c2020-02-02 17:30:07 -0800797 const std::tuple<monotonic_clock::time_point, uint32_t, const MessageHeader *>
798 timestamp = channels_[channel].timestamps[node_index].front_timestamp();
Austin Schuh6f3babe2020-01-26 20:34:50 -0800799 FlatbufferVector<MessageHeader> front =
800 std::move(channels_[channel].timestamps[node_index].front());
Austin Schuh2f8fd752020-09-01 22:38:28 -0700801 channels_[channel].timestamps[node_index].PopFront();
Austin Schuhcde938c2020-02-02 17:30:07 -0800802
Austin Schuh2f8fd752020-09-01 22:38:28 -0700803 VLOG(1) << MaybeNodeName(target_node_) << "Popped timestamp " << this << " "
Austin Schuhee711052020-08-24 16:06:09 -0700804 << std::get<0>(timestamp) << " for "
805 << configuration::StrippedChannelToString(
806 configuration()->channels()->Get(channel))
Austin Schuh2f8fd752020-09-01 22:38:28 -0700807 << " on "
808 << configuration()->nodes()->Get(node_index)->name()->string_view()
809 << " (" << node_index << ")";
Austin Schuhcde938c2020-02-02 17:30:07 -0800810
811 QueueMessages(std::get<0>(timestamp));
Austin Schuh6f3babe2020-01-26 20:34:50 -0800812
813 return std::make_tuple(std::get<0>(timestamp), std::get<1>(timestamp),
814 std::move(front));
815}
816
Austin Schuhcde938c2020-02-02 17:30:07 -0800817bool SplitMessageReader::MessageHeaderQueue::emplace_back(
Austin Schuh6f3babe2020-01-26 20:34:50 -0800818 FlatbufferVector<MessageHeader> &&msg) {
819 CHECK(split_reader != nullptr);
820
821 // If there is no timestamp merger for this queue, nobody is listening. Drop
822 // the message. This happens when a log file from another node is replayed,
823 // and the timestamp mergers down stream just don't care.
824 if (timestamp_merger == nullptr) {
Austin Schuhcde938c2020-02-02 17:30:07 -0800825 return false;
Austin Schuh6f3babe2020-01-26 20:34:50 -0800826 }
827
828 CHECK(timestamps != msg.message().has_data())
829 << ": Got timestamps and data mixed up on a node. "
830 << FlatbufferToJson(msg);
831
832 data_.emplace_back(std::move(msg));
833
834 if (data_.size() == 1u) {
835 // Yup, new data. Notify.
836 if (timestamps) {
837 timestamp_merger->UpdateTimestamp(split_reader, front_timestamp());
838 } else {
839 timestamp_merger->Update(split_reader, front_timestamp());
840 }
841 }
Austin Schuhcde938c2020-02-02 17:30:07 -0800842
843 return true;
Austin Schuh6f3babe2020-01-26 20:34:50 -0800844}
845
Austin Schuh2f8fd752020-09-01 22:38:28 -0700846void SplitMessageReader::MessageHeaderQueue::PopFront() {
Austin Schuh6f3babe2020-01-26 20:34:50 -0800847 data_.pop_front();
848 if (data_.size() != 0u) {
849 // Yup, new data.
850 if (timestamps) {
851 timestamp_merger->UpdateTimestamp(split_reader, front_timestamp());
852 } else {
853 timestamp_merger->Update(split_reader, front_timestamp());
854 }
Austin Schuh2f8fd752020-09-01 22:38:28 -0700855 } else {
856 // Poke anyways to update the heap.
857 if (timestamps) {
858 timestamp_merger->UpdateTimestamp(
859 nullptr, std::make_tuple(monotonic_clock::min_time, 0, nullptr));
860 } else {
861 timestamp_merger->Update(
862 nullptr, std::make_tuple(monotonic_clock::min_time, 0, nullptr));
863 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800864 }
Austin Schuh05b70472020-01-01 17:11:17 -0800865}
866
867namespace {
868
Austin Schuh6f3babe2020-01-26 20:34:50 -0800869bool SplitMessageReaderHeapCompare(
870 const std::tuple<monotonic_clock::time_point, uint32_t,
871 SplitMessageReader *>
872 first,
873 const std::tuple<monotonic_clock::time_point, uint32_t,
874 SplitMessageReader *>
875 second) {
876 if (std::get<0>(first) > std::get<0>(second)) {
877 return true;
878 } else if (std::get<0>(first) == std::get<0>(second)) {
879 if (std::get<1>(first) > std::get<1>(second)) {
880 return true;
881 } else if (std::get<1>(first) == std::get<1>(second)) {
882 return std::get<2>(first) > std::get<2>(second);
883 } else {
884 return false;
885 }
886 } else {
887 return false;
888 }
889}
890
Austin Schuh05b70472020-01-01 17:11:17 -0800891bool ChannelHeapCompare(
892 const std::pair<monotonic_clock::time_point, int> first,
893 const std::pair<monotonic_clock::time_point, int> second) {
894 if (first.first > second.first) {
895 return true;
896 } else if (first.first == second.first) {
897 return first.second > second.second;
898 } else {
899 return false;
900 }
901}
902
903} // namespace
904
Austin Schuh6f3babe2020-01-26 20:34:50 -0800905TimestampMerger::TimestampMerger(
906 const Configuration *configuration,
907 std::vector<SplitMessageReader *> split_message_readers, int channel_index,
908 const Node *target_node, ChannelMerger *channel_merger)
909 : configuration_(configuration),
910 split_message_readers_(std::move(split_message_readers)),
911 channel_index_(channel_index),
912 node_index_(configuration::MultiNode(configuration)
913 ? configuration::GetNodeIndex(configuration, target_node)
914 : -1),
915 channel_merger_(channel_merger) {
916 // Tell the readers we care so they know who to notify.
