1 /* Copyright (c) 2010-2021. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
6 #include "include/xbt/config.hpp"
7 #include "simgrid/Exception.hpp"
8 #include "simgrid/s4u/Engine.hpp"
9 #include "simgrid/version.h"
10 #include "src/instr/instr_private.hpp"
11 #include "surf/surf.hpp"
15 #include <direct.h> // _mkdir
22 XBT_LOG_NEW_CATEGORY(instr, "Logging the behavior of the tracing system (used for Visualization/Analysis of simulations)");
23 XBT_LOG_NEW_DEFAULT_SUBCATEGORY (instr_config, instr, "Configuration");
25 std::ofstream tracing_file;
26 std::map<const simgrid::instr::Container*, std::ofstream*> tracing_files; // TI specific
27 double prefix = 0.0; // TI specific
29 constexpr char OPT_TRACING_BASIC[] = "tracing/basic";
30 constexpr char OPT_TRACING_COMMENT_FILE[] = "tracing/comment-file";
31 constexpr char OPT_TRACING_DISABLE_DESTROY[] = "tracing/disable-destroy";
32 constexpr char OPT_TRACING_FORMAT_TI_ONEFILE[] = "tracing/smpi/format/ti-one-file";
33 constexpr char OPT_TRACING_SMPI[] = "tracing/smpi";
34 constexpr char OPT_TRACING_TOPOLOGY[] = "tracing/platform/topology";
36 static simgrid::config::Flag<bool> trace_enabled{
37 "tracing", "Enable the tracing system. You have to enable this option to use other tracing options.", false};
39 static simgrid::config::Flag<bool> trace_actor_enabled{
41 {"tracing/msg/process"}, // XBT_ATTRIB_DEPRECATED_v330(option alias)
42 "Trace the behavior of all categorized actors, grouping them by host. "
43 "Can be used to track actor location if the simulator does actor migration.",
46 static simgrid::config::Flag<bool> trace_vm_enabled{"tracing/vm", "Trace the behavior of all virtual machines.", false};
48 static simgrid::config::Flag<bool> trace_platform{"tracing/platform",
49 "Register the platform in the trace as a hierarchy.", false};
51 static simgrid::config::Flag<bool> trace_platform_topology{
52 OPT_TRACING_TOPOLOGY, "Register the platform topology in the trace as a graph.", true};
53 static simgrid::config::Flag<bool> trace_smpi_enabled{OPT_TRACING_SMPI, "Tracing of the SMPI interface.", false};
54 static simgrid::config::Flag<bool> trace_smpi_grouped{"tracing/smpi/group", "Group MPI processes by host.", false};
56 static simgrid::config::Flag<bool> trace_smpi_computing{
57 "tracing/smpi/computing", "Generate 'Computing' states to trace the out-of-SMPI parts of the application", false};
59 static simgrid::config::Flag<bool> trace_smpi_sleeping{
60 "tracing/smpi/sleeping", "Generate 'Sleeping' states for the sleeps in the application that do not pertain to SMPI",
63 static simgrid::config::Flag<bool> trace_view_internals{
64 "tracing/smpi/internals",
65 "Generate tracing events corresponding to point-to-point messages sent by SMPI collective communications", false};
67 static simgrid::config::Flag<bool> trace_categorized{
68 "tracing/categorized", "Trace categorized resource utilization of hosts and links.", false};
70 static simgrid::config::Flag<bool> trace_uncategorized{
71 "tracing/uncategorized",
72 "Trace uncategorized resource utilization of hosts and links. "
73 "To use if the simulator does not use tracing categories but resource utilization have to be traced.",
76 static simgrid::config::Flag<bool> trace_disable_destroy{OPT_TRACING_DISABLE_DESTROY,
77 "Disable platform containers destruction.", false};
78 static simgrid::config::Flag<bool> trace_basic{OPT_TRACING_BASIC, "Avoid extended events (impoverished trace file).",
81 static simgrid::config::Flag<bool> trace_display_sizes{
82 "tracing/smpi/display-sizes",
83 "Add message size information (in bytes) to the to links and states (SMPI only). "
84 "For collectives, it usually corresponds to the total number of bytes sent by a process.",
87 static simgrid::config::Flag<bool> trace_disable_link{"tracing/disable_link",
88 "Do not trace link bandwidth and latency.", false};
89 static simgrid::config::Flag<bool> trace_disable_power{"tracing/disable_power", "Do not trace host power.", false};
91 bool TRACE_needs_platform ()
93 return TRACE_actor_is_enabled() || TRACE_vm_is_enabled() || TRACE_categorized() || TRACE_uncategorized() ||
94 TRACE_platform() || (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped());
97 bool TRACE_is_enabled()
102 bool TRACE_platform()
104 return trace_platform;
107 bool TRACE_platform_topology()
109 return trace_platform_topology;
