1 /* Copyright (c) 2010-2020. 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<container_t, 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{
40 "tracing/msg/process", // FIXME rename this flag
41 "Trace the behavior of all categorized actors, grouping them by host. "
42 "Can be used to track actor location if the simulator does actor migration.",
45 static simgrid::config::Flag<bool> trace_vm_enabled{"tracing/vm", "Trace the behavior of all virtual machines.", false};
47 static simgrid::config::Flag<bool> trace_platform{"tracing/platform",
48 "Register the platform in the trace as a hierarchy.", false};
50 static simgrid::config::Flag<bool> trace_platform_topology{
51 OPT_TRACING_TOPOLOGY, "Register the platform topology in the trace as a graph.", true};
52 static simgrid::config::Flag<bool> trace_smpi_enabled{OPT_TRACING_SMPI, "Tracing of the SMPI interface.", false};
53 static simgrid::config::Flag<bool> trace_smpi_grouped{"tracing/smpi/group", "Group MPI processes by host.", false};
55 static simgrid::config::Flag<bool> trace_smpi_computing{
56 "tracing/smpi/computing", "Generate 'Computing' states to trace the out-of-SMPI parts of the application", false};
58 static simgrid::config::Flag<bool> trace_smpi_sleeping{
59 "tracing/smpi/sleeping", "Generate 'Sleeping' states for the sleeps in the application that do not pertain to SMPI",
62 static simgrid::config::Flag<bool> trace_view_internals{
63 "tracing/smpi/internals",
64 "Generate tracing events corresponding to point-to-point messages sent by SMPI collective communications", false};
66 static simgrid::config::Flag<bool> trace_categorized{
67 "tracing/categorized", "Trace categorized resource utilization of hosts and links.", false};
69 static simgrid::config::Flag<bool> trace_uncategorized{
70 "tracing/uncategorized",
71 "Trace uncategorized resource utilization of hosts and links. "
72 "To use if the simulator does not use tracing categories but resource utilization have to be traced.",
75 static simgrid::config::Flag<bool> trace_disable_destroy{
76 OPT_TRACING_DISABLE_DESTROY, {"tracing/disable_destroy"}, "Disable platform containers destruction.", false};
77 static simgrid::config::Flag<bool> trace_basic{OPT_TRACING_BASIC, "Avoid extended events (impoverished trace file).",
80 static simgrid::config::Flag<bool> trace_display_sizes{
81 "tracing/smpi/display-sizes",
82 "Add message size information (in bytes) to the to links and states (SMPI only). "
83 "For collectives, it usually corresponds to the total number of bytes sent by a process.",
86 static simgrid::config::Flag<bool> trace_disable_link{"tracing/disable_link",
87 "Do not trace link bandwidth and latency.", false};
88 static simgrid::config::Flag<bool> trace_disable_power{"tracing/disable_power", "Do not trace host power.", false};
90 bool TRACE_needs_platform ()
92 return TRACE_actor_is_enabled() || TRACE_vm_is_enabled() || TRACE_categorized() || TRACE_uncategorized() ||
93 TRACE_platform() || (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped());
96 bool TRACE_is_enabled()
101 bool TRACE_platform()
103 return trace_platform;
106 bool TRACE_platform_topology()
108 return trace_platform_topology;
111 bool TRACE_smpi_is_enabled()
113 return (trace_smpi_enabled || TRACE_smpi_is_grouped()) && TRACE_is_enabled();
116 bool TRACE_smpi_is_grouped()
118 return trace_smpi_grouped;
121 bool TRACE_smpi_is_computing()
123 return trace_smpi_computing;
126 bool TRACE_smpi_is_sleeping()
128 return trace_smpi_sleeping;
131 bool TRACE_smpi_view_internals()
133 return trace_view_internals;
136 bool TRACE_categorized ()
138 return trace_categorized;
141 bool TRACE_uncategorized ()
143 return trace_uncategorized;
146 bool TRACE_actor_is_enabled()
148 return trace_actor_enabled && trace_enabled;
151 bool TRACE_vm_is_enabled()
153 return trace_vm_enabled && trace_enabled;
156 bool TRACE_disable_link()
158 return trace_disable_link && trace_enabled;
161 bool TRACE_disable_speed()
163 return trace_disable_power && trace_enabled;
166 bool TRACE_display_sizes ()
