src/instr/instr_platform.cpp

   1 /* Copyright (c) 2010-2020. The SimGrid Team. All rights reserved.          */
   2
   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. */
   5
   6 #include "src/instr/instr_private.hpp"
   7
   8 #include "simgrid/kernel/routing/NetPoint.hpp"
   9 #include "simgrid/kernel/routing/NetZoneImpl.hpp"
  10 #include "simgrid/s4u/Actor.hpp"
  11 #include "simgrid/s4u/Comm.hpp"
  12 #include "simgrid/s4u/Engine.hpp"
  13 #include "simgrid/s4u/Exec.hpp"
  14 #include "simgrid/s4u/Host.hpp"
  15 #include "simgrid/s4u/VirtualMachine.hpp"
  16 #include "src/surf/cpu_interface.hpp"
  17 #include "src/surf/network_interface.hpp"
  18 #include "src/surf/surf_interface.hpp"
  19 #include "src/surf/xml/platf_private.hpp"
  20 #include "surf/surf.hpp"
  21 #include "xbt/graph.h"
  22
  23 #include <fstream>
  24
  25 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
  26
  27 std::string instr_pid(simgrid::s4u::Actor const& proc)
  28 {
  29   return std::string(proc.get_name()) + "-" + std::to_string(proc.get_pid());
  30 }
  31
  32 static simgrid::instr::Container* lowestCommonAncestor(const simgrid::instr::Container* a1,
  33                                                        const simgrid::instr::Container* a2)
  34 {
  35   // this is only an optimization (since most of a1 and a2 share the same parent)
  36   if (a1->father_ == a2->father_)
  37     return a1->father_;
  38
  39   // create an array with all ancestors of a1
  40   std::vector<simgrid::instr::Container*> ancestors_a1;
  41   for (auto* p = a1->father_; p != nullptr; p = p->father_)
  42     ancestors_a1.push_back(p);
  43
  44   // create an array with all ancestors of a2
  45   std::vector<simgrid::instr::Container*> ancestors_a2;
  46   for (auto* p = a2->father_; p != nullptr; p = p->father_)
  47     ancestors_a2.push_back(p);
  48
  49   // find the lowest ancestor
  50   simgrid::instr::Container* p = nullptr;
  51   int i = static_cast<int>(ancestors_a1.size()) - 1;
  52   int j = static_cast<int>(ancestors_a2.size()) - 1;
  53   while (i >= 0 && j >= 0) {
  54     simgrid::instr::Container* a1p       = ancestors_a1.at(i);
  55     const simgrid::instr::Container* a2p = ancestors_a2.at(j);
  56     if (a1p == a2p) {
  57       p = a1p;
  58     } else {
  59       break;
  60     }
  61     i--;
  62     j--;
  63   }
  64   return p;
  65 }
  66
  67 static void linkContainers(simgrid::instr::Container* src, simgrid::instr::Container* dst,
  68                            std::set<std::string, std::less<>>* filter)
  69 {
  70   // ignore loopback
  71   if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
  72     XBT_DEBUG("  linkContainers: ignoring loopback link");
  73     return;
  74   }
  75
  76   // find common father
  77   simgrid::instr::Container* father = lowestCommonAncestor(src, dst);
  78   if (not father) {
  79     xbt_die("common father unknown, this is a tracing problem");
  80   }
  81
  82   // check if we already register this pair (we only need one direction)
  83   std::string aux1 = src->get_name() + dst->get_name();
  84   std::string aux2 = dst->get_name() + src->get_name();
  85   if (filter->find(aux1) != filter->end()) {
  86     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
  87     return;
  88   }
  89   if (filter->find(aux2) != filter->end()) {
  90     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
  91     return;
  92   }
  93
  94   // ok, not found, register it
  95   filter->insert(aux1);
  96   filter->insert(aux2);
  97
  98   // declare type
  99   std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
 100                               std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
 101                               std::to_string(dst->type_->get_id());
 102   simgrid::instr::LinkType* link = father->type_->by_name_or_create(link_typename, src->type_, dst->type_);
 103   link->set_calling_container(father);
 104
 105   // create the link
 106   static long long counter = 0;
 107
 108   std::string key = std::to_string(counter);
 109   counter++;
 110
 111   link->start_event(src, "topology", key);
 112   link->end_event(dst, "topology", key);
 113
 114   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
 115 }
 116
