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   simgrid::instr::Container* p = a1->father_;
  42   while (p) {
  43     ancestors_a1.push_back(p);
  44     p = p->father_;
  45   }
  46
  47   // create an array with all ancestors of a2
  48   std::vector<simgrid::instr::Container*> ancestors_a2;
  49   p = a2->father_;
  50   while (p) {
  51     ancestors_a2.push_back(p);
  52     p = p->father_;
  53   }
  54
  55   // find the lowest ancestor
  56   p     = nullptr;
  57   int i = static_cast<int>(ancestors_a1.size()) - 1;
  58   int j = static_cast<int>(ancestors_a2.size()) - 1;
  59   while (i >= 0 && j >= 0) {
  60     simgrid::instr::Container* a1p       = ancestors_a1.at(i);
  61     const simgrid::instr::Container* a2p = ancestors_a2.at(j);
  62     if (a1p == a2p) {
  63       p = a1p;
  64     } else {
  65       break;
  66     }
  67     i--;
  68     j--;
  69   }
  70   return p;
  71 }
  72
  73 static void linkContainers(simgrid::instr::Container* src, simgrid::instr::Container* dst,
  74                            std::set<std::string>* filter)
  75 {
  76   // ignore loopback
  77   if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
  78     XBT_DEBUG("  linkContainers: ignoring loopback link");
  79     return;
  80   }
  81
  82   // find common father
  83   simgrid::instr::Container* father = lowestCommonAncestor(src, dst);
  84   if (not father) {
  85     xbt_die("common father unknown, this is a tracing problem");
  86   }
  87
  88   // check if we already register this pair (we only need one direction)
  89   std::string aux1 = src->get_name() + dst->get_name();
  90   std::string aux2 = dst->get_name() + src->get_name();
  91   if (filter->find(aux1) != filter->end()) {
  92     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
  93     return;
  94   }
  95   if (filter->find(aux2) != filter->end()) {
  96     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
  97     return;
  98   }
  99
 100   // ok, not found, register it
 101   filter->insert(aux1);
 102   filter->insert(aux2);
 103
 104   // declare type
 105   std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
 106                               std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
 107                               std::to_string(dst->type_->get_id());
 108   simgrid::instr::LinkType* link = father->type_->by_name_or_create(link_typename, src->type_, dst->type_);
 109   link->set_calling_container(father);
 110
 111   // create the link
 112   static long long counter = 0;
 113
 114   std::string key = std::to_string(counter);
 115   counter++;
 116
 117   link->start_event(src, "topology", key);
 118   link->end_event(dst, "topology", key);
 119
 120   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
 121 }
 122
 123 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, simgrid::instr::Container* container,
 124                                      std::set<std::string>* filter)
 125 {
 126   if (not TRACE_platform_topology()) {
 127     XBT_DEBUG("Graph extraction disabled by user.");
 128     return;
 129   }
 130   XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
 131   if (not netzone->get_children().empty()) {
 132     // bottom-up recursion
 133     for (auto const& nz_son : netzone->get_children()) {
 134       simgrid::instr::Container* child_container = container->children_.at(nz_son->get_name());
 135       recursiveGraphExtraction(nz_son, child_container, filter);
 136     }
 137   }
 138
 139   auto* graph = xbt_graph_new_graph(0, nullptr);
 140   std::map<std::string, xbt_node_t> nodes;
 141   std::map<std::string, xbt_edge_t> edges;
 142
 143   netzone->get_impl()->get_graph(graph, &nodes, &edges);
 144   for (auto elm : edges) {
 145     const xbt_edge* edge = elm.second;
 146     linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
 147                    simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
 148   }
 149   xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
 150 }
 151
 152 /*
 153  * user categories support
 154  */
 155 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
 156                                      simgrid::instr::Type* root)
 157 {
 158   if (root->get_name() == "HOST" || root->get_name() == "VM")
 159     root->by_name_or_create(std::string("p") + new_typename, color);
 160
 161   if (root->get_name() == "LINK")
 162     root->by_name_or_create(std::string("b") + new_typename, color);
 163
 164   for (auto const& elm : root->get_children()) {
 165     recursiveNewVariableType(new_typename, color, elm.second.get());
 166   }
 167 }
 168
 169 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
 170 {
 171   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
 172 }
 173
 174 static void recursiveNewUserVariableType(const std::string& father_type, const std::string& new_typename,
 175                                          const std::string& color, simgrid::instr::Type* root)
 176 {
 177   if (root->get_name() == father_type) {