168 return trace_display_sizes && trace_smpi_enabled && trace_enabled;
171 static void print_line(const char* option, const char* desc, const char* longdesc)
173 std::string str = std::string("--cfg=") + option + " ";
175 int len = static_cast<int>(str.size());
176 XBT_HELP("%s%*.*s %s", str.c_str(), 30 - len, 30 - len, "", desc);
177 if (longdesc != nullptr) {
178 XBT_HELP("%s\n", longdesc);
184 XBT_HELP("Description of the tracing options accepted by this simulator:\n");
185 print_line(OPT_TRACING_SMPI, "Trace the MPI Interface (SMPI)",
186 " This option only has effect if this simulator is SMPI-based. Traces the MPI\n"
187 " interface and generates a trace that can be analyzed using Gantt-like\n"
188 " visualizations. Every MPI function (implemented by SMPI) is transformed in a\n"
189 " state, and point-to-point communications can be analyzed with arrows.");
190 print_line(OPT_TRACING_DISABLE_DESTROY, "Disable platform containers destruction",
191 " Disable the destruction of containers at the end of simulation. This can be\n"
192 " used with simulators that have a different notion of time (different from\n"
193 " the simulated time).");
194 print_line(OPT_TRACING_BASIC, "Avoid extended events (impoverished trace file).",
195 " Some visualization tools are not able to parse correctly the Paje file format.\n"
196 " Use this option if you are using one of these tools to visualize the simulation\n"
197 " trace. Keep in mind that the trace might be incomplete, without all the\n"
198 " information that would be registered otherwise.");
199 print_line(OPT_TRACING_FORMAT_TI_ONEFILE, "Only works for SMPI now, and TI output format",
200 " By default, each process outputs to a separate file, inside a filename_files folder\n"
201 " By setting this option to yes, all processes will output to only one file\n"
202 " This is meant to avoid opening thousands of files with large simulations");
203 print_line(OPT_TRACING_TOPOLOGY, "Register the platform topology as a graph",
204 " This option (enabled by default) can be used to disable the tracing of\n"
205 " the platform topology in the trace file. Sometimes, such task is really\n"
206 " time consuming, since it must get the route from each host to other hosts\n"
207 " within the same Autonomous System (AS).");
212 static bool trace_active = false;
213 TraceFormat trace_format = TraceFormat::Paje;
219 xbt::signal<void(Container&)> Container::on_creation;
220 xbt::signal<void(Container&)> Container::on_destruction;
221 xbt::signal<void(Type&, e_event_type)> Type::on_creation;
222 xbt::signal<void(LinkType&, Type&, Type&)> LinkType::on_creation;
223 xbt::signal<void(PajeEvent&)> PajeEvent::on_creation;
224 xbt::signal<void(PajeEvent&)> PajeEvent::on_destruction;
225 xbt::signal<void(StateEvent&)> StateEvent::on_destruction;
226 xbt::signal<void(EntityValue&)> EntityValue::on_creation;
228 static void on_container_creation_paje(const Container& c)
230 double timestamp = SIMIX_get_clock();
231 std::stringstream stream;
233 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_CreateContainer, timestamp);
235 stream << std::fixed << std::setprecision(trace_precision) << PAJE_CreateContainer << " ";
236 stream << timestamp << " " << c.get_id() << " " << c.type_->get_id() << " " << c.father_->get_id() << " \"";
237 if (c.get_name().find("rank-") != 0)
238 stream << c.get_name() << "\"";
240 /* Subtract -1 because this is the process id and we transform it to the rank id */
241 stream << "rank-" << stoi(c.get_name().substr(5)) - 1 << "\"";
243 XBT_DEBUG("Dump %s", stream.str().c_str());
244 tracing_file << stream.str() << std::endl;
247 static void on_container_destruction_paje(const Container& c)
249 // trace my destruction, but not if user requests so or if the container is root
250 if (not trace_disable_destroy && &c != Container::get_root()) {
251 std::stringstream stream;
252 double timestamp = SIMIX_get_clock();
254 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_DestroyContainer, timestamp);
