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"
12 #include "xbt/virtu.h" /* xbt::cmdline */
16 #include <direct.h> // _mkdir
23 XBT_LOG_NEW_CATEGORY(instr, "Logging the behavior of the tracing system (used for Visualization/Analysis of simulations)");
24 XBT_LOG_NEW_DEFAULT_SUBCATEGORY (instr_config, instr, "Configuration");
26 std::ofstream tracing_file;
27 std::map<container_t, std::ofstream*> tracing_files; // TI specific
28 double prefix = 0.0; // TI specific
30 constexpr char OPT_TRACING_BASIC[] = "tracing/basic";
31 constexpr char OPT_TRACING_COMMENT_FILE[] = "tracing/comment-file";
32 constexpr char OPT_TRACING_DISABLE_DESTROY[] = "tracing/disable-destroy";
33 constexpr char OPT_TRACING_FORMAT_TI_ONEFILE[] = "tracing/smpi/format/ti-one-file";
34 constexpr char OPT_TRACING_SMPI[] = "tracing/smpi";
35 constexpr char OPT_TRACING_TOPOLOGY[] = "tracing/platform/topology";
37 static simgrid::config::Flag<bool> trace_enabled{
38 "tracing", "Enable the tracing system. You have to enable this option to use other tracing options.", false};
40 static simgrid::config::Flag<bool> trace_actor_enabled{
41 "tracing/msg/process", // FIXME rename this flag
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{
77 OPT_TRACING_DISABLE_DESTROY, {"tracing/disable_destroy"}, "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 simgrid::instr::TraceFormat simgrid::instr::trace_format = simgrid::instr::TraceFormat::Paje;
93 bool TRACE_needs_platform ()
95 return TRACE_actor_is_enabled() || TRACE_vm_is_enabled() || TRACE_categorized() || TRACE_uncategorized() ||
96 TRACE_platform() || (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped());
99 bool TRACE_is_enabled()
101 return trace_enabled;
104 bool TRACE_platform()
106 return trace_platform;
109 bool TRACE_platform_topology()
111 return trace_platform_topology;
114 bool TRACE_smpi_is_enabled()
116 return (trace_smpi_enabled || TRACE_smpi_is_grouped()) && TRACE_is_enabled();
119 bool TRACE_smpi_is_grouped()
121 return trace_smpi_grouped;
124 bool TRACE_smpi_is_computing()
126 return trace_smpi_computing;
129 bool TRACE_smpi_is_sleeping()
131 return trace_smpi_sleeping;
134 bool TRACE_smpi_view_internals()
136 return trace_view_internals;
139 bool TRACE_categorized ()
141 return trace_categorized;
144 bool TRACE_uncategorized ()
146 return trace_uncategorized;
149 bool TRACE_actor_is_enabled()
151 return trace_actor_enabled && trace_enabled;
154 bool TRACE_vm_is_enabled()
156 return trace_vm_enabled && trace_enabled;
159 bool TRACE_disable_link()
161 return trace_disable_link && trace_enabled;
164 bool TRACE_disable_speed()
166 return trace_disable_power && trace_enabled;
171 return trace_basic && trace_enabled;
174 bool TRACE_display_sizes ()
176 return trace_display_sizes && trace_smpi_enabled && trace_enabled;
179 int TRACE_precision ()
181 return simgrid::config::get_value<int>("tracing/precision");
184 std::string TRACE_get_filename()
186 return simgrid::config::get_value<std::string>("tracing/filename");
189 static void print_line(const char* option, const char* desc, const char* longdesc)
191 std::string str = std::string("--cfg=") + option + " ";
193 int len = str.size();
194 XBT_HELP("%s%*.*s %s", str.c_str(), 30 - len, 30 - len, "", desc);
195 if (longdesc != nullptr) {
196 XBT_HELP("%s\n", longdesc);
202 XBT_HELP("Description of the tracing options accepted by this simulator:\n");
203 print_line(OPT_TRACING_SMPI, "Trace the MPI Interface (SMPI)",
204 " This option only has effect if this simulator is SMPI-based. Traces the MPI\n"
205 " interface and generates a trace that can be analyzed using Gantt-like\n"
206 " visualizations. Every MPI function (implemented by SMPI) is transformed in a\n"
