1 /* Copyright (c) 2010-2018. 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/s4u/Engine.hpp"
8 #include "src/instr/instr_private.hpp"
9 #include "surf/surf.hpp"
10 #include "xbt/virtu.h" /* sg_cmdline */
15 XBT_LOG_NEW_CATEGORY(instr, "Logging the behavior of the tracing system (used for Visualization/Analysis of simulations)");
16 XBT_LOG_NEW_DEFAULT_SUBCATEGORY (instr_config, instr, "Configuration");
18 std::ofstream tracing_file;
20 #define OPT_TRACING_BASIC "tracing/basic"
21 #define OPT_TRACING_COMMENT_FILE "tracing/comment-file"
22 #define OPT_TRACING_DISABLE_DESTROY "tracing/disable-destroy"
23 #define OPT_TRACING_FORMAT_TI_ONEFILE "tracing/smpi/format/ti-one-file"
24 #define OPT_TRACING_SMPI_COMPUTING "tracing/smpi/computing"
25 #define OPT_TRACING_SMPI_GROUP "tracing/smpi/group"
26 #define OPT_TRACING_SMPI_INTERNALS "tracing/smpi/internals"
27 #define OPT_TRACING_SMPI_SLEEPING "tracing/smpi/sleeping"
28 #define OPT_TRACING_SMPI "tracing/smpi"
29 #define OPT_TRACING_TOPOLOGY "tracing/platform/topology"
31 static simgrid::config::Flag<bool> trace_enabled{
32 "tracing", "Enable the tracing system. You have to enable this option to use other tracing options.", false};
34 static simgrid::config::Flag<bool> trace_actor_enabled{
35 "tracing/msg/process", // FIXME rename this flag
36 "Trace the behavior of all categorized actors, grouping them by host. "
37 "Can be used to track actor location if the simulator does actor migration.",
40 static simgrid::config::Flag<bool> trace_vm_enabled{"tracing/vm", "Trace the behavior of all virtual machines.", false};
42 static simgrid::config::Flag<bool> trace_platform{"tracing/platform",
43 "Register the platform in the trace as a hierarchy.", false};
45 static simgrid::config::Flag<bool> trace_platform_topology{
46 OPT_TRACING_TOPOLOGY, "Register the platform topology in the trace as a graph.", true};
47 static simgrid::config::Flag<bool> trace_smpi_enabled{OPT_TRACING_SMPI, "Tracing of the SMPI interface.", false};
48 static simgrid::config::Flag<bool> trace_smpi_grouped{OPT_TRACING_SMPI_GROUP, "Group MPI processes by host.", false};
50 static simgrid::config::Flag<bool> trace_smpi_computing{
51 OPT_TRACING_SMPI_COMPUTING, "Generate states for timing out of SMPI parts of the application", false};
53 static simgrid::config::Flag<bool> trace_smpi_sleeping{
54 OPT_TRACING_SMPI_SLEEPING, "Generate states for timing out of SMPI parts of the application", false};
56 static simgrid::config::Flag<bool> trace_view_internals{
57 "tracing/smpi/internals",
58 "Generate tracing events corresponding to point-to-point messages sent by SMPI collective communications", false};
60 static simgrid::config::Flag<bool> trace_categorized{
61 "tracing/categorized", "Trace categorized resource utilization of hosts and links.", false};
63 static simgrid::config::Flag<bool> trace_uncategorized{
64 "tracing/uncategorized",
65 "Trace uncategorized resource utilization of hosts and links. "
66 "To use if the simulator does not use tracing categories but resource utilization have to be traced.",
69 static simgrid::config::Flag<bool> trace_disable_destroy{
70 OPT_TRACING_DISABLE_DESTROY, {"tracing/disable_destroy"}, "Disable platform containers destruction.", false};
71 static simgrid::config::Flag<bool> trace_basic{OPT_TRACING_BASIC, "Avoid extended events (impoverished trace file).",
74 static simgrid::config::Flag<bool> trace_display_sizes{
75 "tracing/smpi/display-sizes",
76 "Add message size information (in bytes) to the to links and states (SMPI only). "
77 "For collectives, it usually corresponds to the total number of bytes sent by a process.",
80 static simgrid::config::Flag<bool> trace_disable_link{"tracing/disable_link",
81 "Do not trace link bandwidth and latency.", false};
82 static simgrid::config::Flag<bool> trace_disable_power{"tracing/disable_power", "Do not trace host power.", false};
84 static bool trace_active = false;
86 simgrid::instr::TraceFormat simgrid::instr::trace_format = simgrid::instr::TraceFormat::Paje;
88 static void TRACE_start()
93 // tracing system must be:
94 // - enabled (with --cfg=tracing:yes)
95 // - already configured (TRACE_global_init already called)
96 if (TRACE_is_enabled()) {
97 instr_define_callbacks();
99 XBT_DEBUG("Tracing starts");
100 /* init the tracing module to generate the right output */
101 std::string format = simgrid::config::get_value<std::string>("tracing/smpi/format");
102 XBT_DEBUG("Tracing format %s", format.c_str());
104 /* open the trace file(s) */
105 std::string filename = TRACE_get_filename();
106 tracing_file.open(filename.c_str(), std::ofstream::out);
107 if (tracing_file.fail()) {
108 THROWF(system_error, 1, "Tracefile %s could not be opened for writing.", filename.c_str());
111 XBT_DEBUG("Filename %s is open for writing", filename.c_str());
113 if (format == "Paje") {
114 /* output generator version */
115 tracing_file << "#This file was generated using SimGrid-" << SIMGRID_VERSION_MAJOR << "." << SIMGRID_VERSION_MINOR
116 << "." << SIMGRID_VERSION_PATCH << std::endl;
117 tracing_file << "#[";
120 xbt_dynar_foreach (xbt_cmdline, cpt, str) {
121 tracing_file << str << " ";
123 tracing_file << "]" << std::endl;
126 /* output one line comment */
127 dump_comment(simgrid::config::get_value<std::string>("tracing/comment"));
129 /* output comment file */
130 dump_comment_file(simgrid::config::get_value<std::string>(OPT_TRACING_COMMENT_FILE));
132 if (format == "Paje") {
133 /* output Pajé header */
134 TRACE_header(TRACE_basic(), TRACE_display_sizes());
136 simgrid::instr::trace_format = simgrid::instr::TraceFormat::Ti;
139 XBT_DEBUG("Tracing is on");
143 static void TRACE_end()
145 if (not trace_active)
148 /* dump trace buffer */
149 TRACE_last_timestamp_to_dump = surf_get_clock();
150 TRACE_paje_dump_buffer(true);
152 simgrid::instr::Type* root_type = simgrid::instr::Container::get_root()->type_;
153 /* destroy all data structures of tracing (and free) */
154 delete simgrid::instr::Container::get_root();
157 /* close the trace files */
158 tracing_file.close();
159 XBT_DEBUG("Filename %s is closed", TRACE_get_filename().c_str());
161 /* de-activate trace */
162 trace_active = false;
163 XBT_DEBUG("Tracing is off");
164 XBT_DEBUG("Tracing system is shutdown");
167 bool TRACE_needs_platform ()
169 return TRACE_actor_is_enabled() || TRACE_vm_is_enabled() || TRACE_categorized() || TRACE_uncategorized() ||
170 TRACE_platform() || (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped());
173 bool TRACE_is_enabled()
175 return trace_enabled;
178 bool TRACE_platform()
180 return trace_platform;
183 bool TRACE_platform_topology()
185 return trace_platform_topology;
188 bool TRACE_smpi_is_enabled()
190 return (trace_smpi_enabled || TRACE_smpi_is_grouped()) && TRACE_is_enabled();
193 bool TRACE_smpi_is_grouped()
195 return trace_smpi_grouped;
198 bool TRACE_smpi_is_computing()
200 return trace_smpi_computing;
203 bool TRACE_smpi_is_sleeping()
205 return trace_smpi_sleeping;
208 bool TRACE_smpi_view_internals()
210 return trace_view_internals;
213 bool TRACE_categorized ()
215 return trace_categorized;
218 bool TRACE_uncategorized ()
220 return trace_uncategorized;
223 bool TRACE_actor_is_enabled()
225 return trace_actor_enabled && trace_enabled;
228 bool TRACE_vm_is_enabled()
230 return trace_vm_enabled && trace_enabled;
233 bool TRACE_disable_link()
235 return trace_disable_link && trace_enabled;
238 bool TRACE_disable_speed()
240 return trace_disable_power && trace_enabled;
243 bool TRACE_disable_destroy ()
245 return trace_disable_destroy && trace_enabled;
250 return trace_basic && trace_enabled;
253 bool TRACE_display_sizes ()
255 return trace_display_sizes && trace_smpi_enabled && trace_enabled;
258 int TRACE_precision ()
260 return simgrid::config::get_value<int>("tracing/precision");
263 std::string TRACE_get_filename()
265 return simgrid::config::get_value<std::string>("tracing/filename");
268 void TRACE_global_init()
270 static bool is_initialised = false;
274 is_initialised = true;
276 /* name of the tracefile */
277 simgrid::config::declare_flag<std::string>("tracing/filename", "Trace file created by the instrumented SimGrid.",