Austin Schuhcde938c2020-02-02 17:30:07 -0800917 VLOG(1) << "Configuring channel " << channel_index << " target node "
918 << FlatbufferToJson(target_node);
Austin Schuh6f3babe2020-01-26 20:34:50 -0800919 for (SplitMessageReader *reader : split_message_readers_) {
920 reader->SetTimestampMerger(this, channel_index, target_node);
921 }
922
923 // And then determine if we need to track timestamps.
924 const Channel *channel = configuration->channels()->Get(channel_index);
925 if (!configuration::ChannelIsSendableOnNode(channel, target_node) &&
926 configuration::ChannelIsReadableOnNode(channel, target_node)) {
927 has_timestamps_ = true;
928 }
929}
930
931void TimestampMerger::PushMessageHeap(
Austin Schuhcde938c2020-02-02 17:30:07 -0800932 std::tuple<monotonic_clock::time_point, uint32_t, const MessageHeader *>
933 timestamp,
Austin Schuh6f3babe2020-01-26 20:34:50 -0800934 SplitMessageReader *split_message_reader) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700935 if (split_message_reader != nullptr) {
936 DCHECK(std::find_if(message_heap_.begin(), message_heap_.end(),
937 [split_message_reader](
938 const std::tuple<monotonic_clock::time_point,
939 uint32_t, SplitMessageReader *>
940 x) {
941 return std::get<2>(x) == split_message_reader;
942 }) == message_heap_.end())
943 << ": Pushing message when it is already in the heap.";
Austin Schuh6f3babe2020-01-26 20:34:50 -0800944
Austin Schuh2f8fd752020-09-01 22:38:28 -0700945 message_heap_.push_back(std::make_tuple(
946 std::get<0>(timestamp), std::get<1>(timestamp), split_message_reader));
Austin Schuh6f3babe2020-01-26 20:34:50 -0800947
Austin Schuh2f8fd752020-09-01 22:38:28 -0700948 std::push_heap(message_heap_.begin(), message_heap_.end(),
949 &SplitMessageReaderHeapCompare);
950 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800951
952 // If we are just a data merger, don't wait for timestamps.
953 if (!has_timestamps_) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700954 if (!message_heap_.empty()) {
955 channel_merger_->Update(std::get<0>(message_heap_[0]), channel_index_);
956 pushed_ = true;
957 } else {
958 // Remove ourselves if we are empty.
959 channel_merger_->Update(monotonic_clock::min_time, channel_index_);
960 }
Austin Schuh6f3babe2020-01-26 20:34:50 -0800961 }
962}
963
Austin Schuhcde938c2020-02-02 17:30:07 -0800964std::tuple<monotonic_clock::time_point, uint32_t, const MessageHeader *>
965TimestampMerger::oldest_message() const {
966 CHECK_GT(message_heap_.size(), 0u);
967 std::tuple<monotonic_clock::time_point, uint32_t, SplitMessageReader *>
968 oldest_message_reader = message_heap_.front();
969 return std::get<2>(oldest_message_reader)->oldest_message(channel_index_);
970}
971
972std::tuple<monotonic_clock::time_point, uint32_t, const MessageHeader *>
973TimestampMerger::oldest_timestamp() const {
974 CHECK_GT(timestamp_heap_.size(), 0u);
975 std::tuple<monotonic_clock::time_point, uint32_t, SplitMessageReader *>
976 oldest_message_reader = timestamp_heap_.front();
977 return std::get<2>(oldest_message_reader)
978 ->oldest_message(channel_index_, node_index_);
979}
980
Austin Schuh6f3babe2020-01-26 20:34:50 -0800981void TimestampMerger::PushTimestampHeap(
Austin Schuhcde938c2020-02-02 17:30:07 -0800982 std::tuple<monotonic_clock::time_point, uint32_t, const MessageHeader *>
983 timestamp,
Austin Schuh6f3babe2020-01-26 20:34:50 -0800984 SplitMessageReader *split_message_reader) {
Austin Schuh2f8fd752020-09-01 22:38:28 -0700985 if (split_message_reader != nullptr) {
986 DCHECK(std::find_if(timestamp_heap_.begin(), timestamp_heap_.end(),
987 [split_message_reader](
988 const std::tuple<monotonic_clock::time_point,
989 uint32_t, SplitMessageReader *>
990 x) {
991 return std::get<2>(x) == split_message_reader;
992 }) == timestamp_heap_.end())
993 << ": Pushing timestamp when it is already in the heap.";
Austin Schuh6f3babe2020-01-26 20:34:50 -0800994
Austin Schuh2f8fd752020-09-01 22:38:28 -0700995 timestamp_heap_.push_back(std::make_tuple(
996 std::get<0>(timestamp), std::get<1>(timestamp), split_message_reader));
Austin Schuh6f3babe2020-01-26 20:34:50 -0800997
Austin Schuh2f8fd752020-09-01 22:38:28 -0700998 std::push_heap(timestamp_heap_.begin(), timestamp_heap_.end(),
999 SplitMessageReaderHeapCompare);
1000 }
Austin Schuh6f3babe2020-01-26 20:34:50 -08001001
1002 // If we are a timestamp merger, don't wait for data. Missing data will be
1003 // caught at read time.
1004 if (has_timestamps_) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001005 if (!timestamp_heap_.empty()) {
1006 channel_merger_->Update(std::get<0>(timestamp_heap_[0]), channel_index_);
1007 pushed_ = true;
1008 } else {
1009 // Remove ourselves if we are empty.
1010 channel_merger_->Update(monotonic_clock::min_time, channel_index_);
1011 }
Austin Schuh6f3babe2020-01-26 20:34:50 -08001012 }
1013}
1014
1015std::tuple<monotonic_clock::time_point, uint32_t,
1016 FlatbufferVector<MessageHeader>>
1017TimestampMerger::PopMessageHeap() {
1018 // Pop the oldest message reader pointer off the heap.
