src/instr/instr_platform.cpp

   1 /* Copyright (c) 2010-2021. 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   xbt_assert(father, "common father unknown, this is a tracing problem");
  79
  80   // check if we already register this pair (we only need one direction)
  81   std::string aux1 = src->get_name() + dst->get_name();
  82   std::string aux2 = dst->get_name() + src->get_name();
  83   if (filter->find(aux1) != filter->end()) {
  84     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
  85     return;
  86   }
  87   if (filter->find(aux2) != filter->end()) {
  88     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
  89     return;
  90   }
  91
  92   // ok, not found, register it
  93   filter->insert(aux1);
  94   filter->insert(aux2);
  95
  96   // declare type
  97   std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
  98                               std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
  99                               std::to_string(dst->type_->get_id());
 100   simgrid::instr::LinkType* link = father->type_->by_name_or_create(link_typename, src->type_, dst->type_);
 101   link->set_calling_container(father);
 102
 103   // create the link
 104   static long long counter = 0;
 105
 106   std::string key = std::to_string(counter);
 107   counter++;
 108
 109   link->start_event(src, "topology", key);
 110   link->end_event(dst, "topology", key);
 111
 112   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
 113 }
 114
 115 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, simgrid::instr::Container* container,
 116                                      std::set<std::string, std::less<>>* filter)
 117 {
 118   if (not TRACE_platform_topology()) {
 119     XBT_DEBUG("Graph extraction disabled by user.");
 120     return;
 121   }
 122   XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
 123   if (not netzone->get_children().empty()) {
 124     // bottom-up recursion
 125     for (auto const& nz_son : netzone->get_children()) {
 126       simgrid::instr::Container* child_container = container->children_.at(nz_son->get_name());
 127       recursiveGraphExtraction(nz_son, child_container, filter);
 128     }
 129   }
 130
 131   auto* graph = xbt_graph_new_graph(0, nullptr);
 132   std::map<std::string, xbt_node_t, std::less<>> nodes;
 133   std::map<std::string, xbt_edge_t, std::less<>> edges;
 134
 135   netzone->get_impl()->get_graph(graph, &nodes, &edges);
 136   for (auto const& elm : edges) {
 137     const xbt_edge* edge = elm.second;
 138     linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
 139                    simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
 140   }
 141   xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
 142 }
 143
 144 /*
 145  * user categories support
 146  */
 147 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
 148                                      simgrid::instr::Type* root)
 149 {
 150   if (root->get_name() == "HOST" || root->get_name() == "VM")
 151     root->by_name_or_create(std::string("p") + new_typename, color);
 152
 153   if (root->get_name() == "LINK")
 154     root->by_name_or_create(std::string("b") + new_typename, color);
 155
 156   for (auto const& elm : root->get_children()) {
 157     recursiveNewVariableType(new_typename, color, elm.second.get());
 158   }
 159 }
 160
 161 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
 162 {
 163   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
 164 }
 165
 166 static void recursiveNewUserVariableType(const std::string& father_type, const std::string& new_typename,
 167                                          const std::string& color, simgrid::instr::Type* root)
 168 {
 169   if (root->get_name() == father_type) {
 170     root->by_name_or_create(new_typename, color);
 171   }
 172   for (auto const& elm : root->get_children())
 173     recursiveNewUserVariableType(father_type, new_typename, color, elm.second.get());
 174 }
 175
 176 void instr_new_user_variable_type(const std::string& father_type, const std::string& new_typename,
 177                                   const std::string& color)
 178 {
 179   recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
 180 }
 181
 182 static void recursiveNewUserStateType(const std::string& father_type, const std::string& new_typename,
 183                                       simgrid::instr::Type* root)
 184 {
 185   if (root->get_name() == father_type)
 186     root->by_name_or_create<simgrid::instr::StateType>(new_typename);
 187
 188   for (auto const& elm : root->get_children())