256 stream << std::fixed << std::setprecision(trace_precision) << PAJE_DestroyContainer << " ";
257 stream << timestamp << " " << c.type_->get_id() << " " << c.get_id();
258 XBT_DEBUG("Dump %s", stream.str().c_str());
259 tracing_file << stream.str() << std::endl;
263 static void on_container_creation_ti(Container& c)
265 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_CreateContainer, SIMIX_get_clock());
266 // if we are in the mode with only one file
267 static std::ofstream* ti_unique_file = nullptr;
269 if (tracing_files.empty()) {
270 // generate unique run id with time
271 prefix = xbt_os_time();
274 if (not simgrid::config::get_value<bool>("tracing/smpi/format/ti-one-file") || ti_unique_file == nullptr) {
275 std::string folder_name = simgrid::config::get_value<std::string>("tracing/filename") + "_files";
276 std::string filename = folder_name + "/" + std::to_string(prefix) + "_" + c.get_name() + ".txt";
278 _mkdir(folder_name.c_str());
280 mkdir(folder_name.c_str(), S_IRWXU | S_IRWXG | S_IRWXO);
282 ti_unique_file = new std::ofstream(filename.c_str(), std::ofstream::out);
283 xbt_assert(not ti_unique_file->fail(), "Tracefile %s could not be opened for writing", filename.c_str());
284 tracing_file << filename << std::endl;
286 tracing_files.insert({&c, ti_unique_file});
289 static void on_container_destruction_ti(Container& c)
291 if (not trace_disable_destroy && &c != Container::get_root()) {
292 if (not simgrid::config::get_value<bool>("tracing/smpi/format/ti-one-file") || tracing_files.size() == 1) {
293 tracing_files.at(&c)->close();
294 delete tracing_files.at(&c);
296 tracing_files.erase(&c);
300 static void on_entity_value_creation(const EntityValue& value)
302 std::stringstream stream;
303 XBT_DEBUG("%s: event_type=%u", __func__, PAJE_DefineEntityValue);
304 stream << std::fixed << std::setprecision(trace_precision) << PAJE_DefineEntityValue;
305 stream << " " << value.get_id() << " " << value.get_father()->get_id() << " " << value.get_name();
306 if (not value.get_color().empty())
307 stream << " \"" << value.get_color() << "\"";
308 XBT_DEBUG("Dump %s", stream.str().c_str());
309 tracing_file << stream.str() << std::endl;
312 static void on_event_creation(PajeEvent& event)
314 XBT_DEBUG("%s: event_type=%u, timestamp=%.*f", __func__, event.eventType_, trace_precision, event.timestamp_);
315 event.stream_ << std::fixed << std::setprecision(trace_precision);
316 event.stream_ << event.eventType_ << " " << event.timestamp_ << " ";
317 event.stream_ << event.get_type()->get_id() << " " << event.get_container()->get_id();
320 static void on_event_destruction(const PajeEvent& event)
322 XBT_DEBUG("Dump %s", event.stream_.str().c_str());
323 tracing_file << event.stream_.str() << std::endl;
326 static void on_state_event_destruction(const StateEvent& event)
328 if (event.has_extra())
329 *tracing_files.at(event.get_container()) << event.stream_.str() << std::endl;
332 static void on_type_creation(const Type& type, e_event_type event_type)
334 if (event_type == PAJE_DefineLinkType)
335 return; // this kind of type has to be handled differently
337 std::stringstream stream;
338 stream << std::fixed << std::setprecision(trace_precision);
339 XBT_DEBUG("%s: event_type=%u, timestamp=%.*f", __func__, event_type, trace_precision, 0.);
340 stream << event_type << " " << type.get_id() << " " << type.get_father()->get_id() << " " << type.get_name();
341 if (type.is_colored())
342 stream << " \"" << type.get_color() << "\"";
343 XBT_DEBUG("Dump %s", stream.str().c_str());
344 tracing_file << stream.str() << std::endl;
347 static void on_link_type_creation(const Type& type, const Type& source, const Type& dest)
349 std::stringstream stream;
350 XBT_DEBUG("%s: event_type=%u, timestamp=%.*f", __func__, PAJE_DefineLinkType, trace_precision, 0.);
351 stream << PAJE_DefineLinkType << " " << type.get_id() << " " << type.get_father()->get_id();
352 stream << " " << source.get_id() << " " << dest.get_id() << " " << type.get_name();
353 XBT_DEBUG("Dump %s", stream.str().c_str());