1019 CHECK_GT(message_heap_.size(), 0u);
1020 std::tuple<monotonic_clock::time_point, uint32_t, SplitMessageReader *>
1021 oldest_message_reader = message_heap_.front();
1022
1023 std::pop_heap(message_heap_.begin(), message_heap_.end(),
1024 &SplitMessageReaderHeapCompare);
1025 message_heap_.pop_back();
1026
1027 // Pop the oldest message. This re-pushes any messages from the reader to the
1028 // message heap.
1029 std::tuple<monotonic_clock::time_point, uint32_t,
1030 FlatbufferVector<MessageHeader>>
1031 oldest_message =
1032 std::get<2>(oldest_message_reader)->PopOldest(channel_index_);
1033
1034 // Confirm that the time and queue_index we have recorded matches.
1035 CHECK_EQ(std::get<0>(oldest_message), std::get<0>(oldest_message_reader));
1036 CHECK_EQ(std::get<1>(oldest_message), std::get<1>(oldest_message_reader));
1037
1038 // Now, keep reading until we have found all duplicates.
Brian Silverman8a32ce62020-08-12 12:02:38 -07001039 while (!message_heap_.empty()) {
Austin Schuh6f3babe2020-01-26 20:34:50 -08001040 // See if it is a duplicate.
1041 std::tuple<monotonic_clock::time_point, uint32_t, SplitMessageReader *>
1042 next_oldest_message_reader = message_heap_.front();
1043
Austin Schuhcde938c2020-02-02 17:30:07 -08001044 std::tuple<monotonic_clock::time_point, uint32_t, const MessageHeader *>
1045 next_oldest_message_time = std::get<2>(next_oldest_message_reader)
1046 ->oldest_message(channel_index_);
Austin Schuh6f3babe2020-01-26 20:34:50 -08001047
1048 if (std::get<0>(next_oldest_message_time) == std::get<0>(oldest_message) &&
1049 std::get<1>(next_oldest_message_time) == std::get<1>(oldest_message)) {
1050 // Pop the message reader pointer.
1051 std::pop_heap(message_heap_.begin(), message_heap_.end(),
1052 &SplitMessageReaderHeapCompare);
1053 message_heap_.pop_back();
1054
1055 // Pop the next oldest message. This re-pushes any messages from the
1056 // reader.
1057 std::tuple<monotonic_clock::time_point, uint32_t,
1058 FlatbufferVector<MessageHeader>>
1059 next_oldest_message = std::get<2>(next_oldest_message_reader)
1060 ->PopOldest(channel_index_);
1061
1062 // And make sure the message matches in it's entirety.
1063 CHECK(std::get<2>(oldest_message).span() ==
1064 std::get<2>(next_oldest_message).span())
1065 << ": Data at the same timestamp doesn't match.";
1066 } else {
1067 break;
1068 }
1069 }
1070
1071 return oldest_message;
1072}
1073
1074std::tuple<monotonic_clock::time_point, uint32_t,
1075 FlatbufferVector<MessageHeader>>
1076TimestampMerger::PopTimestampHeap() {
1077 // Pop the oldest message reader pointer off the heap.
1078 CHECK_GT(timestamp_heap_.size(), 0u);
1079
1080 std::tuple<monotonic_clock::time_point, uint32_t, SplitMessageReader *>
1081 oldest_timestamp_reader = timestamp_heap_.front();
1082
1083 std::pop_heap(timestamp_heap_.begin(), timestamp_heap_.end(),
1084 &SplitMessageReaderHeapCompare);
1085 timestamp_heap_.pop_back();
1086
1087 CHECK(node_index_ != -1) << ": Timestamps in a single node environment";
1088
1089 // Pop the oldest message. This re-pushes any timestamps from the reader to
1090 // the timestamp heap.
1091 std::tuple<monotonic_clock::time_point, uint32_t,
1092 FlatbufferVector<MessageHeader>>
1093 oldest_timestamp = std::get<2>(oldest_timestamp_reader)
Austin Schuh2f8fd752020-09-01 22:38:28 -07001094 ->PopOldestTimestamp(channel_index_, node_index_);
Austin Schuh6f3babe2020-01-26 20:34:50 -08001095
1096 // Confirm that the time we have recorded matches.
1097 CHECK_EQ(std::get<0>(oldest_timestamp), std::get<0>(oldest_timestamp_reader));
1098 CHECK_EQ(std::get<1>(oldest_timestamp), std::get<1>(oldest_timestamp_reader));
1099
Austin Schuh2f8fd752020-09-01 22:38:28 -07001100 // Now, keep reading until we have found all duplicates.
1101 while (!timestamp_heap_.empty()) {
1102 // See if it is a duplicate.
1103 std::tuple<monotonic_clock::time_point, uint32_t, SplitMessageReader *>
1104 next_oldest_timestamp_reader = timestamp_heap_.front();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001105
Austin Schuh2f8fd752020-09-01 22:38:28 -07001106 std::tuple<monotonic_clock::time_point, uint32_t, const MessageHeader *>
1107 next_oldest_timestamp_time =
1108 std::get<2>(next_oldest_timestamp_reader)
1109 ->oldest_message(channel_index_, node_index_);
Austin Schuh6f3babe2020-01-26 20:34:50 -08001110
Austin Schuh2f8fd752020-09-01 22:38:28 -07001111 if (std::get<0>(next_oldest_timestamp_time) ==
1112 std::get<0>(oldest_timestamp) &&
1113 std::get<1>(next_oldest_timestamp_time) ==
1114 std::get<1>(oldest_timestamp)) {
1115 // Pop the timestamp reader pointer.
1116 std::pop_heap(timestamp_heap_.begin(), timestamp_heap_.end(),
1117 &SplitMessageReaderHeapCompare);
1118 timestamp_heap_.pop_back();
1119
1120 // Pop the next oldest timestamp. This re-pushes any messages from the
1121 // reader.