 189     recursiveNewUserStateType(father_type, new_typename, elm.second.get());
 190 }
 191
 192 void instr_new_user_state_type(const std::string& father_type, const std::string& new_typename)
 193 {
 194   recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
 195 }
 196
 197 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
 198                                               simgrid::instr::Type* root)
 199 {
 200   if (root->get_name() == type_name)
 201     static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
 202
 203   for (auto const& elm : root->get_children())
 204     recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
 205 }
 206
 207 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
 208 {
 209   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
 210 }
 211
 212 namespace simgrid {
 213 namespace instr {
 214
 215 void platform_graph_export_graphviz(const std::string& output_filename)
 216 {
 217   auto* g     = xbt_graph_new_graph(0, nullptr);
 218   std::map<std::string, xbt_node_t, std::less<>> nodes;
 219   std::map<std::string, xbt_edge_t, std::less<>> edges;
 220   s4u::Engine::get_instance()->get_netzone_root()->extract_xbt_graph(g, &nodes, &edges);
 221
 222   std::ofstream fs;
 223   fs.open(output_filename, std::ofstream::out);
 224   xbt_assert(not fs.fail(), "Failed to open %s", output_filename.c_str());
 225
 226   if (g->directed)
 227     fs << "digraph test {" << std::endl;
 228   else
 229     fs << "graph test {" << std::endl;
 230
 231   fs << "  graph [overlap=scale]" << std::endl;
 232
 233   fs << "  node [shape=box, style=filled]" << std::endl;
 234   fs << "  node [width=.3, height=.3, style=filled, color=skyblue]" << std::endl << std::endl;
 235
 236   for (auto const& elm : nodes)
 237     fs << "  \"" << elm.first << "\";" << std::endl;
 238
 239   for (auto const& elm : edges) {
 240     const char* src_s = static_cast<char*>(elm.second->src->data);
 241     const char* dst_s = static_cast<char*>(elm.second->dst->data);
 242     if (g->directed)
 243       fs << "  \"" << src_s << "\" -> \"" << dst_s << "\";" << std::endl;
 244     else
 245       fs << "  \"" << src_s << "\" -- \"" << dst_s << "\";" << std::endl;
 246   }
 247   fs << "}" << std::endl;
 248   fs.close();
 249
 250   xbt_graph_free_graph(g, xbt_free_f, xbt_free_f, nullptr);
 251 }
 252
 253 /* Callbacks */
 254 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
 255 static void on_netzone_creation(s4u::NetZone const& netzone)
 256 {
 257   std::string id = netzone.get_name();
 258   if (Container::get_root() == nullptr) {
 259     auto* root = new NetZoneContainer(id, 0, nullptr);
 260     xbt_assert(Container::get_root() == root);
 261
 262     if (TRACE_smpi_is_enabled()) {
 263       auto* mpi = root->type_->by_name_or_create<ContainerType>("MPI");
 264       if (not TRACE_smpi_is_grouped())
 265         mpi->by_name_or_create<StateType>("MPI_STATE");
 266       root->type_->by_name_or_create("MPI_LINK", mpi, mpi);
 267       root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
 268       mpi->by_name_or_create<StateType>("MIGRATE_STATE");
 269     }
 270
 271     if (TRACE_needs_platform()) {
 272       currentContainer.push_back(root);
 273     }
 274     return;
 275   }
 276
 277   if (TRACE_needs_platform()) {
 278     auto level      = static_cast<unsigned>(currentContainer.size());
 279     auto* container = new NetZoneContainer(id, level, currentContainer.back());
 280     currentContainer.push_back(container);
 281   }
 282 }
 283
 284 static void on_link_creation(s4u::Link const& link)
 285 {
 286   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
 287     return;
 288
 289   auto* container = new Container(link.get_name(), "LINK", currentContainer.back());
 290
 291   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
 292     VariableType* bandwidth = container->type_->by_name_or_create("bandwidth", "");
 293     bandwidth->set_calling_container(container);
 294     bandwidth->set_event(0, link.get_bandwidth());
 295     VariableType* latency = container->type_->by_name_or_create("latency", "");
 296     latency->set_calling_container(container);
 297     latency->set_event(0, link.get_latency());
 298   }
 299
 300   if (TRACE_uncategorized()) {
 301     container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
 302   }
 303 }
 304
 305 static void on_host_creation(s4u::Host const& host)
 306 {
 307   Container* container  = new HostContainer(host, currentContainer.back());
 308   const Container* root = Container::get_root();
 309
 310   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