 178     root->by_name_or_create(new_typename, color);
 179   }
 180   for (auto const& elm : root->get_children())
 181     recursiveNewUserVariableType(father_type, new_typename, color, elm.second.get());
 182 }
 183
 184 void instr_new_user_variable_type(const std::string& father_type, const std::string& new_typename,
 185                                   const std::string& color)
 186 {
 187   recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
 188 }
 189
 190 static void recursiveNewUserStateType(const std::string& father_type, const std::string& new_typename,
 191                                       simgrid::instr::Type* root)
 192 {
 193   if (root->get_name() == father_type)
 194     root->by_name_or_create<simgrid::instr::StateType>(new_typename);
 195
 196   for (auto const& elm : root->get_children())
 197     recursiveNewUserStateType(father_type, new_typename, elm.second.get());
 198 }
 199
 200 void instr_new_user_state_type(const std::string& father_type, const std::string& new_typename)
 201 {
 202   recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
 203 }
 204
 205 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
 206                                               simgrid::instr::Type* root)
 207 {
 208   if (root->get_name() == type_name)
 209     static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
 210
 211   for (auto const& elm : root->get_children())
 212     recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
 213 }
 214
 215 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
 216 {
 217   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
 218 }
 219
 220 namespace simgrid {
 221 namespace instr {
 222
 223 void platform_graph_export_graphviz(const std::string& output_filename)
 224 {
 225   auto* g     = xbt_graph_new_graph(0, nullptr);
 226   std::map<std::string, xbt_node_t> nodes;
 227   std::map<std::string, xbt_edge_t> edges;
 228   s4u::Engine::get_instance()->get_netzone_root()->extract_xbt_graph(g, &nodes, &edges);
 229
 230   std::ofstream fs;
 231   fs.open(output_filename, std::ofstream::out);
 232   xbt_assert(not fs.fail(), "Failed to open %s", output_filename.c_str());
 233
 234   if (g->directed)
 235     fs << "digraph test {" << std::endl;
 236   else
 237     fs << "graph test {" << std::endl;
 238
 239   fs << "  graph [overlap=scale]" << std::endl;
 240
 241   fs << "  node [shape=box, style=filled]" << std::endl;
 242   fs << "  node [width=.3, height=.3, style=filled, color=skyblue]" << std::endl << std::endl;
 243
 244   for (auto const& elm : nodes)
 245     fs << "  \"" << elm.first << "\";" << std::endl;
 246
 247   for (auto const& elm : edges) {
 248     const char* src_s = static_cast<char*>(elm.second->src->data);
 249     const char* dst_s = static_cast<char*>(elm.second->dst->data);
 250     if (g->directed)
 251       fs << "  \"" << src_s << "\" -> \"" << dst_s << "\";" << std::endl;
 252     else
 253       fs << "  \"" << src_s << "\" -- \"" << dst_s << "\";" << std::endl;
 254   }
 255   fs << "}" << std::endl;
 256   fs.close();
 257
 258   xbt_graph_free_graph(g, xbt_free_f, xbt_free_f, nullptr);
 259 }
 260
 261 /* Callbacks */
 262 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
 263 static void on_netzone_creation(s4u::NetZone const& netzone)
 264 {
 265   std::string id = netzone.get_name();
 266   if (Container::get_root() == nullptr) {
 267     auto* root = new NetZoneContainer(id, 0, nullptr);
 268     xbt_assert(Container::get_root() == root);
 269
 270     if (TRACE_smpi_is_enabled()) {
 271       auto* mpi = root->type_->by_name_or_create<ContainerType>("MPI");
 272       if (not TRACE_smpi_is_grouped())
 273         mpi->by_name_or_create<StateType>("MPI_STATE");
 274       root->type_->by_name_or_create("MPI_LINK", mpi, mpi);
 275       // TODO See if we can move this to the LoadBalancer plugin
 276       root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
 277       mpi->by_name_or_create<StateType>("MIGRATE_STATE");
 278     }
 279
 280     if (TRACE_needs_platform()) {
 281       currentContainer.push_back(root);
 282     }
 283     return;
 284   }
 285
 286   if (TRACE_needs_platform()) {
 287     auto level      = static_cast<unsigned>(currentContainer.size());
 288     auto* container = new NetZoneContainer(id, level, currentContainer.back());
 289     currentContainer.push_back(container);
 290   }
 291 }
 292
 293 static void on_link_creation(s4u::Link const& link)
 294 {
 295   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
 296     return;
 297
 298   auto* container = new Container(link.get_name(), "LINK", currentContainer.back());
 299
 300   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
 301     VariableType* bandwidth = container->type_->by_name_or_create("bandwidth", "");
 302     bandwidth->set_calling_container(container);
 303     bandwidth->set_event(0, link.get_bandwidth());
 304     VariableType* latency = container->type_->by_name_or_create("latency", "");