1122 std::tuple<monotonic_clock::time_point, uint32_t,
1123 FlatbufferVector<MessageHeader>>
1124 next_oldest_timestamp =
1125 std::get<2>(next_oldest_timestamp_reader)
1126 ->PopOldestTimestamp(channel_index_, node_index_);
1127
1128 // And make sure the contents matches in it's entirety.
1129 CHECK(std::get<2>(oldest_timestamp).span() ==
1130 std::get<2>(next_oldest_timestamp).span())
1131 << ": Data at the same timestamp doesn't match, "
1132 << aos::FlatbufferToJson(std::get<2>(oldest_timestamp)) << " vs "
1133 << aos::FlatbufferToJson(std::get<2>(next_oldest_timestamp)) << " "
1134 << absl::BytesToHexString(std::string_view(
1135 reinterpret_cast<const char *>(
1136 std::get<2>(oldest_timestamp).span().data()),
1137 std::get<2>(oldest_timestamp).span().size()))
1138 << " vs "
1139 << absl::BytesToHexString(std::string_view(
1140 reinterpret_cast<const char *>(
1141 std::get<2>(next_oldest_timestamp).span().data()),
1142 std::get<2>(next_oldest_timestamp).span().size()));
1143
1144 } else {
1145 break;
1146 }
Austin Schuh8bd96322020-02-13 21:18:22 -08001147 }
1148
Austin Schuh2f8fd752020-09-01 22:38:28 -07001149 return oldest_timestamp;
Austin Schuh8bd96322020-02-13 21:18:22 -08001150}
1151
Austin Schuh6f3babe2020-01-26 20:34:50 -08001152std::tuple<TimestampMerger::DeliveryTimestamp, FlatbufferVector<MessageHeader>>
1153TimestampMerger::PopOldest() {
1154 if (has_timestamps_) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001155 VLOG(1) << "Looking for matching timestamp for "
1156 << configuration::StrippedChannelToString(
1157 configuration_->channels()->Get(channel_index_))
1158 << " (" << channel_index_ << ") "
1159 << " at " << std::get<0>(oldest_timestamp());
1160
Austin Schuh8bd96322020-02-13 21:18:22 -08001161 // Read the timestamps.
Austin Schuh6f3babe2020-01-26 20:34:50 -08001162 std::tuple<monotonic_clock::time_point, uint32_t,
1163 FlatbufferVector<MessageHeader>>
1164 oldest_timestamp = PopTimestampHeap();
1165
1166 TimestampMerger::DeliveryTimestamp timestamp;
1167 timestamp.monotonic_event_time =
1168 monotonic_clock::time_point(chrono::nanoseconds(
1169 std::get<2>(oldest_timestamp).message().monotonic_sent_time()));
1170 timestamp.realtime_event_time =
1171 realtime_clock::time_point(chrono::nanoseconds(
1172 std::get<2>(oldest_timestamp).message().realtime_sent_time()));
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001173 timestamp.queue_index =
1174 std::get<2>(oldest_timestamp).message().queue_index();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001175
1176 // Consistency check.
1177 CHECK_EQ(timestamp.monotonic_event_time, std::get<0>(oldest_timestamp));
1178 CHECK_EQ(std::get<2>(oldest_timestamp).message().queue_index(),
1179 std::get<1>(oldest_timestamp));
1180
1181 monotonic_clock::time_point remote_timestamp_monotonic_time(
1182 chrono::nanoseconds(
1183 std::get<2>(oldest_timestamp).message().monotonic_remote_time()));
1184
Austin Schuh8bd96322020-02-13 21:18:22 -08001185 // See if we have any data. If not, pass the problem up the chain.
Brian Silverman8a32ce62020-08-12 12:02:38 -07001186 if (message_heap_.empty()) {
Austin Schuhee711052020-08-24 16:06:09 -07001187 LOG(WARNING) << MaybeNodeName(configuration_->nodes()->Get(node_index_))
1188 << "No data to match timestamp on "
1189 << configuration::CleanedChannelToString(
1190 configuration_->channels()->Get(channel_index_))
1191 << " (" << channel_index_ << ")";
Austin Schuh8bd96322020-02-13 21:18:22 -08001192 return std::make_tuple(timestamp,
1193 std::move(std::get<2>(oldest_timestamp)));
1194 }
1195
Austin Schuh6f3babe2020-01-26 20:34:50 -08001196 while (true) {
Austin Schuhcde938c2020-02-02 17:30:07 -08001197 {
1198 // Ok, now try grabbing data until we find one which matches.
1199 std::tuple<monotonic_clock::time_point, uint32_t, const MessageHeader *>
1200 oldest_message_ref = oldest_message();
1201
1202 // Time at which the message was sent (this message is written from the
1203 // sending node's perspective.