112 bool TRACE_smpi_is_enabled()
114 return (trace_smpi_enabled || TRACE_smpi_is_grouped()) && TRACE_is_enabled();
117 bool TRACE_smpi_is_grouped()
119 return trace_smpi_grouped;
122 bool TRACE_smpi_is_computing()
124 return trace_smpi_computing;
127 bool TRACE_smpi_is_sleeping()
129 return trace_smpi_sleeping;
132 bool TRACE_smpi_view_internals()
134 return trace_view_internals;
137 bool TRACE_categorized ()
139 return trace_categorized;
142 bool TRACE_uncategorized ()
144 return trace_uncategorized;
147 bool TRACE_actor_is_enabled()
149 return trace_actor_enabled && trace_enabled;
152 bool TRACE_vm_is_enabled()
154 return trace_vm_enabled && trace_enabled;
157 bool TRACE_disable_link()
159 return trace_disable_link && trace_enabled;
162 bool TRACE_disable_speed()
164 return trace_disable_power && trace_enabled;
167 bool TRACE_display_sizes ()
169 return trace_display_sizes && trace_smpi_enabled && trace_enabled;
172 static void print_line(const char* option, const char* desc, const char* longdesc)
174 std::string str = std::string("--cfg=") + option + " ";
176 int len = static_cast<int>(str.size());
177 XBT_HELP("%s%*.*s %s", str.c_str(), 30 - len, 30 - len, "", desc);
178 if (longdesc != nullptr) {
179 XBT_HELP("%s\n", longdesc);
185 XBT_HELP("Description of the tracing options accepted by this simulator:\n");
186 print_line(OPT_TRACING_SMPI, "Trace the MPI Interface (SMPI)",
187 " This option only has effect if this simulator is SMPI-based. Traces the MPI\n"
188 " interface and generates a trace that can be analyzed using Gantt-like\n"
189 " visualizations. Every MPI function (implemented by SMPI) is transformed in a\n"
190 " state, and point-to-point communications can be analyzed with arrows.");
191 print_line(OPT_TRACING_DISABLE_DESTROY, "Disable platform containers destruction",
192 " Disable the destruction of containers at the end of simulation. This can be\n"
193 " used with simulators that have a different notion of time (different from\n"
194 " the simulated time).");
195 print_line(OPT_TRACING_BASIC, "Avoid extended events (impoverished trace file).",
196 " Some visualization tools are not able to parse correctly the Paje file format.\n"
197 " Use this option if you are using one of these tools to visualize the simulation\n"
198 " trace. Keep in mind that the trace might be incomplete, without all the\n"
199 " information that would be registered otherwise.");
200 print_line(OPT_TRACING_FORMAT_TI_ONEFILE, "Only works for SMPI now, and TI output format",
201 " By default, each process outputs to a separate file, inside a filename_files folder\n"
202 " By setting this option to yes, all processes will output to only one file\n"
203 " This is meant to avoid opening thousands of files with large simulations");
204 print_line(OPT_TRACING_TOPOLOGY, "Register the platform topology as a graph",
205 " This option (enabled by default) can be used to disable the tracing of\n"
206 " the platform topology in the trace file. Sometimes, such task is really\n"
207 " time consuming, since it must get the route from each host to other hosts\n"
208 " within the same Autonomous System (AS).");
213 static bool trace_active = false;
214 TraceFormat trace_format = TraceFormat::Paje;
220 xbt::signal<void(Container const&)> Container::on_creation;
221 xbt::signal<void(Container const&)> Container::on_destruction;
222 xbt::signal<void(Type const&, PajeEventType)> Type::on_creation;
223 xbt::signal<void(LinkType const&, Type const&, Type const&)> LinkType::on_creation;
224 xbt::signal<void(PajeEvent&)> PajeEvent::on_creation;
225 xbt::signal<void(PajeEvent const&)> PajeEvent::on_destruction;
226 xbt::signal<void(StateEvent const&)> StateEvent::on_destruction;
227 xbt::signal<void(EntityValue const&)> EntityValue::on_creation;
229 static void on_container_creation_paje(const Container& c)
231 double timestamp = SIMIX_get_clock();
232 std::stringstream stream;
234 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, static_cast<unsigned>(PajeEventType::CreateContainer),
237 stream << std::fixed << std::setprecision(trace_precision) << PajeEventType::CreateContainer << " ";
238 stream << timestamp << " " << c.get_id() << " " << c.type_->get_id() << " " << c.father_->get_id() << " \"";
239 if (c.get_name().find("rank-") != 0)
240 stream << c.get_name() << "\"";
