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;
169 bool TRACE_disable_destroy ()
171 return trace_disable_destroy && trace_enabled;
176 return trace_basic && trace_enabled;
179 bool TRACE_display_sizes ()
181 return trace_display_sizes && trace_smpi_enabled && trace_enabled;
184 int TRACE_precision ()
186 return simgrid::config::get_value<int>("tracing/precision");
189 std::string TRACE_get_filename()
191 return simgrid::config::get_value<std::string>("tracing/filename");
194 static void print_line(const char* option, const char* desc, const char* longdesc)
196 std::string str = std::string("--cfg=") + option + " ";
198 int len = str.size();
199 XBT_HELP("%s%*.*s %s", str.c_str(), 30 - len, 30 - len, "", desc);
200 if (longdesc != nullptr) {
201 XBT_HELP("%s\n", longdesc);
207 XBT_HELP("Description of the tracing options accepted by this simulator:\n");
208 print_line(OPT_TRACING_SMPI, "Trace the MPI Interface (SMPI)",
209 " This option only has effect if this simulator is SMPI-based. Traces the MPI\n"
210 " interface and generates a trace that can be analyzed using Gantt-like\n"
211 " visualizations. Every MPI function (implemented by SMPI) is transformed in a\n"
212 " state, and point-to-point communications can be analyzed with arrows.");
213 print_line(OPT_TRACING_DISABLE_DESTROY, "Disable platform containers destruction",
214 " Disable the destruction of containers at the end of simulation. This can be\n"
215 " used with simulators that have a different notion of time (different from\n"
216 " the simulated time).");
217 print_line(OPT_TRACING_BASIC, "Avoid extended events (impoverished trace file).",
218 " Some visualization tools are not able to parse correctly the Paje file format.\n"
219 " Use this option if you are using one of these tools to visualize the simulation\n"
220 " trace. Keep in mind that the trace might be incomplete, without all the\n"
221 " information that would be registered otherwise.");
222 print_line(OPT_TRACING_FORMAT_TI_ONEFILE, "Only works for SMPI now, and TI output format",
223 " By default, each process outputs to a separate file, inside a filename_files folder\n"
224 " By setting this option to yes, all processes will output to only one file\n"
225 " This is meant to avoid opening thousands of files with large simulations");
226 print_line(OPT_TRACING_TOPOLOGY, "Register the platform topology as a graph",
227 " This option (enabled by default) can be used to disable the tracing of\n"
228 " the platform topology in the trace file. Sometimes, such task is really\n"
229 " time consuming, since it must get the route from each host to other hosts\n"
230 " within the same Autonomous System (AS).");
236 static bool trace_active = false;
240 xbt::signal<void(Container&)> Container::on_creation;
241 xbt::signal<void(Container&)> Container::on_destruction;
243 static void on_container_creation_paje(Container& c)
245 double timestamp = SIMIX_get_clock();
246 std::stringstream stream;
248 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_CreateContainer, timestamp);
250 stream << std::fixed << std::setprecision(TRACE_precision()) << PAJE_CreateContainer << " ";
251 stream << timestamp << " " << c.get_id() << " " << c.type_->get_id() << " " << c.father_->get_id() << " \"";
252 if (c.get_name().find("rank-") != 0)
253 stream << c.get_name() << "\"";
255 /* Subtract -1 because this is the process id and we transform it to the rank id */
256 stream << "rank-" << stoi(c.get_name().substr(5)) - 1 << "\"";
258 XBT_DEBUG("Dump %s", stream.str().c_str());
259 tracing_file << stream.str() << std::endl;
262 static void on_container_destruction_paje(Container& c)
264 // obligation to dump previous events because they might reference the container that is about to be destroyed
265 TRACE_last_timestamp_to_dump = SIMIX_get_clock();
266 TRACE_paje_dump_buffer(true);
268 // trace my destruction, but not if user requests so or if the container is root
269 if (not TRACE_disable_destroy() && &c != Container::get_root()) {
270 std::stringstream stream;
271 double timestamp = SIMIX_get_clock();
273 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_DestroyContainer, timestamp);
275 stream << std::fixed << std::setprecision(TRACE_precision()) << PAJE_DestroyContainer << " ";
276 stream << timestamp << " " << c.type_->get_id() << " " << c.get_id();
277 XBT_DEBUG("Dump %s", stream.str().c_str());
278 tracing_file << stream.str() << std::endl;
282 static void on_container_creation_ti(Container& c)
284 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_CreateContainer, SIMIX_get_clock());
285 // if we are in the mode with only one file
286 static std::ofstream* ti_unique_file = nullptr;
288 if (tracing_files.empty()) {
289 // generate unique run id with time
290 prefix = xbt_os_time();
293 if (not simgrid::config::get_value<bool>("tracing/smpi/format/ti-one-file") || ti_unique_file == nullptr) {
294 std::string folder_name = simgrid::config::get_value<std::string>("tracing/filename") + "_files";
295 std::string filename = folder_name + "/" + std::to_string(prefix) + "_" + c.get_name() + ".txt";