 305     latency->set_calling_container(container);
 306     latency->set_event(0, link.get_latency());
 307   }
 308
 309   if (TRACE_uncategorized()) {
 310     container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
 311   }
 312 }
 313
 314 static void on_host_creation(s4u::Host const& host)
 315 {
 316   Container* container  = new HostContainer(host, currentContainer.back());
 317   const Container* root = Container::get_root();
 318
 319   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
 320     VariableType* speed = container->type_->by_name_or_create("speed", "");
 321     speed->set_calling_container(container);
 322     speed->set_event(0, host.get_speed());
 323
 324     VariableType* cores = container->type_->by_name_or_create("core_count", "");
 325     cores->set_calling_container(container);
 326     cores->set_event(0, host.get_core_count());
 327   }
 328
 329   if (TRACE_uncategorized())
 330     container->type_->by_name_or_create("speed_used", "0.5 0.5 0.5");
 331
 332   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
 333     auto* mpi = container->type_->by_name_or_create<ContainerType>("MPI");
 334     mpi->by_name_or_create<StateType>("MPI_STATE");
 335     // TODO See if we can move this to the LoadBalancer plugin
 336     root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
 337     mpi->by_name_or_create<StateType>("MIGRATE_STATE");
 338   }
 339 }
 340
 341 static void on_action_state_change(kernel::resource::Action const& action,
 342                                    kernel::resource::Action::State /* previous */)
 343 {
 344   auto n = static_cast<unsigned>(action.get_variable()->get_number_of_constraint());
 345
 346   for (unsigned i = 0; i < n; i++) {
 347     double value = action.get_variable()->get_value() * action.get_variable()->get_constraint_weight(i);
 348     /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
 349     kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
 350     const kernel::resource::Cpu* cpu     = dynamic_cast<kernel::resource::Cpu*>(resource);
 351
 352     if (cpu != nullptr)
 353       resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
 354                                action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
 355
 356     const kernel::resource::LinkImpl* link = dynamic_cast<kernel::resource::LinkImpl*>(resource);
 357
 358     if (link != nullptr)
 359       resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
 360                                action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
 361   }
 362 }
 363
 364 static void on_platform_created()
 365 {
 366   currentContainer.clear();
 367   std::set<std::string> filter;
 368   XBT_DEBUG("Starting graph extraction.");
 369   recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), &filter);
 370   XBT_DEBUG("Graph extraction finished.");
 371   dump_buffer(true);
 372 }
 373
 374 static void on_actor_creation(s4u::Actor const& actor)
 375 {
 376   const Container* root      = Container::get_root();
 377   Container* container       = Container::by_name(actor.get_host()->get_name());
 378   std::string container_name = instr_pid(actor);
 379
 380   container->create_child(container_name, "ACTOR");
 381   auto* actor_type = container->type_->by_name_or_create<ContainerType>("ACTOR");
 382   auto* state      = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
 383   state->add_entity_value("suspend", "1 0 1");
 384   state->add_entity_value("sleep", "1 1 0");
 385   state->add_entity_value("receive", "1 0 0");
 386   state->add_entity_value("send", "0 0 1");
 387   state->add_entity_value("execute", "0 1 1");
 388   root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
 389
 390   actor.on_exit([container_name](bool failed) {
 391     if (failed)
 392       // kill means that this actor no longer exists, let's destroy it
 393       Container::by_name(container_name)->remove_from_parent();
 394   });
 395 }
 396
 397 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
 398 {
 399   static long long int counter = 0;
 400   Container* container         = Container::by_name(instr_pid(actor));
 401   LinkType* link               = Container::get_root()->get_link("ACTOR_LINK");
 402
 403   // start link
 404   link->start_event(container, "M", std::to_string(counter));
 405   // destroy existing container of this process
 406   container->remove_from_parent();
 407   // create new container on the new_host location
 408   Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
 409   // end link
 410   link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
 411   counter++;
 412 }
 413
 414 static void on_vm_creation(s4u::Host const& host)
 415 {
 416   const Container* container = new HostContainer(host, currentContainer.back());
 417   const Container* root      = Container::get_root();
 418   auto* vm                   = container->type_->by_name_or_create<ContainerType>("VM");
