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 = 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(EntityValue&)> EntityValue::on_creation;
222 xbt::signal<void(Type&, e_event_type)> Type::on_creation;
223 xbt::signal<void(LinkType&, Type&, Type&)> LinkType::on_creation;
225 static void on_container_creation_paje(Container& c)
227 double timestamp = SIMIX_get_clock();
228 std::stringstream stream;
230 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_CreateContainer, timestamp);
232 stream << std::fixed << std::setprecision(trace_precision) << PAJE_CreateContainer << " ";
233 stream << timestamp << " " << c.get_id() << " " << c.type_->get_id() << " " << c.father_->get_id() << " \"";
234 if (c.get_name().find("rank-") != 0)
235 stream << c.get_name() << "\"";
237 /* Subtract -1 because this is the process id and we transform it to the rank id */
238 stream << "rank-" << stoi(c.get_name().substr(5)) - 1 << "\"";
240 XBT_DEBUG("Dump %s", stream.str().c_str());
241 tracing_file << stream.str() << std::endl;
244 static void on_container_destruction_paje(Container& c)
246 // obligation to dump previous events because they might reference the container that is about to be destroyed
247 last_timestamp_to_dump = SIMIX_get_clock();
250 // trace my destruction, but not if user requests so or if the container is root
251 if (not trace_disable_destroy && &c != Container::get_root()) {
252 std::stringstream stream;
253 double timestamp = SIMIX_get_clock();
255 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_DestroyContainer, timestamp);
257 stream << std::fixed << std::setprecision(trace_precision) << PAJE_DestroyContainer << " ";
258 stream << timestamp << " " << c.type_->get_id() << " " << c.get_id();
259 XBT_DEBUG("Dump %s", stream.str().c_str());
260 tracing_file << stream.str() << std::endl;
264 static void on_container_creation_ti(Container& c)
266 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_CreateContainer, SIMIX_get_clock());
267 // if we are in the mode with only one file
268 static std::ofstream* ti_unique_file = nullptr;
270 if (tracing_files.empty()) {
271 // generate unique run id with time
272 prefix = xbt_os_time();
275 if (not simgrid::config::get_value<bool>("tracing/smpi/format/ti-one-file") || ti_unique_file == nullptr) {
276 std::string folder_name = simgrid::config::get_value<std::string>("tracing/filename") + "_files";
277 std::string filename = folder_name + "/" + std::to_string(prefix) + "_" + c.get_name() + ".txt";
279 _mkdir(folder_name.c_str());
281 mkdir(folder_name.c_str(), S_IRWXU | S_IRWXG | S_IRWXO);
283 ti_unique_file = new std::ofstream(filename.c_str(), std::ofstream::out);
284 xbt_assert(not ti_unique_file->fail(), "Tracefile %s could not be opened for writing", filename.c_str());
285 tracing_file << filename << std::endl;
287 tracing_files.insert({&c, ti_unique_file});
290 static void on_container_destruction_ti(Container& c)
292 // obligation to dump previous events because they might reference the container that is about to be destroyed
293 last_timestamp_to_dump = SIMIX_get_clock();
296 if (not trace_disable_destroy && &c != Container::get_root()) {
297 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_DestroyContainer, SIMIX_get_clock());
298 if (not simgrid::config::get_value<bool>("tracing/smpi/format/ti-one-file") || tracing_files.size() == 1) {
299 tracing_files.at(&c)->close();
300 delete tracing_files.at(&c);
302 tracing_files.erase(&c);
306 static void on_entity_value_creation(EntityValue& value)
308 std::stringstream stream;
309 XBT_DEBUG("%s: event_type=%u", __func__, PAJE_DefineEntityValue);
310 stream << std::fixed << std::setprecision(trace_precision) << PAJE_DefineEntityValue;
311 stream << " " << value.get_id() << " " << value.get_father()->get_id() << " " << value.get_name();
312 if (not value.get_color().empty())
313 stream << " \"" << value.get_color() << "\"";
314 XBT_DEBUG("Dump %s", stream.str().c_str());
315 tracing_file << stream.str() << std::endl;
318 static void on_type_creation(Type& type, e_event_type event_type)
320 if (event_type == PAJE_DefineLinkType)
321 return; // this kind of type has to be handled differently
323 std::stringstream stream;
324 stream << std::fixed << std::setprecision(trace_precision);
325 XBT_DEBUG("%s: event_type=%u, timestamp=%.*f", __func__, event_type, trace_precision, 0.);
326 stream << event_type << " " << type.get_id() << " " << type.get_father()->get_id() << " " << type.get_name();