1204 monotonic_clock::time_point remote_monotonic_time(chrono::nanoseconds(
1205 std::get<2>(oldest_message_ref)->monotonic_sent_time()));
1206
1207 if (remote_monotonic_time < remote_timestamp_monotonic_time) {
Austin Schuhee711052020-08-24 16:06:09 -07001208 LOG(WARNING) << configuration_->nodes()
1209 ->Get(node_index_)
1210 ->name()
1211 ->string_view()
1212 << " Undelivered message, skipping. Remote time is "
1213 << remote_monotonic_time << " timestamp is "
1214 << remote_timestamp_monotonic_time << " on channel "
1215 << configuration::StrippedChannelToString(
1216 configuration_->channels()->Get(channel_index_))
1217 << " (" << channel_index_ << ")";
Austin Schuhcde938c2020-02-02 17:30:07 -08001218 PopMessageHeap();
1219 continue;
1220 } else if (remote_monotonic_time > remote_timestamp_monotonic_time) {
Austin Schuhee711052020-08-24 16:06:09 -07001221 LOG(WARNING) << configuration_->nodes()
1222 ->Get(node_index_)
1223 ->name()
1224 ->string_view()
1225 << " Data not found. Remote time should be "
1226 << remote_timestamp_monotonic_time
1227 << ", message time is " << remote_monotonic_time
1228 << " on channel "
1229 << configuration::StrippedChannelToString(
1230 configuration_->channels()->Get(channel_index_))
Austin Schuh2f8fd752020-09-01 22:38:28 -07001231 << " (" << channel_index_ << ")"
1232 << (VLOG_IS_ON(1) ? DebugString() : "");
Austin Schuhcde938c2020-02-02 17:30:07 -08001233 return std::make_tuple(timestamp,
1234 std::move(std::get<2>(oldest_timestamp)));
1235 }
1236
1237 timestamp.monotonic_remote_time = remote_monotonic_time;
1238 }
1239
Austin Schuh2f8fd752020-09-01 22:38:28 -07001240 VLOG(1) << "Found matching data "
1241 << configuration::StrippedChannelToString(
1242 configuration_->channels()->Get(channel_index_))
1243 << " (" << channel_index_ << ")";
Austin Schuh6f3babe2020-01-26 20:34:50 -08001244 std::tuple<monotonic_clock::time_point, uint32_t,
1245 FlatbufferVector<MessageHeader>>
1246 oldest_message = PopMessageHeap();
1247
Austin Schuh6f3babe2020-01-26 20:34:50 -08001248 timestamp.realtime_remote_time =
1249 realtime_clock::time_point(chrono::nanoseconds(
1250 std::get<2>(oldest_message).message().realtime_sent_time()));
1251 timestamp.remote_queue_index =
1252 std::get<2>(oldest_message).message().queue_index();
1253
Austin Schuhcde938c2020-02-02 17:30:07 -08001254 CHECK_EQ(timestamp.monotonic_remote_time,
1255 remote_timestamp_monotonic_time);
1256
1257 CHECK_EQ(timestamp.remote_queue_index,
1258 std::get<2>(oldest_timestamp).message().remote_queue_index())
1259 << ": " << FlatbufferToJson(&std::get<2>(oldest_timestamp).message())
1260 << " data "
1261 << FlatbufferToJson(&std::get<2>(oldest_message).message());
Austin Schuh6f3babe2020-01-26 20:34:50 -08001262
Austin Schuh30dd5c52020-08-01 14:43:44 -07001263 return std::make_tuple(timestamp, std::move(std::get<2>(oldest_message)));
Austin Schuh6f3babe2020-01-26 20:34:50 -08001264 }
1265 } else {
1266 std::tuple<monotonic_clock::time_point, uint32_t,
1267 FlatbufferVector<MessageHeader>>
1268 oldest_message = PopMessageHeap();
1269
1270 TimestampMerger::DeliveryTimestamp timestamp;
1271 timestamp.monotonic_event_time =
1272 monotonic_clock::time_point(chrono::nanoseconds(
1273 std::get<2>(oldest_message).message().monotonic_sent_time()));
1274 timestamp.realtime_event_time =
1275 realtime_clock::time_point(chrono::nanoseconds(
1276 std::get<2>(oldest_message).message().realtime_sent_time()));
Austin Schuh8d7e0bb2020-10-02 17:57:00 -07001277 timestamp.queue_index = std::get<2>(oldest_message).message().queue_index();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001278 timestamp.remote_queue_index = 0xffffffff;
1279
1280 CHECK_EQ(std::get<0>(oldest_message), timestamp.monotonic_event_time);
1281 CHECK_EQ(std::get<1>(oldest_message),
1282 std::get<2>(oldest_message).message().queue_index());
1283
Austin Schuh30dd5c52020-08-01 14:43:44 -07001284 return std::make_tuple(timestamp, std::move(std::get<2>(oldest_message)));
Austin Schuh6f3babe2020-01-26 20:34:50 -08001285 }
1286}
1287
Austin Schuh8bd96322020-02-13 21:18:22 -08001288void TimestampMerger::NoticeAtEnd() { channel_merger_->NoticeAtEnd(); }
1289
Austin Schuh6f3babe2020-01-26 20:34:50 -08001290namespace {
1291std::vector<std::unique_ptr<SplitMessageReader>> MakeSplitMessageReaders(
1292 const std::vector<std::vector<std::string>> &filenames) {
1293 CHECK_GT(filenames.size(), 0u);
1294 // Build up all the SplitMessageReaders.
1295 std::vector<std::unique_ptr<SplitMessageReader>> result;
1296 for (const std::vector<std::string> &filenames : filenames) {
1297 result.emplace_back(std::make_unique<SplitMessageReader>(filenames));
1298 }
1299 return result;
1300}
1301} // namespace
1302
1303ChannelMerger::ChannelMerger(
1304 const std::vector<std::vector<std::string>> &filenames)
1305 : split_message_readers_(MakeSplitMessageReaders(filenames)),
Austin Schuh97789fc2020-08-01 14:42:45 -07001306 log_file_header_(split_message_readers_[0]->raw_log_file_header()) {
Austin Schuh6f3babe2020-01-26 20:34:50 -08001307 // Now, confirm that the configuration matches for each and pick a start time.
1308 // Also return the list of possible nodes.
1309 for (const std::unique_ptr<SplitMessageReader> &reader :
1310 split_message_readers_) {
1311 CHECK(CompareFlatBuffer(log_file_header_.message().configuration(),
1312 reader->log_file_header()->configuration()))
1313 << ": Replaying log files with different configurations isn't "
1314 "supported";
1315 }
1316
1317 nodes_ = configuration::GetNodes(configuration());
1318}
1319
1320bool ChannelMerger::SetNode(const Node *target_node) {
1321 std::vector<SplitMessageReader *> split_message_readers;
1322 for (const std::unique_ptr<SplitMessageReader> &reader :
1323 split_message_readers_) {
1324 split_message_readers.emplace_back(reader.get());
1325 }
1326
1327 // Go find a log_file_header for this node.