297 _mkdir(folder_name.c_str());
299 mkdir(folder_name.c_str(), S_IRWXU | S_IRWXG | S_IRWXO);
301 ti_unique_file = new std::ofstream(filename.c_str(), std::ofstream::out);
302 xbt_assert(not ti_unique_file->fail(), "Tracefile %s could not be opened for writing", filename.c_str());
303 tracing_file << filename << std::endl;
305 tracing_files.insert({&c, ti_unique_file});
308 static void on_container_destruction_ti(Container& c)
310 // obligation to dump previous events because they might reference the container that is about to be destroyed
311 TRACE_last_timestamp_to_dump = SIMIX_get_clock();
312 TRACE_paje_dump_buffer(true);
314 if (not TRACE_disable_destroy() && &c != Container::get_root()) {
315 XBT_DEBUG("%s: event_type=%u, timestamp=%f", __func__, PAJE_DestroyContainer, SIMIX_get_clock());
316 if (not simgrid::config::get_value<bool>("tracing/smpi/format/ti-one-file") || tracing_files.size() == 1) {
317 tracing_files.at(&c)->close();
318 delete tracing_files.at(&c);
320 tracing_files.erase(&c);
324 static void on_simulation_start()
329 // tracing system must be:
330 // - enabled (with --cfg=tracing:yes)
331 // - already configured (simgrid::instr::init already called)
332 if (TRACE_is_enabled()) {
335 XBT_DEBUG("Tracing starts");
337 /* init the tracing module to generate the right output */
338 std::string format = config::get_value<std::string>("tracing/smpi/format");
339 XBT_DEBUG("Tracing format %s", format.c_str());
341 /* Connect the callbacks associated to the creation/destruction of containers*/
342 if (format == "Paje") {
343 Container::on_creation.connect(on_container_creation_paje);
344 Container::on_destruction.connect(on_container_destruction_paje);
346 Container::on_creation.connect(on_container_creation_ti);
347 Container::on_destruction.connect(on_container_destruction_ti);
350 /* open the trace file(s) */
351 std::string filename = TRACE_get_filename();
352 tracing_file.open(filename.c_str(), std::ofstream::out);
353 if (tracing_file.fail()) {
354 throw TracingError(XBT_THROW_POINT,
355 xbt::string_printf("Tracefile %s could not be opened for writing.", filename.c_str()));
358 XBT_DEBUG("Filename %s is open for writing", filename.c_str());
360 if (format == "Paje") {
361 /* output generator version */
362 tracing_file << "#This file was generated using SimGrid-" << SIMGRID_VERSION_MAJOR << "." << SIMGRID_VERSION_MINOR
363 << "." << SIMGRID_VERSION_PATCH << std::endl;
364 tracing_file << "#[";
365 for (auto str : simgrid::xbt::cmdline) {
366 tracing_file << str << " ";
368 tracing_file << "]" << std::endl;
371 /* output one line comment */
372 std::string comment = simgrid::config::get_value<std::string>("tracing/comment");
373 if (not comment.empty())
374 tracing_file << "# " << comment << std::endl;
376 /* output comment file */
377 dump_comment_file(simgrid::config::get_value<std::string>(OPT_TRACING_COMMENT_FILE));
379 if (format == "Paje") {
380 /* output Pajé header */
381 TRACE_header(TRACE_basic(), TRACE_display_sizes());
383 trace_format = TraceFormat::Ti;
386 XBT_DEBUG("Tracing is on");
390 static void on_simulation_end()
392 if (not trace_active)
395 /* dump trace buffer */
396 TRACE_last_timestamp_to_dump = surf_get_clock();
397 TRACE_paje_dump_buffer(true);
399 const Type* root_type = Container::get_root()->type_;
400 /* destroy all data structures of tracing (and free) */
401 delete Container::get_root();
404 /* close the trace files */
405 tracing_file.close();
406 XBT_DEBUG("Filename %s is closed", TRACE_get_filename().c_str());
408 /* de-activate trace */
409 trace_active = false;
410 XBT_DEBUG("Tracing is off");
411 XBT_DEBUG("Tracing system is shutdown");
416 static bool is_initialized = false;
420 is_initialized = true;
422 /* name of the tracefile */
423 config::declare_flag<std::string>("tracing/filename", "Trace file created by the instrumented SimGrid.",
425 config::declare_flag<std::string>("tracing/smpi/format",
426 "Select trace output format used by SMPI. The default is the 'Paje' format. "
427 "The 'TI' (Time-Independent) format allows for trace replay.",
430 config::declare_flag<bool>(OPT_TRACING_FORMAT_TI_ONEFILE,
431 "(smpi only) For replay format only : output to one file only", false);
432 config::alias(OPT_TRACING_FORMAT_TI_ONEFILE, {"tracing/smpi/format/ti_one_file"});
433 config::declare_flag<std::string>("tracing/comment", "Add a comment line to the top of the trace file.", "");
434 config::declare_flag<std::string>(OPT_TRACING_COMMENT_FILE,
435 "Add the contents of a file as comments to the top of the trace.", "");
436 config::alias(OPT_TRACING_COMMENT_FILE, {"tracing/comment_file"});
437 config::declare_flag<int>("tracing/precision",
438 "Numerical precision used when timestamping events "
439 "(expressed in number of digits after decimal point)",
442 /* Connect callbacks */
443 s4u::Engine::on_platform_creation.connect(on_simulation_start);
444 s4u::Engine::on_deadlock.connect(on_simulation_end);
445 s4u::Engine::on_simulation_end.connect(on_simulation_end);
448 } // namespace simgrid