242 /* Subtract -1 because this is the process id and we transform it to the rank id */
243 stream << "rank-" << stoi(c.get_name().substr(5)) - 1 << "\"";
245 XBT_DEBUG("Dump %s", stream.str().c_str());
246 tracing_file << stream.str() << std::endl;
249 static void on_container_destruction_paje(const Container& c)
251 // trace my destruction, but not if user requests so or if the container is root
252 if (not trace_disable_destroy && &c != Container::get_root()) {
253 std::stringstream stream;
254 double timestamp = SIMIX_get_clock();
256 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, static_cast<unsigned>(PajeEventType::DestroyContainer),
259 stream << std::fixed << std::setprecision(trace_precision) << PajeEventType::DestroyContainer << " ";
260 stream << timestamp << " " << c.type_->get_id() << " " << c.get_id();
261 XBT_DEBUG("Dump %s", stream.str().c_str());
262 tracing_file << stream.str() << std::endl;
266 static void on_container_creation_ti(const Container& c)
268 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, static_cast<unsigned>(PajeEventType::CreateContainer),
270 // if we are in the mode with only one file
271 static std::ofstream* ti_unique_file = nullptr;
273 if (tracing_files.empty()) {
274 // generate unique run id with time
275 prefix = xbt_os_time();
278 if (not simgrid::config::get_value<bool>("tracing/smpi/format/ti-one-file") || ti_unique_file == nullptr) {
279 std::string folder_name = simgrid::config::get_value<std::string>("tracing/filename") + "_files";
280 std::string filename = folder_name + "/" + std::to_string(prefix) + "_" + c.get_name() + ".txt";
282 _mkdir(folder_name.c_str());
284 mkdir(folder_name.c_str(), S_IRWXU | S_IRWXG | S_IRWXO);
286 ti_unique_file = new std::ofstream(filename.c_str(), std::ofstream::out);
287 xbt_assert(not ti_unique_file->fail(), "Tracefile %s could not be opened for writing", filename.c_str());
288 tracing_file << filename << std::endl;
290 tracing_files.insert({&c, ti_unique_file});
293 static void on_container_destruction_ti(const Container& c)
295 if (not trace_disable_destroy && &c != Container::get_root()) {
296 if (not simgrid::config::get_value<bool>("tracing/smpi/format/ti-one-file") || tracing_files.size() == 1) {
297 tracing_files.at(&c)->close();
298 delete tracing_files.at(&c);
300 tracing_files.erase(&c);
304 static void on_entity_value_creation(const EntityValue& value)
306 std::stringstream stream;
307 XBT_DEBUG("%s: event_type=%u", __func__, static_cast<unsigned>(PajeEventType::DefineEntityValue));
308 stream << std::fixed << std::setprecision(trace_precision) << PajeEventType::DefineEntityValue;
309 stream << " " << value.get_id() << " " << value.get_father()->get_id() << " " << value.get_name();
310 if (not value.get_color().empty())
311 stream << " \"" << value.get_color() << "\"";
312 XBT_DEBUG("Dump %s", stream.str().c_str());
313 tracing_file << stream.str() << std::endl;
316 static void on_event_creation(PajeEvent& event)
318 XBT_DEBUG("%s: event_type=%u, timestamp=%.*f", __func__, static_cast<unsigned>(event.eventType_), trace_precision,
320 event.stream_ << std::fixed << std::setprecision(trace_precision);
321 event.stream_ << event.eventType_ << " " << event.timestamp_ << " ";
322 event.stream_ << event.get_type()->get_id() << " " << event.get_container()->get_id();
325 static void on_event_destruction(const PajeEvent& event)
327 XBT_DEBUG("Dump %s", event.stream_.str().c_str());
328 tracing_file << event.stream_.str() << std::endl;
331 static void on_state_event_destruction(const StateEvent& event)
333 if (event.has_extra())
334 *tracing_files.at(event.get_container()) << event.stream_.str() << std::endl;
337 static void on_type_creation(const Type& type, PajeEventType event_type)
339 if (event_type == PajeEventType::DefineLinkType)
340 return; // this kind of type has to be handled differently
342 std::stringstream stream;
343 stream << std::fixed << std::setprecision(trace_precision);
344 XBT_DEBUG("%s: event_type=%u, timestamp=%.*f", __func__, static_cast<unsigned>(event_type), trace_precision, 0.);
345 stream << event_type << " " << type.get_id() << " " << type.get_father()->get_id() << " " << type.get_name();
346 if (type.is_colored())
347 stream << " \"" << type.get_color() << "\"";
348 XBT_DEBUG("Dump %s", stream.str().c_str());
349 tracing_file << stream.str() << std::endl;