1328 {
1329 bool found_node = false;
1330
1331 for (const std::unique_ptr<SplitMessageReader> &reader :
1332 split_message_readers_) {
James Kuszmaulfc273dc2020-05-09 17:56:19 -07001333 // In order to identify which logfile(s) map to the target node, do a
1334 // logical comparison of the nodes, by confirming that we are either in a
1335 // single-node setup (where the nodes will both be nullptr) or that the
1336 // node names match (but the other node fields--e.g., hostname lists--may
1337 // not).
1338 const bool both_null =
1339 reader->node() == nullptr && target_node == nullptr;
1340 const bool both_have_name =
1341 (reader->node() != nullptr) && (target_node != nullptr) &&
1342 (reader->node()->has_name() && target_node->has_name());
1343 const bool node_names_identical =
Brian Silvermand90905f2020-09-23 14:42:56 -07001344 both_have_name && (reader->node()->name()->string_view() ==
1345 target_node->name()->string_view());
James Kuszmaulfc273dc2020-05-09 17:56:19 -07001346 if (both_null || node_names_identical) {
Austin Schuh6f3babe2020-01-26 20:34:50 -08001347 if (!found_node) {
1348 found_node = true;
1349 log_file_header_ = CopyFlatBuffer(reader->log_file_header());
Austin Schuhcde938c2020-02-02 17:30:07 -08001350 VLOG(1) << "Found log file " << reader->filename() << " with node "
1351 << FlatbufferToJson(reader->node()) << " start_time "
1352 << monotonic_start_time();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001353 } else {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001354 // Find the earliest start time. That way, if we get a full log file
1355 // directly from the node, and a partial later, we start with the
1356 // full. Update our header to match that.
1357 const monotonic_clock::time_point new_monotonic_start_time(
1358 chrono::nanoseconds(
1359 reader->log_file_header()->monotonic_start_time()));
1360 const realtime_clock::time_point new_realtime_start_time(
1361 chrono::nanoseconds(
1362 reader->log_file_header()->realtime_start_time()));
1363
1364 if (monotonic_start_time() == monotonic_clock::min_time ||
1365 (new_monotonic_start_time != monotonic_clock::min_time &&
1366 new_monotonic_start_time < monotonic_start_time())) {
1367 log_file_header_.mutable_message()->mutate_monotonic_start_time(
1368 new_monotonic_start_time.time_since_epoch().count());
1369 log_file_header_.mutable_message()->mutate_realtime_start_time(
1370 new_realtime_start_time.time_since_epoch().count());
1371 VLOG(1) << "Updated log file " << reader->filename()
1372 << " with node " << FlatbufferToJson(reader->node())
1373 << " start_time " << new_monotonic_start_time;
1374 }
Austin Schuh6f3babe2020-01-26 20:34:50 -08001375 }
1376 }
1377 }
1378
1379 if (!found_node) {
1380 LOG(WARNING) << "Failed to find log file for node "
1381 << FlatbufferToJson(target_node);
1382 return false;
1383 }
1384 }
1385
1386 // Build up all the timestamp mergers. This connects up all the
1387 // SplitMessageReaders.
1388 timestamp_mergers_.reserve(configuration()->channels()->size());
1389 for (size_t channel_index = 0;
1390 channel_index < configuration()->channels()->size(); ++channel_index) {
1391 timestamp_mergers_.emplace_back(
1392 configuration(), split_message_readers, channel_index,
1393 configuration::GetNode(configuration(), target_node), this);
1394 }
1395
1396 // And prime everything.
Austin Schuh6f3babe2020-01-26 20:34:50 -08001397 for (std::unique_ptr<SplitMessageReader> &split_message_reader :
1398 split_message_readers_) {
Austin Schuhcde938c2020-02-02 17:30:07 -08001399 split_message_reader->QueueMessages(
1400 split_message_reader->monotonic_start_time());
Austin Schuh6f3babe2020-01-26 20:34:50 -08001401 }
1402
1403 node_ = configuration::GetNodeOrDie(configuration(), target_node);
1404 return true;
1405}
1406
Austin Schuh858c9f32020-08-31 16:56:12 -07001407monotonic_clock::time_point ChannelMerger::OldestMessageTime() const {
Brian Silverman8a32ce62020-08-12 12:02:38 -07001408 if (channel_heap_.empty()) {
Austin Schuh6f3babe2020-01-26 20:34:50 -08001409 return monotonic_clock::max_time;
1410 }
1411 return channel_heap_.front().first;
1412}
1413
1414void ChannelMerger::PushChannelHeap(monotonic_clock::time_point timestamp,
1415 int channel_index) {
1416 // Pop and recreate the heap if it has already been pushed. And since we are
1417 // pushing again, we don't need to clear pushed.
1418 if (timestamp_mergers_[channel_index].pushed()) {
Brian Silverman8a32ce62020-08-12 12:02:38 -07001419 const auto channel_iterator = std::find_if(
Austin Schuh6f3babe2020-01-26 20:34:50 -08001420 channel_heap_.begin(), channel_heap_.end(),
1421 [channel_index](const std::pair<monotonic_clock::time_point, int> x) {
1422 return x.second == channel_index;
Brian Silverman8a32ce62020-08-12 12:02:38 -07001423 });
1424 DCHECK(channel_iterator != channel_heap_.end());
1425 if (std::get<0>(*channel_iterator) == timestamp) {
1426 // It's already in the heap, in the correct spot, so nothing
1427 // more for us to do here.