 311     VariableType* speed = container->type_->by_name_or_create("speed", "");
 312     speed->set_calling_container(container);
 313     speed->set_event(0, host.get_speed());
 314
 315     VariableType* cores = container->type_->by_name_or_create("core_count", "");
 316     cores->set_calling_container(container);
 317     cores->set_event(0, host.get_core_count());
 318   }
 319
 320   if (TRACE_uncategorized())
 321     container->type_->by_name_or_create("speed_used", "0.5 0.5 0.5");
 322
 323   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
 324     auto* mpi = container->type_->by_name_or_create<ContainerType>("MPI");
 325     mpi->by_name_or_create<StateType>("MPI_STATE");
 326     root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
 327     mpi->by_name_or_create<StateType>("MIGRATE_STATE");
 328   }
 329 }
 330
 331 static void on_action_state_change(kernel::resource::Action const& action,
 332                                    kernel::resource::Action::State /* previous */)
 333 {
 334   auto n = static_cast<unsigned>(action.get_variable()->get_number_of_constraint());
 335
 336   for (unsigned i = 0; i < n; i++) {
 337     double value = action.get_variable()->get_value() * action.get_variable()->get_constraint_weight(i);
 338     /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
 339     kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
 340     const kernel::resource::Cpu* cpu     = dynamic_cast<kernel::resource::Cpu*>(resource);
 341
 342     if (cpu != nullptr)
 343       resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
 344                                action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
 345
 346     const kernel::resource::LinkImpl* link = dynamic_cast<kernel::resource::LinkImpl*>(resource);
 347
 348     if (link != nullptr)
 349       resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
 350                                action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
 351   }
 352 }
 353
 354 static void on_platform_created()
 355 {
 356   currentContainer.clear();
 357   std::set<std::string, std::less<>> filter;
 358   XBT_DEBUG("Starting graph extraction.");
 359   recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), &filter);
 360   XBT_DEBUG("Graph extraction finished.");
 361   dump_buffer(true);
 362 }
 363
 364 static void on_actor_creation(s4u::Actor const& actor)
 365 {
 366   const Container* root      = Container::get_root();
 367   Container* container       = Container::by_name(actor.get_host()->get_name());
 368   std::string container_name = instr_pid(actor);
 369
 370   container->create_child(container_name, "ACTOR");
 371   auto* actor_type = container->type_->by_name_or_create<ContainerType>("ACTOR");
 372   auto* state      = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
 373   state->add_entity_value("suspend", "1 0 1");
 374   state->add_entity_value("sleep", "1 1 0");
 375   state->add_entity_value("receive", "1 0 0");
 376   state->add_entity_value("send", "0 0 1");
 377   state->add_entity_value("execute", "0 1 1");
 378   root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
 379
 380   actor.on_exit([container_name](bool failed) {
 381     if (failed)
 382       // kill means that this actor no longer exists, let's destroy it
 383       Container::by_name(container_name)->remove_from_parent();
 384   });
 385 }
 386
 387 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
 388 {
 389   static long long int counter = 0;
 390   Container* container         = Container::by_name(instr_pid(actor));
 391   LinkType* link               = Container::get_root()->get_link("ACTOR_LINK");
 392
 393   // start link
 394   link->start_event(container, "M", std::to_string(counter));
 395   // destroy existing container of this process
 396   container->remove_from_parent();
 397   // create new container on the new_host location
 398   Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
 399   // end link
 400   link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
 401   counter++;
 402 }
 403
 404 static void on_vm_creation(s4u::Host const& host)
 405 {
 406   const Container* container = new HostContainer(host, currentContainer.back());
 407   const Container* root      = Container::get_root();
 408   auto* vm                   = container->type_->by_name_or_create<ContainerType>("VM");
 409   auto* state                = vm->by_name_or_create<StateType>("VM_STATE");
 410   state->add_entity_value("suspend", "1 0 1");
 411   state->add_entity_value("sleep", "1 1 0");
 412   state->add_entity_value("receive", "1 0 0");