 117 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, simgrid::instr::Container* container,
 118                                      std::set<std::string, std::less<>>* filter)
 119 {
 120   if (not TRACE_platform_topology()) {
 121     XBT_DEBUG("Graph extraction disabled by user.");
 122     return;
 123   }
 124   XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
 125   if (not netzone->get_children().empty()) {
 126     // bottom-up recursion
 127     for (auto const& nz_son : netzone->get_children()) {
 128       simgrid::instr::Container* child_container = container->children_.at(nz_son->get_name());
 129       recursiveGraphExtraction(nz_son, child_container, filter);
 130     }
 131   }
 132
 133   auto* graph = xbt_graph_new_graph(0, nullptr);
 134   std::map<std::string, xbt_node_t, std::less<>> nodes;
 135   std::map<std::string, xbt_edge_t, std::less<>> edges;
 136
 137   netzone->get_impl()->get_graph(graph, &nodes, &edges);
 138   for (auto const& elm : edges) {
 139     const xbt_edge* edge = elm.second;
 140     linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
 141                    simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
 142   }
 143   xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
 144 }
 145
 146 /*
 147  * user categories support
 148  */
 149 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
 150                                      simgrid::instr::Type* root)
 151 {
 152   if (root->get_name() == "HOST" || root->get_name() == "VM")
 153     root->by_name_or_create(std::string("p") + new_typename, color);
 154
 155   if (root->get_name() == "LINK")
 156     root->by_name_or_create(std::string("b") + new_typename, color);
 157
 158   for (auto const& elm : root->get_children()) {
 159     recursiveNewVariableType(new_typename, color, elm.second.get());
 160   }
 161 }
 162
 163 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
 164 {
 165   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
 166 }
 167
 168 static void recursiveNewUserVariableType(const std::string& father_type, const std::string& new_typename,
 169                                          const std::string& color, simgrid::instr::Type* root)
 170 {
 171   if (root->get_name() == father_type) {
 172     root->by_name_or_create(new_typename, color);
 173   }
 174   for (auto const& elm : root->get_children())
 175     recursiveNewUserVariableType(father_type, new_typename, color, elm.second.get());
 176 }
 177
 178 void instr_new_user_variable_type(const std::string& father_type, const std::string& new_typename,
 179                                   const std::string& color)
 180 {
 181   recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
 182 }
 183
 184 static void recursiveNewUserStateType(const std::string& father_type, const std::string& new_typename,
 185                                       simgrid::instr::Type* root)
 186 {
 187   if (root->get_name() == father_type)
 188     root->by_name_or_create<simgrid::instr::StateType>(new_typename);
 189
 190   for (auto const& elm : root->get_children())
 191     recursiveNewUserStateType(father_type, new_typename, elm.second.get());
 192 }
 193
 194 void instr_new_user_state_type(const std::string& father_type, const std::string& new_typename)
 195 {
 196   recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
 197 }
 198
 199 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
 200                                               simgrid::instr::Type* root)
 201 {
 202   if (root->get_name() == type_name)
 203     static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
 204
 205   for (auto const& elm : root->get_children())
 206     recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
 207 }
 208
 209 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
 210 {
 211   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
 212 }
 213
 214 namespace simgrid {
 215 namespace instr {
 216
 217 void platform_graph_export_graphviz(const std::string& output_filename)
 218 {
 219   auto* g     = xbt_graph_new_graph(0, nullptr);
 220   std::map<std::string, xbt_node_t, std::less<>> nodes;
 221   std::map<std::string, xbt_edge_t, std::less<>> edges;
 222   s4u::Engine::get_instance()->get_netzone_root()->extract_xbt_graph(g, &nodes, &edges);
 223
 224   std::ofstream fs;