1428 return;
1429 }
1430 channel_heap_.erase(channel_iterator);
Austin Schuh6f3babe2020-01-26 20:34:50 -08001431 std::make_heap(channel_heap_.begin(), channel_heap_.end(),
1432 ChannelHeapCompare);
1433 }
1434
Austin Schuh2f8fd752020-09-01 22:38:28 -07001435 if (timestamp == monotonic_clock::min_time) {
1436 timestamp_mergers_[channel_index].set_pushed(false);
1437 return;
1438 }
1439
Austin Schuh05b70472020-01-01 17:11:17 -08001440 channel_heap_.push_back(std::make_pair(timestamp, channel_index));
1441
1442 // The default sort puts the newest message first. Use a custom comparator to
1443 // put the oldest message first.
1444 std::push_heap(channel_heap_.begin(), channel_heap_.end(),
1445 ChannelHeapCompare);
1446}
1447
Austin Schuh2f8fd752020-09-01 22:38:28 -07001448void ChannelMerger::VerifyHeaps() {
Austin Schuh661a8d82020-09-13 17:25:56 -07001449 std::vector<std::pair<monotonic_clock::time_point, int>> channel_heap =
1450 channel_heap_;
1451 std::make_heap(channel_heap.begin(), channel_heap.end(), &ChannelHeapCompare);
Austin Schuh2f8fd752020-09-01 22:38:28 -07001452
Austin Schuh661a8d82020-09-13 17:25:56 -07001453 for (size_t i = 0; i < channel_heap_.size(); ++i) {
1454 CHECK(channel_heap_[i] == channel_heap[i]) << ": Heaps diverged...";
1455 CHECK_EQ(
1456 std::get<0>(channel_heap[i]),
1457 timestamp_mergers_[std::get<1>(channel_heap[i])].channel_merger_time());
Austin Schuh2f8fd752020-09-01 22:38:28 -07001458 }
1459}
1460
Austin Schuh6f3babe2020-01-26 20:34:50 -08001461std::tuple<TimestampMerger::DeliveryTimestamp, int,
1462 FlatbufferVector<MessageHeader>>
1463ChannelMerger::PopOldest() {
Austin Schuh8bd96322020-02-13 21:18:22 -08001464 CHECK_GT(channel_heap_.size(), 0u);
Austin Schuh05b70472020-01-01 17:11:17 -08001465 std::pair<monotonic_clock::time_point, int> oldest_channel_data =
1466 channel_heap_.front();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001467 int channel_index = oldest_channel_data.second;
Austin Schuh05b70472020-01-01 17:11:17 -08001468 std::pop_heap(channel_heap_.begin(), channel_heap_.end(),
1469 &ChannelHeapCompare);
1470 channel_heap_.pop_back();
Austin Schuh8bd96322020-02-13 21:18:22 -08001471
Austin Schuh6f3babe2020-01-26 20:34:50 -08001472 timestamp_mergers_[channel_index].set_pushed(false);
Austin Schuh05b70472020-01-01 17:11:17 -08001473
Austin Schuh6f3babe2020-01-26 20:34:50 -08001474 TimestampMerger *merger = &timestamp_mergers_[channel_index];
Austin Schuh05b70472020-01-01 17:11:17 -08001475
Austin Schuhcde938c2020-02-02 17:30:07 -08001476 // Merger handles any queueing needed from here.
Austin Schuh6f3babe2020-01-26 20:34:50 -08001477 std::tuple<TimestampMerger::DeliveryTimestamp,
1478 FlatbufferVector<MessageHeader>>
1479 message = merger->PopOldest();
Brian Silverman8a32ce62020-08-12 12:02:38 -07001480 DCHECK_EQ(std::get<0>(message).monotonic_event_time,
1481 oldest_channel_data.first)
1482 << ": channel_heap_ was corrupted for " << channel_index << ": "
1483 << DebugString();
Austin Schuh05b70472020-01-01 17:11:17 -08001484
Austin Schuh2f8fd752020-09-01 22:38:28 -07001485 CHECK_GE(std::get<0>(message).monotonic_event_time, last_popped_time_)
1486 << ": " << MaybeNodeName(log_file_header()->node())
1487 << "Messages came off the queue out of order. " << DebugString();
1488 last_popped_time_ = std::get<0>(message).monotonic_event_time;
1489
1490 VLOG(1) << "Popped " << last_popped_time_ << " "
1491 << configuration::StrippedChannelToString(
1492 configuration()->channels()->Get(channel_index))
1493 << " (" << channel_index << ")";
1494
Austin Schuh6f3babe2020-01-26 20:34:50 -08001495 return std::make_tuple(std::get<0>(message), channel_index,
1496 std::move(std::get<1>(message)));
1497}
1498
Austin Schuhcde938c2020-02-02 17:30:07 -08001499std::string SplitMessageReader::MessageHeaderQueue::DebugString() const {
1500 std::stringstream ss;
1501 for (size_t i = 0; i < data_.size(); ++i) {
Austin Schuh2f8fd752020-09-01 22:38:28 -07001502 if (i < 5 || i + 5 > data_.size()) {
1503 if (timestamps) {
1504 ss << " msg: ";
1505 } else {
1506 ss << " timestamp: ";
1507 }
1508 ss << monotonic_clock::time_point(
1509 chrono::nanoseconds(data_[i].message().monotonic_sent_time()))
Austin Schuhcde938c2020-02-02 17:30:07 -08001510 << " ("
Austin Schuh2f8fd752020-09-01 22:38:28 -07001511 << realtime_clock::time_point(
1512 chrono::nanoseconds(data_[i].message().realtime_sent_time()))
1513 << ") " << data_[i].message().queue_index();
1514 if (timestamps) {
1515 ss << " <- remote "
1516 << monotonic_clock::time_point(chrono::nanoseconds(
1517 data_[i].message().monotonic_remote_time()))