354 tracing_file << stream.str() << std::endl;
357 static void on_simulation_start()
359 if (trace_active || not TRACE_is_enabled())
364 XBT_DEBUG("Tracing starts");
365 trace_precision = config::get_value<int>("tracing/precision");
367 /* init the tracing module to generate the right output */
368 std::string format = config::get_value<std::string>("tracing/smpi/format");
369 XBT_DEBUG("Tracing format %s", format.c_str());
371 /* open the trace file(s) */
372 std::string filename = simgrid::config::get_value<std::string>("tracing/filename");
373 tracing_file.open(filename.c_str(), std::ofstream::out);
374 if (tracing_file.fail()) {
375 throw TracingError(XBT_THROW_POINT,
376 xbt::string_printf("Tracefile %s could not be opened for writing.", filename.c_str()));
379 XBT_DEBUG("Filename %s is open for writing", filename.c_str());
381 if (format == "Paje") {
382 Container::on_creation.connect(on_container_creation_paje);
383 Container::on_destruction.connect(on_container_destruction_paje);
384 EntityValue::on_creation.connect(on_entity_value_creation);
385 Type::on_creation.connect(on_type_creation);
386 LinkType::on_creation.connect(on_link_type_creation);
387 PajeEvent::on_creation.connect(on_event_creation);
388 PajeEvent::on_destruction.connect(on_event_destruction);
390 paje::dump_generator_version();
392 /* output one line comment */
393 std::string comment = simgrid::config::get_value<std::string>("tracing/comment");
394 if (not comment.empty())
395 tracing_file << "# " << comment << std::endl;
397 /* output comment file */
398 paje::dump_comment_file(config::get_value<std::string>(OPT_TRACING_COMMENT_FILE));
399 paje::dump_header(trace_basic, TRACE_display_sizes());
401 trace_format = TraceFormat::Ti;
402 Container::on_creation.connect(on_container_creation_ti);
403 Container::on_destruction.connect(on_container_destruction_ti);
404 StateEvent::on_destruction.connect(on_state_event_destruction);
408 XBT_DEBUG("Tracing is on");
411 static void on_simulation_end()
413 if (not trace_active)
416 /* dump trace buffer */
417 last_timestamp_to_dump = surf_get_clock();
420 const Type* root_type = Container::get_root()->type_;
421 /* destroy all data structures of tracing (and free) */
422 delete Container::get_root();
425 /* close the trace files */
426 tracing_file.close();
427 XBT_DEBUG("Filename %s is closed", config::get_value<std::string>("tracing/filename").c_str());
429 /* de-activate trace */
430 trace_active = false;
431 XBT_DEBUG("Tracing is off");
432 XBT_DEBUG("Tracing system is shutdown");
437 static bool is_initialized = false;
441 is_initialized = true;
443 /* name of the tracefile */
444 config::declare_flag<std::string>("tracing/filename", "Trace file created by the instrumented SimGrid.",
446 config::declare_flag<std::string>("tracing/smpi/format",
447 "Select trace output format used by SMPI. The default is the 'Paje' format. "
448 "The 'TI' (Time-Independent) format allows for trace replay.",
451 config::declare_flag<bool>(OPT_TRACING_FORMAT_TI_ONEFILE,
452 "(smpi only) For replay format only : output to one file only", false);
453 config::alias(OPT_TRACING_FORMAT_TI_ONEFILE, {"tracing/smpi/format/ti_one_file"});
454 config::declare_flag<std::string>("tracing/comment", "Add a comment line to the top of the trace file.", "");
455 config::declare_flag<std::string>(OPT_TRACING_COMMENT_FILE,
456 "Add the contents of a file as comments to the top of the trace.", "");
457 config::alias(OPT_TRACING_COMMENT_FILE, {"tracing/comment_file"});
458 config::declare_flag<int>("tracing/precision",
459 "Numerical precision used when timestamping events "
460 "(expressed in number of digits after decimal point)",
463 /* Connect Engine callbacks */
464 s4u::Engine::on_platform_creation.connect(on_simulation_start);
465 s4u::Engine::on_time_advance.connect([](double /*time_delta*/) { dump_buffer(false); });
466 s4u::Engine::on_deadlock.connect(on_simulation_end);
467 s4u::Engine::on_simulation_end.connect(on_simulation_end);
470 } // namespace simgrid