 225   fs.open(output_filename, std::ofstream::out);
 226   xbt_assert(not fs.fail(), "Failed to open %s", output_filename.c_str());
 227
 228   if (g->directed)
 229     fs << "digraph test {" << std::endl;
 230   else
 231     fs << "graph test {" << std::endl;
 232
 233   fs << "  graph [overlap=scale]" << std::endl;
 234
 235   fs << "  node [shape=box, style=filled]" << std::endl;
 236   fs << "  node [width=.3, height=.3, style=filled, color=skyblue]" << std::endl << std::endl;
 237
 238   for (auto const& elm : nodes)
 239     fs << "  \"" << elm.first << "\";" << std::endl;
 240
 241   for (auto const& elm : edges) {
 242     const char* src_s = static_cast<char*>(elm.second->src->data);
 243     const char* dst_s = static_cast<char*>(elm.second->dst->data);
 244     if (g->directed)
 245       fs << "  \"" << src_s << "\" -> \"" << dst_s << "\";" << std::endl;
 246     else
 247       fs << "  \"" << src_s << "\" -- \"" << dst_s << "\";" << std::endl;
 248   }
 249   fs << "}" << std::endl;
 250   fs.close();
 251
 252   xbt_graph_free_graph(g, xbt_free_f, xbt_free_f, nullptr);
 253 }
 254
 255 /* Callbacks */
 256 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
 257 static void on_netzone_creation(s4u::NetZone const& netzone)
 258 {
 259   std::string id = netzone.get_name();
 260   if (Container::get_root() == nullptr) {
 261     auto* root = new NetZoneContainer(id, 0, nullptr);
 262     xbt_assert(Container::get_root() == root);
 263
 264     if (TRACE_smpi_is_enabled()) {
 265       auto* mpi = root->type_->by_name_or_create<ContainerType>("MPI");
 266       if (not TRACE_smpi_is_grouped())
 267         mpi->by_name_or_create<StateType>("MPI_STATE");
 268       root->type_->by_name_or_create("MPI_LINK", mpi, mpi);
 269       root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
 270       mpi->by_name_or_create<StateType>("MIGRATE_STATE");
 271     }
 272
 273     if (TRACE_needs_platform()) {
 274       currentContainer.push_back(root);
 275     }
 276     return;
 277   }
 278
 279   if (TRACE_needs_platform()) {
 280     auto level      = static_cast<unsigned>(currentContainer.size());
 281     auto* container = new NetZoneContainer(id, level, currentContainer.back());
 282     currentContainer.push_back(container);
 283   }
 284 }
 285
 286 static void on_link_creation(s4u::Link const& link)
 287 {
 288   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
 289     return;
 290
 291   auto* container = new Container(link.get_name(), "LINK", currentContainer.back());
 292
 293   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
 294     VariableType* bandwidth = container->type_->by_name_or_create("bandwidth", "");
 295     bandwidth->set_calling_container(container);
 296     bandwidth->set_event(0, link.get_bandwidth());
 297     VariableType* latency = container->type_->by_name_or_create("latency", "");
 298     latency->set_calling_container(container);
 299     latency->set_event(0, link.get_latency());
 300   }
 301
 302   if (TRACE_uncategorized()) {
 303     container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
 304   }
 305 }
 306
 307 static void on_host_creation(s4u::Host const& host)
 308 {
 309   Container* container  = new HostContainer(host, currentContainer.back());
 310   const Container* root = Container::get_root();
 311
 312   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
 313     VariableType* speed = container->type_->by_name_or_create("speed", "");
 314     speed->set_calling_container(container);
 315     speed->set_event(0, host.get_speed());
 316
 317     VariableType* cores = container->type_->by_name_or_create("core_count", "");
 318     cores->set_calling_container(container);
 319     cores->set_event(0, host.get_core_count());
 320   }
 321
 322   if (TRACE_uncategorized())
 323     container->type_->by_name_or_create("speed_used", "0.5 0.5 0.5");
 324
 325   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
 326     auto* mpi = container->type_->by_name_or_create<ContainerType>("MPI");
 327     mpi->by_name_or_create<StateType>("MPI_STATE");
 328     root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
 329     mpi->by_name_or_create<StateType>("MIGRATE_STATE");
 330   }
 331 }
 332
 333 static void on_action_state_change(kernel::resource::Action const& action,