1518 << " ("
1519 << realtime_clock::time_point(chrono::nanoseconds(
1520 data_[i].message().realtime_remote_time()))
1521 << ")";
1522 }
1523 ss << "\n";
1524 } else if (i == 5) {
1525 ss << " ...\n";
Austin Schuh6f3babe2020-01-26 20:34:50 -08001526 }
Austin Schuhcde938c2020-02-02 17:30:07 -08001527 }
Austin Schuh6f3babe2020-01-26 20:34:50 -08001528
Austin Schuhcde938c2020-02-02 17:30:07 -08001529 return ss.str();
1530}
Austin Schuh6f3babe2020-01-26 20:34:50 -08001531
Austin Schuhcde938c2020-02-02 17:30:07 -08001532std::string SplitMessageReader::DebugString(int channel) const {
1533 std::stringstream ss;
1534 ss << "[\n";
1535 ss << channels_[channel].data.DebugString();
1536 ss << " ]";
1537 return ss.str();
1538}
Austin Schuh6f3babe2020-01-26 20:34:50 -08001539
Austin Schuhcde938c2020-02-02 17:30:07 -08001540std::string SplitMessageReader::DebugString(int channel, int node_index) const {
1541 std::stringstream ss;
1542 ss << "[\n";
1543 ss << channels_[channel].timestamps[node_index].DebugString();
1544 ss << " ]";
1545 return ss.str();
1546}
1547
1548std::string TimestampMerger::DebugString() const {
1549 std::stringstream ss;
1550
1551 if (timestamp_heap_.size() > 0) {
1552 ss << " timestamp_heap {\n";
1553 std::vector<
1554 std::tuple<monotonic_clock::time_point, uint32_t, SplitMessageReader *>>
1555 timestamp_heap = timestamp_heap_;
1556 while (timestamp_heap.size() > 0u) {
1557 std::tuple<monotonic_clock::time_point, uint32_t, SplitMessageReader *>
1558 oldest_timestamp_reader = timestamp_heap.front();
1559
1560 ss << " " << std::get<2>(oldest_timestamp_reader) << " "
1561 << std::get<0>(oldest_timestamp_reader) << " queue_index ("
1562 << std::get<1>(oldest_timestamp_reader) << ") ttq "
1563 << std::get<2>(oldest_timestamp_reader)->time_to_queue() << " "
1564 << std::get<2>(oldest_timestamp_reader)->filename() << " -> "
1565 << std::get<2>(oldest_timestamp_reader)
1566 ->DebugString(channel_index_, node_index_)
1567 << "\n";
1568
1569 std::pop_heap(timestamp_heap.begin(), timestamp_heap.end(),
1570 &SplitMessageReaderHeapCompare);
1571 timestamp_heap.pop_back();
1572 }
1573 ss << " }\n";
1574 }
1575
1576 ss << " message_heap {\n";
1577 {
1578 std::vector<
1579 std::tuple<monotonic_clock::time_point, uint32_t, SplitMessageReader *>>
1580 message_heap = message_heap_;
Brian Silverman8a32ce62020-08-12 12:02:38 -07001581 while (!message_heap.empty()) {
Austin Schuhcde938c2020-02-02 17:30:07 -08001582 std::tuple<monotonic_clock::time_point, uint32_t, SplitMessageReader *>
1583 oldest_message_reader = message_heap.front();
1584
1585 ss << " " << std::get<2>(oldest_message_reader) << " "
1586 << std::get<0>(oldest_message_reader) << " queue_index ("
1587 << std::get<1>(oldest_message_reader) << ") ttq "
1588 << std::get<2>(oldest_message_reader)->time_to_queue() << " "
1589 << std::get<2>(oldest_message_reader)->filename() << " -> "
1590 << std::get<2>(oldest_message_reader)->DebugString(channel_index_)
1591 << "\n";
1592
1593 std::pop_heap(message_heap.begin(), message_heap.end(),
1594 &SplitMessageReaderHeapCompare);
1595 message_heap.pop_back();
Austin Schuh6f3babe2020-01-26 20:34:50 -08001596 }
Austin Schuh05b70472020-01-01 17:11:17 -08001597 }
Austin Schuhcde938c2020-02-02 17:30:07 -08001598 ss << " }";
1599
1600 return ss.str();
1601}
1602
1603std::string ChannelMerger::DebugString() const {
1604 std::stringstream ss;
1605 ss << "start_time " << realtime_start_time() << " " << monotonic_start_time()
1606 << "\n";
1607 ss << "channel_heap {\n";
1608 std::vector<std::pair<monotonic_clock::time_point, int>> channel_heap =
1609 channel_heap_;
Brian Silverman8a32ce62020-08-12 12:02:38 -07001610 while (!channel_heap.empty()) {
Austin Schuhcde938c2020-02-02 17:30:07 -08001611 std::tuple<monotonic_clock::time_point, int> channel = channel_heap.front();
1612 ss << " " << std::get<0>(channel) << " (" << std::get<1>(channel) << ") "
1613 << configuration::CleanedChannelToString(
1614 configuration()->channels()->Get(std::get<1>(channel)))
1615 << "\n";
1616
1617 ss << timestamp_mergers_[std::get<1>(channel)].DebugString() << "\n";
1618
1619 std::pop_heap(channel_heap.begin(), channel_heap.end(),
1620 &ChannelHeapCompare);
1621 channel_heap.pop_back();
1622 }
1623 ss << "}";
1624
1625 return ss.str();
Austin Schuh05b70472020-01-01 17:11:17 -08001626}
1627
Austin Schuhee711052020-08-24 16:06:09 -07001628std::string MaybeNodeName(const Node *node) {
1629 if (node != nullptr) {
1630 return node->name()->str() + " ";
1631 }
1632 return "";
1633}
1634
Brian Silvermanf51499a2020-09-21 12:49:08 -07001635} // namespace aos::logger