207 " state, and point-to-point communications can be analyzed with arrows.");
208 print_line(OPT_TRACING_DISABLE_DESTROY, "Disable platform containers destruction",
209 " Disable the destruction of containers at the end of simulation. This can be\n"
210 " used with simulators that have a different notion of time (different from\n"
211 " the simulated time).");
212 print_line(OPT_TRACING_BASIC, "Avoid extended events (impoverished trace file).",
213 " Some visualization tools are not able to parse correctly the Paje file format.\n"
214 " Use this option if you are using one of these tools to visualize the simulation\n"
215 " trace. Keep in mind that the trace might be incomplete, without all the\n"
216 " information that would be registered otherwise.");
217 print_line(OPT_TRACING_FORMAT_TI_ONEFILE, "Only works for SMPI now, and TI output format",
218 " By default, each process outputs to a separate file, inside a filename_files folder\n"
219 " By setting this option to yes, all processes will output to only one file\n"
220 " This is meant to avoid opening thousands of files with large simulations");
221 print_line(OPT_TRACING_TOPOLOGY, "Register the platform topology as a graph",
222 " This option (enabled by default) can be used to disable the tracing of\n"
223 " the platform topology in the trace file. Sometimes, such task is really\n"
224 " time consuming, since it must get the route from each host to other hosts\n"
225 " within the same Autonomous System (AS).");
231 static bool trace_active = false;
235 xbt::signal<void(Container&)> Container::on_creation;
236 xbt::signal<void(Container&)> Container::on_destruction;
237 xbt::signal<void(EntityValue&)> EntityValue::on_creation;
238 xbt::signal<void(Type&, e_event_type)> Type::on_creation;
239 xbt::signal<void(LinkType&, Type&, Type&)> LinkType::on_creation;
241 static void on_container_creation_paje(Container& c)
243 double timestamp = SIMIX_get_clock();
244 std::stringstream stream;
246 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_CreateContainer, timestamp);
248 stream << std::fixed << std::setprecision(TRACE_precision()) << PAJE_CreateContainer << " ";
249 stream << timestamp << " " << c.get_id() << " " << c.type_->get_id() << " " << c.father_->get_id() << " \"";
250 if (c.get_name().find("rank-") != 0)
251 stream << c.get_name() << "\"";
253 /* Subtract -1 because this is the process id and we transform it to the rank id */
254 stream << "rank-" << stoi(c.get_name().substr(5)) - 1 << "\"";
256 XBT_DEBUG("Dump %s", stream.str().c_str());
257 tracing_file << stream.str() << std::endl;
260 static void on_container_destruction_paje(Container& c)
262 // obligation to dump previous events because they might reference the container that is about to be destroyed
263 TRACE_last_timestamp_to_dump = SIMIX_get_clock();
264 TRACE_paje_dump_buffer(true);
266 // trace my destruction, but not if user requests so or if the container is root
267 if (not trace_disable_destroy && &c != Container::get_root()) {
268 std::stringstream stream;
269 double timestamp = SIMIX_get_clock();
271 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_DestroyContainer, timestamp);
273 stream << std::fixed << std::setprecision(TRACE_precision()) << PAJE_DestroyContainer << " ";
274 stream << timestamp << " " << c.type_->get_id() << " " << c.get_id();
275 XBT_DEBUG("Dump %s", stream.str().c_str());
276 tracing_file << stream.str() << std::endl;
280 static void on_container_creation_ti(Container& c)
282 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_CreateContainer, SIMIX_get_clock());
283 // if we are in the mode with only one file