 413   state->add_entity_value("send", "0 0 1");
 414   state->add_entity_value("execute", "0 1 1");
 415   root->type_->by_name_or_create("VM_LINK", vm, vm);
 416   root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
 417 }
 418
 419 void define_callbacks()
 420 {
 421   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
 422   // properly
 423   if (TRACE_needs_platform()) {
 424     s4u::Engine::on_platform_created.connect(on_platform_created);
 425     s4u::Host::on_creation.connect(on_host_creation);
 426     s4u::Host::on_speed_change.connect([](s4u::Host const& host) {
 427       Container::by_name(host.get_name())
 428           ->get_variable("speed")
 429           ->set_event(surf_get_clock(), host.get_core_count() * host.get_available_speed());
 430     });
 431     s4u::Link::on_creation.connect(on_link_creation);
 432     s4u::Link::on_bandwidth_change.connect([](s4u::Link const& link) {
 433       Container::by_name(link.get_name())
 434           ->get_variable("bandwidth")
 435           ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
 436     });
 437     s4u::NetZone::on_seal.connect([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
 438     kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
 439       if (netpoint.is_router())
 440         new RouterContainer(netpoint.get_name(), currentContainer.back());
 441     });
 442   }
 443
 444   s4u::NetZone::on_creation.connect(on_netzone_creation);
 445
 446   kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
 447   s4u::Link::on_communication_state_change.connect(on_action_state_change);
 448
 449   if (TRACE_actor_is_enabled()) {
 450     s4u::Actor::on_creation.connect(on_actor_creation);
 451     s4u::Actor::on_destruction.connect([](s4u::Actor const& actor) {
 452       auto container = Container::by_name_or_null(instr_pid(actor));
 453       if (container != nullptr)
 454         container->remove_from_parent();
 455     });
 456     s4u::Actor::on_suspend.connect([](s4u::Actor const& actor) {
 457       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
 458     });
 459     s4u::Actor::on_resume.connect(
 460         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 461     s4u::Actor::on_sleep.connect([](s4u::Actor const& actor) {
 462       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
 463     });
 464     s4u::Actor::on_wake_up.connect(
 465         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 466     s4u::Exec::on_start.connect([](s4u::Exec const&) {
 467       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("execute");
 468     });
 469     s4u::Exec::on_completion.connect([](s4u::Exec const&) {
 470       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->pop_event();
 471     });
 472     s4u::Comm::on_start.connect([](s4u::Comm const&, bool is_sender) {
 473       Container::by_name(instr_pid(*s4u::Actor::self()))
 474           ->get_state("ACTOR_STATE")
 475           ->push_event(is_sender ? "send" : "receive");
 476     });
 477     s4u::Comm::on_completion.connect([](s4u::Comm const&) {
 478       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->pop_event();
 479     });
 480     s4u::Actor::on_host_change.connect(on_actor_host_change);
 481   }
 482
 483   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing()) {
 484     s4u::Exec::on_start.connect([](s4u::Exec const& exec) {
 485       Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
 486           ->get_state("MPI_STATE")
 487           ->push_event("computing", new CpuTIData("compute", exec.get_cost()));
 488     });
 489     s4u::Exec::on_completion.connect([](s4u::Exec const&) {
 490       Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
 491           ->get_state("MPI_STATE")
 492           ->pop_event();
 493     });
 494   }
 495
 496   if (TRACE_vm_is_enabled()) {
 497     s4u::Host::on_creation.connect(on_vm_creation);
 498     s4u::VirtualMachine::on_start.connect([](s4u::VirtualMachine const& vm) {
 499       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
 500     });
 501     s4u::VirtualMachine::on_started.connect(
 502         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
 503     s4u::VirtualMachine::on_suspend.connect([](s4u::VirtualMachine const& vm) {
 504       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
 505     });
 506     s4u::VirtualMachine::on_resume.connect(
 507         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
 508     s4u::Host::on_destruction.connect(
 509         [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
 510   }
 511 }
 512 } // namespace instr
 513 } // namespace simgrid