352 static void on_link_type_creation(const Type& type, const Type& source, const Type& dest)
354 std::stringstream stream;
355 XBT_DEBUG("%s: event_type=%u, timestamp=%.*f", __func__, static_cast<unsigned>(PajeEventType::DefineLinkType),
356 trace_precision, 0.);
357 stream << PajeEventType::DefineLinkType << " " << type.get_id() << " " << type.get_father()->get_id();
358 stream << " " << source.get_id() << " " << dest.get_id() << " " << type.get_name();
359 XBT_DEBUG("Dump %s", stream.str().c_str());
360 tracing_file << stream.str() << std::endl;
363 static void on_simulation_start()
365 if (trace_active || not TRACE_is_enabled())
370 XBT_DEBUG("Tracing starts");
371 trace_precision = config::get_value<int>("tracing/precision");
373 /* init the tracing module to generate the right output */
374 std::string format = config::get_value<std::string>("tracing/smpi/format");
375 XBT_DEBUG("Tracing format %s", format.c_str());
377 /* open the trace file(s) */
378 std::string filename = simgrid::config::get_value<std::string>("tracing/filename");
379 tracing_file.open(filename.c_str(), std::ofstream::out);
380 if (tracing_file.fail()) {
381 throw TracingError(XBT_THROW_POINT,
382 xbt::string_printf("Tracefile %s could not be opened for writing.", filename.c_str()));
385 XBT_DEBUG("Filename %s is open for writing", filename.c_str());
387 if (format == "Paje") {
388 Container::on_creation.connect(on_container_creation_paje);
389 Container::on_destruction.connect(on_container_destruction_paje);
390 EntityValue::on_creation.connect(on_entity_value_creation);
391 Type::on_creation.connect(on_type_creation);
392 LinkType::on_creation.connect(on_link_type_creation);
393 PajeEvent::on_creation.connect(on_event_creation);
394 PajeEvent::on_destruction.connect(on_event_destruction);
396 paje::dump_generator_version();
398 /* output one line comment */
399 std::string comment = simgrid::config::get_value<std::string>("tracing/comment");
400 if (not comment.empty())
401 tracing_file << "# " << comment << std::endl;
403 /* output comment file */
404 paje::dump_comment_file(config::get_value<std::string>(OPT_TRACING_COMMENT_FILE));
405 paje::dump_header(trace_basic, TRACE_display_sizes());
407 trace_format = TraceFormat::Ti;
408 Container::on_creation.connect(on_container_creation_ti);
409 Container::on_destruction.connect(on_container_destruction_ti);
410 StateEvent::on_destruction.connect(on_state_event_destruction);
414 XBT_DEBUG("Tracing is on");
417 static void on_simulation_end()
419 if (not trace_active)
422 /* dump trace buffer */
423 last_timestamp_to_dump = surf_get_clock();
426 const Type* root_type = Container::get_root()->type_;
427 /* destroy all data structures of tracing (and free) */
428 delete Container::get_root();
431 /* close the trace files */
432 tracing_file.close();
433 XBT_DEBUG("Filename %s is closed", config::get_value<std::string>("tracing/filename").c_str());
435 /* de-activate trace */
436 trace_active = false;
437 XBT_DEBUG("Tracing is off");
438 XBT_DEBUG("Tracing system is shutdown");
443 static bool is_initialized = false;
447 is_initialized = true;
449 /* name of the tracefile */
450 config::declare_flag<std::string>("tracing/filename", "Trace file created by the instrumented SimGrid.",
452 config::declare_flag<std::string>("tracing/smpi/format",
453 "Select trace output format used by SMPI. The default is the 'Paje' format. "
454 "The 'TI' (Time-Independent) format allows for trace replay.",
457 config::declare_flag<bool>(OPT_TRACING_FORMAT_TI_ONEFILE,
458 "(smpi only) For replay format only : output to one file only", false);
459 config::declare_flag<std::string>("tracing/comment", "Add a comment line to the top of the trace file.", "");
460 config::declare_flag<std::string>(OPT_TRACING_COMMENT_FILE,
461 "Add the contents of a file as comments to the top of the trace.", "");
462 config::declare_flag<int>("tracing/precision",
463 "Numerical precision used when timestamping events "
464 "(expressed in number of digits after decimal point)",
467 /* Connect Engine callbacks */
468 s4u::Engine::on_platform_creation.connect(on_simulation_start);
469 s4u::Engine::on_time_advance.connect([](double /*time_delta*/) { dump_buffer(false); });
470 s4u::Engine::on_deadlock.connect(on_simulation_end);
471 s4u::Engine::on_simulation_end.connect(on_simulation_end);
474 } // namespace simgrid