279 simgrid::config::declare_flag<std::string>(
280 "tracing/smpi/format", "Select trace output format used by SMPI. The default is the 'Paje' format. "
281 "The 'TI' (Time-Independent) format allows for trace replay.",
284 simgrid::config::declare_flag<bool>(OPT_TRACING_FORMAT_TI_ONEFILE,
285 "(smpi only) For replay format only : output to one file only", false);
286 simgrid::config::alias(OPT_TRACING_FORMAT_TI_ONEFILE, {"tracing/smpi/format/ti_one_file"});
287 simgrid::config::declare_flag<std::string>("tracing/comment", "Add a comment line to the top of the trace file.", "");
288 simgrid::config::declare_flag<std::string>(OPT_TRACING_COMMENT_FILE,
289 "Add the contents of a file as comments to the top of the trace.", "");
290 simgrid::config::alias(OPT_TRACING_COMMENT_FILE, {"tracing/comment_file"});
291 simgrid::config::declare_flag<int>("tracing/precision", "Numerical precision used when timestamping events "
292 "(expressed in number of digits after decimal point)",
295 /* Connect callbacks */
296 simgrid::s4u::on_platform_creation.connect(TRACE_start);
297 simgrid::s4u::on_deadlock.connect(TRACE_end);
298 simgrid::s4u::on_simulation_end.connect(TRACE_end);
301 static void print_line(const char* option, const char* desc, const char* longdesc)
303 std::string str = std::string("--cfg=") + option + " ";
305 int len = str.size();
306 printf("%s%*.*s %s\n", str.c_str(), 30 - len, 30 - len, "", desc);
307 if (longdesc != nullptr) {
308 printf ("%s\n\n", longdesc);
314 printf("Description of the tracing options accepted by this simulator:\n\n");
315 print_line(OPT_TRACING_SMPI, "Trace the MPI Interface (SMPI)",
316 " This option only has effect if this simulator is SMPI-based. Traces the MPI\n"
317 " interface and generates a trace that can be analyzed using Gantt-like\n"
318 " visualizations. Every MPI function (implemented by SMPI) is transformed in a\n"
319 " state, and point-to-point communications can be analyzed with arrows.");
320 print_line(OPT_TRACING_SMPI_GROUP, "Group MPI processes by host (SMPI)",
321 " This option only has effect if this simulator is SMPI-based. The processes\n"
322 " are grouped by the hosts where they were executed.");
323 print_line(OPT_TRACING_SMPI_COMPUTING, "Generates a \" Computing \" State",
324 " This option aims at tracing computations in the application, outside SMPI\n"
325 " to allow further study of simulated or real computation time");
326 print_line(OPT_TRACING_SMPI_SLEEPING, "Generates a \" Sleeping \" State",
327 " This option aims at tracing sleeps in the application, outside SMPI\n"
328 " to allow further study of simulated or real sleep time");
329 print_line(OPT_TRACING_DISABLE_DESTROY, "Disable platform containers destruction",
330 " Disable the destruction of containers at the end of simulation. This can be\n"
331 " used with simulators that have a different notion of time (different from\n"
332 " the simulated time).");
333 print_line(OPT_TRACING_BASIC, "Avoid extended events (impoverished trace file).",
334 " Some visualization tools are not able to parse correctly the Paje file format.\n"
335 " Use this option if you are using one of these tools to visualize the simulation\n"
336 " trace. Keep in mind that the trace might be incomplete, without all the\n"
337 " information that would be registered otherwise.");
338 print_line(OPT_TRACING_FORMAT_TI_ONEFILE, "Only works for SMPI now, and TI output format",
339 " By default, each process outputs to a separate file, inside a filename_files folder\n"
340 " By setting this option to yes, all processes will output to only one file\n"
341 " This is meant to avoid opening thousands of files with large simulations");
342 print_line(OPT_TRACING_TOPOLOGY, "Register the platform topology as a graph",
343 " This option (enabled by default) can be used to disable the tracing of\n"
344 " the platform topology in the trace file. Sometimes, such task is really\n"
345 " time consuming, since it must get the route from each host to other hosts\n"
346 " within the same Autonomous System (AS).");