 419   auto* state                = vm->by_name_or_create<StateType>("VM_STATE");
 420   state->add_entity_value("suspend", "1 0 1");
 421   state->add_entity_value("sleep", "1 1 0");
 422   state->add_entity_value("receive", "1 0 0");
 423   state->add_entity_value("send", "0 0 1");
 424   state->add_entity_value("execute", "0 1 1");
 425   root->type_->by_name_or_create("VM_LINK", vm, vm);
 426   root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
 427 }
 428
 429 void define_callbacks()
 430 {
 431   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
 432   // properly
 433   if (TRACE_needs_platform()) {
 434     s4u::Engine::on_platform_created.connect(on_platform_created);
 435     s4u::Host::on_creation.connect(on_host_creation);
 436     s4u::Host::on_speed_change.connect([](s4u::Host const& host) {
 437       Container::by_name(host.get_name())
 438           ->get_variable("speed")
 439           ->set_event(surf_get_clock(), host.get_core_count() * host.get_available_speed());
 440     });
 441     s4u::Link::on_creation.connect(on_link_creation);
 442     s4u::Link::on_bandwidth_change.connect([](s4u::Link const& link) {
 443       Container::by_name(link.get_name())
 444           ->get_variable("bandwidth")
 445           ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
 446     });
 447     s4u::NetZone::on_seal.connect([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
 448     kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
 449       if (netpoint.is_router())
 450         new RouterContainer(netpoint.get_name(), currentContainer.back());
 451     });
 452   }
 453
 454   s4u::NetZone::on_creation.connect(on_netzone_creation);
 455
 456   kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
 457   s4u::Link::on_communication_state_change.connect(on_action_state_change);
 458
 459   if (TRACE_actor_is_enabled()) {
 460     s4u::Actor::on_creation.connect(on_actor_creation);
 461     s4u::Actor::on_destruction.connect([](s4u::Actor const& actor) {
 462       auto container = Container::by_name_or_null(instr_pid(actor));
 463       if (container != nullptr)
 464         container->remove_from_parent();
 465     });
 466     s4u::Actor::on_suspend.connect([](s4u::Actor const& actor) {
 467       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
 468     });
 469     s4u::Actor::on_resume.connect(
 470         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 471     s4u::Actor::on_sleep.connect([](s4u::Actor const& actor) {
 472       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
 473     });
 474     s4u::Actor::on_wake_up.connect(
 475         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 476     s4u::Exec::on_start.connect([](simgrid::s4u::Actor const& actor, s4u::Exec const&) {
 477       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("execute");
 478     });
 479     s4u::Exec::on_completion.connect([](s4u::Actor const& actor, s4u::Exec const&) {
 480       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
 481     });
 482     s4u::Comm::on_sender_start.connect([](s4u::Actor const& actor) {
 483       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("send");
 484     });
 485     s4u::Comm::on_receiver_start.connect([](s4u::Actor const& actor) {
 486       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("receive");
 487     });
 488     s4u::Comm::on_completion.connect(
 489         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 490     s4u::Actor::on_host_change.connect(on_actor_host_change);
 491   }
 492
 493   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing()) {
 494     s4u::Exec::on_start.connect([](simgrid::s4u::Actor const& actor, s4u::Exec const& exec) {
 495       Container::by_name(std::string("rank-") + std::to_string(actor.get_pid()))
 496           ->get_state("MPI_STATE")
 497           ->push_event("computing", new CpuTIData("compute", exec.get_cost()));
 498     });
 499     s4u::Exec::on_completion.connect([](s4u::Actor const& actor, s4u::Exec const&) {
 500       Container::by_name(std::string("rank-") + std::to_string(actor.get_pid()))->get_state("MPI_STATE")->pop_event();
 501     });
 502   }
 503
 504   if (TRACE_vm_is_enabled()) {
 505     s4u::Host::on_creation.connect(on_vm_creation);
 506     s4u::VirtualMachine::on_start.connect([](s4u::VirtualMachine const& vm) {
 507       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
 508     });
 509     s4u::VirtualMachine::on_started.connect(
 510         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
 511     s4u::VirtualMachine::on_suspend.connect([](s4u::VirtualMachine const& vm) {
 512       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
 513     });
 514     s4u::VirtualMachine::on_resume.connect(
 515         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
 516     s4u::Host::on_destruction.connect(
 517         [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
 518   }
 519 }
 520 } // namespace instr
 521 } // namespace simgrid