284 static std::ofstream* ti_unique_file = nullptr;
286 if (tracing_files.empty()) {
287 // generate unique run id with time
288 prefix = xbt_os_time();
291 if (not simgrid::config::get_value<bool>("tracing/smpi/format/ti-one-file") || ti_unique_file == nullptr) {
292 std::string folder_name = simgrid::config::get_value<std::string>("tracing/filename") + "_files";
293 std::string filename = folder_name + "/" + std::to_string(prefix) + "_" + c.get_name() + ".txt";
295 _mkdir(folder_name.c_str());
297 mkdir(folder_name.c_str(), S_IRWXU | S_IRWXG | S_IRWXO);
299 ti_unique_file = new std::ofstream(filename.c_str(), std::ofstream::out);
300 xbt_assert(not ti_unique_file->fail(), "Tracefile %s could not be opened for writing", filename.c_str());
301 tracing_file << filename << std::endl;
303 tracing_files.insert({&c, ti_unique_file});
306 static void on_container_destruction_ti(Container& c)
308 // obligation to dump previous events because they might reference the container that is about to be destroyed
309 TRACE_last_timestamp_to_dump = SIMIX_get_clock();
310 TRACE_paje_dump_buffer(true);
312 if (not trace_disable_destroy && &c != Container::get_root()) {
313 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_DestroyContainer, SIMIX_get_clock());
314 if (not simgrid::config::get_value<bool>("tracing/smpi/format/ti-one-file") || tracing_files.size() == 1) {
315 tracing_files.at(&c)->close();
316 delete tracing_files.at(&c);
318 tracing_files.erase(&c);
322 static void on_entity_value_creation(EntityValue& value)
324 std::stringstream stream;
325 XBT_DEBUG("%s: event_type=%u", __func__, PAJE_DefineEntityValue);
326 stream << std::fixed << std::setprecision(TRACE_precision()) << PAJE_DefineEntityValue;
327 stream << " " << value.get_id() << " " << value.get_father()->get_id() << " " << value.get_name();
328 if (not value.get_color().empty())
329 stream << " \"" << value.get_color() << "\"";
330 XBT_DEBUG("Dump %s", stream.str().c_str());
331 tracing_file << stream.str() << std::endl;
334 static void on_type_creation(Type& type, e_event_type event_type)
336 if (event_type == PAJE_DefineLinkType)
337 return; // this kind of type has to be handled differently
339 std::stringstream stream;
340 stream << std::fixed << std::setprecision(TRACE_precision());
341 XBT_DEBUG("%s: event_type=%u, timestamp=%.*f", __func__, event_type, TRACE_precision(), 0.);
342 stream << event_type << " " << type.get_id() << " " << type.get_father()->get_id() << " " << type.get_name();
343 if (type.is_colored())
344 stream << " \"" << type.get_color() << "\"";
345 XBT_DEBUG("Dump %s", stream.str().c_str());
346 tracing_file << stream.str() << std::endl;
349 static void on_link_type_creation(Type& type, Type& source, Type& dest)
351 std::stringstream stream;
352 XBT_DEBUG("%s: event_type=%u, timestamp=%.*f", __func__, PAJE_DefineLinkType, TRACE_precision(), 0.);
353 stream << PAJE_DefineLinkType << " " << type.get_id() << " " << type.get_father()->get_id();
354 stream << " " << source.get_id() << " " << dest.get_id() << " " << type.get_name();
355 XBT_DEBUG("Dump %s", stream.str().c_str());
356 tracing_file << stream.str() << std::endl;
358 static void on_simulation_start()
363 // tracing system must be:
364 // - enabled (with --cfg=tracing:yes)
365 // - already configured (simgrid::instr::init already called)
366 if (TRACE_is_enabled()) {
369 XBT_DEBUG("Tracing starts");
371 /* init the tracing module to generate the right output */
372 std::string format = config::get_value<std::string>("tracing/smpi/format");
373 XBT_DEBUG("Tracing format %s", format.c_str());
375 /* Connect the callbacks associated to the creation/destruction of containers*/