327 if (type.is_colored())
328 stream << " \"" << type.get_color() << "\"";
329 XBT_DEBUG("Dump %s", stream.str().c_str());
330 tracing_file << stream.str() << std::endl;
333 static void on_link_type_creation(Type& type, Type& source, Type& dest)
335 std::stringstream stream;
336 XBT_DEBUG("%s: event_type=%u, timestamp=%.*f", __func__, PAJE_DefineLinkType, trace_precision, 0.);
337 stream << PAJE_DefineLinkType << " " << type.get_id() << " " << type.get_father()->get_id();
338 stream << " " << source.get_id() << " " << dest.get_id() << " " << type.get_name();
339 XBT_DEBUG("Dump %s", stream.str().c_str());
340 tracing_file << stream.str() << std::endl;
343 static void on_simulation_start()
345 if (trace_active || not TRACE_is_enabled())
350 XBT_DEBUG("Tracing starts");
351 trace_precision = config::get_value<int>("tracing/precision");
353 /* init the tracing module to generate the right output */
354 std::string format = config::get_value<std::string>("tracing/smpi/format");
355 XBT_DEBUG("Tracing format %s", format.c_str());
357 /* open the trace file(s) */
358 std::string filename = simgrid::config::get_value<std::string>("tracing/filename");
359 tracing_file.open(filename.c_str(), std::ofstream::out);
360 if (tracing_file.fail()) {
361 throw TracingError(XBT_THROW_POINT,
362 xbt::string_printf("Tracefile %s could not be opened for writing.", filename.c_str()));
365 XBT_DEBUG("Filename %s is open for writing", filename.c_str());
367 if (format == "Paje") {
368 Container::on_creation.connect(on_container_creation_paje);
369 Container::on_destruction.connect(on_container_destruction_paje);
370 EntityValue::on_creation.connect(on_entity_value_creation);
371 Type::on_creation.connect(on_type_creation);
372 LinkType::on_creation.connect(on_link_type_creation);
374 paje::dump_generator_version();
376 /* output one line comment */
377 std::string comment = simgrid::config::get_value<std::string>("tracing/comment");
378 if (not comment.empty())
379 tracing_file << "# " << comment << std::endl;
381 /* output comment file */
382 paje::dump_comment_file(config::get_value<std::string>(OPT_TRACING_COMMENT_FILE));
383 paje::dump_header(trace_basic, TRACE_display_sizes());
385 trace_format = TraceFormat::Ti;
386 Container::on_creation.connect(on_container_creation_ti);
387 Container::on_destruction.connect(on_container_destruction_ti);
391 XBT_DEBUG("Tracing is on");
394 static void on_simulation_end()
396 if (not trace_active)
399 /* dump trace buffer */
400 last_timestamp_to_dump = surf_get_clock();
403 const Type* root_type = Container::get_root()->type_;
404 /* destroy all data structures of tracing (and free) */
405 delete Container::get_root();
408 /* close the trace files */
409 tracing_file.close();
410 XBT_DEBUG("Filename %s is closed", config::get_value<std::string>("tracing/filename").c_str());
412 /* de-activate trace */
413 trace_active = false;
414 XBT_DEBUG("Tracing is off");
415 XBT_DEBUG("Tracing system is shutdown");
420 static bool is_initialized = false;
424 is_initialized = true;
426 /* name of the tracefile */
427 config::declare_flag<std::string>("tracing/filename", "Trace file created by the instrumented SimGrid.",
429 config::declare_flag<std::string>("tracing/smpi/format",
430 "Select trace output format used by SMPI. The default is the 'Paje' format. "
431 "The 'TI' (Time-Independent) format allows for trace replay.",
434 config::declare_flag<bool>(OPT_TRACING_FORMAT_TI_ONEFILE,
435 "(smpi only) For replay format only : output to one file only", false);
436 config::alias(OPT_TRACING_FORMAT_TI_ONEFILE, {"tracing/smpi/format/ti_one_file"});
437 config::declare_flag<std::string>("tracing/comment", "Add a comment line to the top of the trace file.", "");
438 config::declare_flag<std::string>(OPT_TRACING_COMMENT_FILE,
439 "Add the contents of a file as comments to the top of the trace.", "");
440 config::alias(OPT_TRACING_COMMENT_FILE, {"tracing/comment_file"});
441 config::declare_flag<int>("tracing/precision",
442 "Numerical precision used when timestamping events "
443 "(expressed in number of digits after decimal point)",
446 /* Connect Engine callbacks */
447 s4u::Engine::on_platform_creation.connect(on_simulation_start);
448 s4u::Engine::on_time_advance.connect([](double /*time_delta*/) { dump_buffer(false); });
449 s4u::Engine::on_deadlock.connect(on_simulation_end);
450 s4u::Engine::on_simulation_end.connect(on_simulation_end);
453 } // namespace simgrid