 334                                    kernel::resource::Action::State /* previous */)
 335 {
 336   auto n = static_cast<unsigned>(action.get_variable()->get_number_of_constraint());
 337
 338   for (unsigned i = 0; i < n; i++) {
 339     double value = action.get_variable()->get_value() * action.get_variable()->get_constraint_weight(i);
 340     /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
 341     kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
 342     const kernel::resource::Cpu* cpu     = dynamic_cast<kernel::resource::Cpu*>(resource);
 343
 344     if (cpu != nullptr)
 345       resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
 346                                action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
 347
 348     const kernel::resource::LinkImpl* link = dynamic_cast<kernel::resource::LinkImpl*>(resource);
 349
 350     if (link != nullptr)
 351       resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
 352                                action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
 353   }
 354 }
 355
 356 static void on_platform_created()
 357 {
 358   currentContainer.clear();
 359   std::set<std::string, std::less<>> filter;
 360   XBT_DEBUG("Starting graph extraction.");
 361   recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), &filter);
 362   XBT_DEBUG("Graph extraction finished.");
 363   dump_buffer(true);
 364 }
 365
 366 static void on_actor_creation(s4u::Actor const& actor)
 367 {
 368   const Container* root      = Container::get_root();
 369   Container* container       = Container::by_name(actor.get_host()->get_name());
 370   std::string container_name = instr_pid(actor);
 371
 372   container->create_child(container_name, "ACTOR");
 373   auto* actor_type = container->type_->by_name_or_create<ContainerType>("ACTOR");
 374   auto* state      = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
 375   state->add_entity_value("suspend", "1 0 1");
 376   state->add_entity_value("sleep", "1 1 0");
 377   state->add_entity_value("receive", "1 0 0");
 378   state->add_entity_value("send", "0 0 1");
 379   state->add_entity_value("execute", "0 1 1");
 380   root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
 381
 382   actor.on_exit([container_name](bool failed) {
 383     if (failed)
 384       // kill means that this actor no longer exists, let's destroy it
 385       Container::by_name(container_name)->remove_from_parent();
 386   });
 387 }
 388
 389 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
 390 {
 391   static long long int counter = 0;
 392   Container* container         = Container::by_name(instr_pid(actor));
 393   LinkType* link               = Container::get_root()->get_link("ACTOR_LINK");
 394
 395   // start link
 396   link->start_event(container, "M", std::to_string(counter));
 397   // destroy existing container of this process
 398   container->remove_from_parent();
 399   // create new container on the new_host location
 400   Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
 401   // end link
 402   link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
 403   counter++;
 404 }
 405
 406 static void on_vm_creation(s4u::Host const& host)
 407 {
 408   const Container* container = new HostContainer(host, currentContainer.back());
 409   const Container* root      = Container::get_root();
 410   auto* vm                   = container->type_->by_name_or_create<ContainerType>("VM");
 411   auto* state                = vm->by_name_or_create<StateType>("VM_STATE");
 412   state->add_entity_value("suspend", "1 0 1");
 413   state->add_entity_value("sleep", "1 1 0");
 414   state->add_entity_value("receive", "1 0 0");
 415   state->add_entity_value("send", "0 0 1");
 416   state->add_entity_value("execute", "0 1 1");
 417   root->type_->by_name_or_create("VM_LINK", vm, vm);
 418   root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
 419 }
 420
 421 void define_callbacks()
 422 {
 423   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
 424   // properly
 425   if (TRACE_needs_platform()) {
 426     s4u::Engine::on_platform_created.connect(on_platform_created);
 427     s4u::Host::on_creation.connect(on_host_creation);
 428     s4u::Host::on_speed_change.connect([](s4u::Host const& host) {