376 if (format == "Paje") {
377 Container::on_creation.connect(on_container_creation_paje);
378 Container::on_destruction.connect(on_container_destruction_paje);
379 EntityValue::on_creation.connect(on_entity_value_creation);
380 Type::on_creation.connect(on_type_creation);
381 LinkType::on_creation.connect(on_link_type_creation);
383 Container::on_creation.connect(on_container_creation_ti);
384 Container::on_destruction.connect(on_container_destruction_ti);
387 /* open the trace file(s) */
388 std::string filename = TRACE_get_filename();
389 tracing_file.open(filename.c_str(), std::ofstream::out);
390 if (tracing_file.fail()) {
391 throw TracingError(XBT_THROW_POINT,
392 xbt::string_printf("Tracefile %s could not be opened for writing.", filename.c_str()));
395 XBT_DEBUG("Filename %s is open for writing", filename.c_str());
397 if (format == "Paje") {
398 /* output generator version */
399 tracing_file << "#This file was generated using SimGrid-" << SIMGRID_VERSION_MAJOR << "." << SIMGRID_VERSION_MINOR
400 << "." << SIMGRID_VERSION_PATCH << std::endl;
401 tracing_file << "#[";
402 for (auto str : simgrid::xbt::cmdline) {
403 tracing_file << str << " ";
405 tracing_file << "]" << std::endl;
408 /* output one line comment */
409 std::string comment = simgrid::config::get_value<std::string>("tracing/comment");
410 if (not comment.empty())
411 tracing_file << "# " << comment << std::endl;
413 /* output comment file */
414 dump_comment_file(simgrid::config::get_value<std::string>(OPT_TRACING_COMMENT_FILE));
416 if (format == "Paje") {
417 /* output Pajé header */
418 TRACE_header(TRACE_basic(), TRACE_display_sizes());
420 trace_format = TraceFormat::Ti;
423 XBT_DEBUG("Tracing is on");
427 static void on_simulation_end()
429 if (not trace_active)
432 /* dump trace buffer */
433 TRACE_last_timestamp_to_dump = surf_get_clock();
434 TRACE_paje_dump_buffer(true);
436 const Type* root_type = Container::get_root()->type_;
437 /* destroy all data structures of tracing (and free) */
438 delete Container::get_root();
441 /* close the trace files */
442 tracing_file.close();
443 XBT_DEBUG("Filename %s is closed", TRACE_get_filename().c_str());
445 /* de-activate trace */
446 trace_active = false;
447 XBT_DEBUG("Tracing is off");
448 XBT_DEBUG("Tracing system is shutdown");
453 static bool is_initialized = false;
457 is_initialized = true;
459 /* name of the tracefile */
460 config::declare_flag<std::string>("tracing/filename", "Trace file created by the instrumented SimGrid.",
462 config::declare_flag<std::string>("tracing/smpi/format",
463 "Select trace output format used by SMPI. The default is the 'Paje' format. "
464 "The 'TI' (Time-Independent) format allows for trace replay.",
467 config::declare_flag<bool>(OPT_TRACING_FORMAT_TI_ONEFILE,
468 "(smpi only) For replay format only : output to one file only", false);
469 config::alias(OPT_TRACING_FORMAT_TI_ONEFILE, {"tracing/smpi/format/ti_one_file"});
470 config::declare_flag<std::string>("tracing/comment", "Add a comment line to the top of the trace file.", "");
471 config::declare_flag<std::string>(OPT_TRACING_COMMENT_FILE,
472 "Add the contents of a file as comments to the top of the trace.", "");
473 config::alias(OPT_TRACING_COMMENT_FILE, {"tracing/comment_file"});
474 config::declare_flag<int>("tracing/precision",
475 "Numerical precision used when timestamping events "
476 "(expressed in number of digits after decimal point)",
479 /* Connect callbacks */
480 s4u::Engine::on_platform_creation.connect(on_simulation_start);
481 s4u::Engine::on_deadlock.connect(on_simulation_end);
482 s4u::Engine::on_simulation_end.connect(on_simulation_end);
485 } // namespace simgrid