 429       Container::by_name(host.get_name())
 430           ->get_variable("speed")
 431           ->set_event(surf_get_clock(), host.get_core_count() * host.get_available_speed());
 432     });
 433     s4u::Link::on_creation.connect(on_link_creation);
 434     s4u::Link::on_bandwidth_change.connect([](s4u::Link const& link) {
 435       Container::by_name(link.get_name())
 436           ->get_variable("bandwidth")
 437           ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
 438     });
 439     s4u::NetZone::on_seal.connect([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
 440     kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
 441       if (netpoint.is_router())
 442         new RouterContainer(netpoint.get_name(), currentContainer.back());
 443     });
 444   }
 445
 446   s4u::NetZone::on_creation.connect(on_netzone_creation);
 447
 448   kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
 449   s4u::Link::on_communication_state_change.connect(on_action_state_change);
 450
 451   if (TRACE_actor_is_enabled()) {
 452     s4u::Actor::on_creation.connect(on_actor_creation);
 453     s4u::Actor::on_destruction.connect([](s4u::Actor const& actor) {
 454       auto container = Container::by_name_or_null(instr_pid(actor));
 455       if (container != nullptr)
 456         container->remove_from_parent();
 457     });
 458     s4u::Actor::on_suspend.connect([](s4u::Actor const& actor) {
 459       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
 460     });
 461     s4u::Actor::on_resume.connect(
 462         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 463     s4u::Actor::on_sleep.connect([](s4u::Actor const& actor) {
 464       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
 465     });
 466     s4u::Actor::on_wake_up.connect(
 467         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 468     s4u::Exec::on_start.connect([](s4u::Exec const&, simgrid::s4u::Actor const& actor) {
 469       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("execute");
 470     });
 471     s4u::Exec::on_completion.connect([](s4u::Exec const&, s4u::Actor const& actor) {
 472       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
 473     });
 474     s4u::Comm::on_sender_start.connect([](s4u::Comm const&, s4u::Actor const& actor) {
 475       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("send");
 476     });
 477     s4u::Comm::on_receiver_start.connect([](s4u::Comm const&, s4u::Actor const& actor) {
 478       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("receive");
 479     });
 480     s4u::Comm::on_completion.connect([](s4u::Comm const&, s4u::Actor const& actor) {
 481       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
 482     });
 483     s4u::Actor::on_host_change.connect(on_actor_host_change);
 484   }
 485
 486   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing()) {
 487     s4u::Exec::on_start.connect([](s4u::Exec const& exec, simgrid::s4u::Actor const& actor) {
 488       Container::by_name(std::string("rank-") + std::to_string(actor.get_pid()))
 489           ->get_state("MPI_STATE")
 490           ->push_event("computing", new CpuTIData("compute", exec.get_cost()));
 491     });
 492     s4u::Exec::on_completion.connect([](s4u::Exec const&, s4u::Actor const& actor) {
 493       Container::by_name(std::string("rank-") + std::to_string(actor.get_pid()))->get_state("MPI_STATE")->pop_event();
 494     });
 495   }
 496
 497   if (TRACE_vm_is_enabled()) {
 498     s4u::Host::on_creation.connect(on_vm_creation);
 499     s4u::VirtualMachine::on_start.connect([](s4u::VirtualMachine const& vm) {
 500       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
 501     });
 502     s4u::VirtualMachine::on_started.connect(
 503         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
 504     s4u::VirtualMachine::on_suspend.connect([](s4u::VirtualMachine const& vm) {
 505       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
 506     });
 507     s4u::VirtualMachine::on_resume.connect(
 508         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
 509     s4u::Host::on_destruction.connect(
 510         [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
 511   }
 512 }
 513 } // namespace instr
 514 } // namespace simgrid