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 container_t lowestCommonAncestor(const simgrid::instr::Container* a1, const simgrid::instr::Container* a2)
  33 {
  34   // this is only an optimization (since most of a1 and a2 share the same parent)
  35   if (a1->father_ == a2->father_)
  36     return a1->father_;
  37
  38   // create an array with all ancestors of a1
  39   std::vector<container_t> ancestors_a1;
  40   container_t p = a1->father_;
  41   while (p) {
  42     ancestors_a1.push_back(p);
  43     p = p->father_;
  44   }
  45
  46   // create an array with all ancestors of a2
  47   std::vector<container_t> ancestors_a2;
  48   p = a2->father_;
  49   while (p) {
  50     ancestors_a2.push_back(p);
  51     p = p->father_;
  52   }
  53
  54   // find the lowest ancestor
  55   p     = nullptr;
  56   int i = static_cast<int>(ancestors_a1.size()) - 1;
  57   int j = static_cast<int>(ancestors_a2.size()) - 1;
  58   while (i >= 0 && j >= 0) {
  59     container_t a1p = ancestors_a1.at(i);
  60     const simgrid::instr::Container* a2p = ancestors_a2.at(j);
  61     if (a1p == a2p) {
  62       p = a1p;
  63     } else {
  64       break;
  65     }
  66     i--;
  67     j--;
  68   }
  69   return p;
  70 }
  71
  72 static void linkContainers(container_t src, container_t dst, std::set<std::string>* filter)
  73 {
  74   // ignore loopback
  75   if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
  76     XBT_DEBUG("  linkContainers: ignoring loopback link");
  77     return;
  78   }
  79
  80   // find common father
  81   container_t father = lowestCommonAncestor(src, dst);
  82   if (not father) {
  83     xbt_die("common father unknown, this is a tracing problem");
  84   }
  85
  86   // check if we already register this pair (we only need one direction)
  87   std::string aux1 = src->get_name() + dst->get_name();
  88   std::string aux2 = dst->get_name() + src->get_name();
  89   if (filter->find(aux1) != filter->end()) {
  90     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
  91     return;
  92   }
  93   if (filter->find(aux2) != filter->end()) {
  94     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
  95     return;
  96   }
  97
  98   // ok, not found, register it
  99   filter->insert(aux1);
 100   filter->insert(aux2);
 101
 102   // declare type
 103   std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
 104                               std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
 105                               std::to_string(dst->type_->get_id());
 106   simgrid::instr::LinkType* link = father->type_->by_name_or_create(link_typename, src->type_, dst->type_);
 107   link->set_calling_container(father);
 108
 109   // create the link
 110   static long long counter = 0;
 111
 112   std::string key = std::to_string(counter);
 113   counter++;
 114
 115   link->start_event(src, "topology", key);
 116   link->end_event(dst, "topology", key);
 117
 118   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
 119 }
 120
 121 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, container_t container,
 122                                      std::set<std::string>* filter)
 123 {
 124   if (not TRACE_platform_topology()) {
 125     XBT_DEBUG("Graph extraction disabled by user.");
 126     return;
 127   }
 128   XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
 129   if (not netzone->get_children().empty()) {
 130     // bottom-up recursion
 131     for (auto const& nz_son : netzone->get_children()) {
 132       container_t child_container = container->children_.at(nz_son->get_name());
 133       recursiveGraphExtraction(nz_son, child_container, filter);
 134     }
 135   }
 136
 137   auto* graph = xbt_graph_new_graph(0, nullptr);
 138   auto* nodes = new std::map<std::string, xbt_node_t>();
 139   auto* edges = new std::map<std::string, xbt_edge_t>();
 140
 141   netzone->get_impl()->get_graph(graph, nodes, edges);
 142   for (auto elm : *edges) {
 143     const xbt_edge* edge = elm.second;
 144     linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
 145                    simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
 146   }
 147   delete nodes;
 148   delete edges;
 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   auto* nodes = new std::map<std::string, xbt_node_t>();
 227   auto* edges = new std::map<std::string, xbt_edge_t>();
 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   delete nodes;
 260   delete edges;
 261 }
 262
 263 /* Callbacks */
 264 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
 265 static void on_netzone_creation(s4u::NetZone const& netzone)
 266 {
 267   std::string id = netzone.get_name();
 268   if (Container::get_root() == nullptr) {
 269     auto* root = new NetZoneContainer(id, 0, nullptr);
 270     xbt_assert(Container::get_root() == root);
 271
 272     if (TRACE_smpi_is_enabled()) {
 273       ContainerType* mpi = root->type_->by_name_or_create<ContainerType>("MPI");
 274       if (not TRACE_smpi_is_grouped())
 275         mpi->by_name_or_create<StateType>("MPI_STATE");
 276       root->type_->by_name_or_create("MPI_LINK", mpi, mpi);
 277       // TODO See if we can move this to the LoadBalancer plugin
 278       root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
 279       mpi->by_name_or_create<StateType>("MIGRATE_STATE");
 280     }
 281
 282     if (TRACE_needs_platform()) {
 283       currentContainer.push_back(root);
 284     }
 285     return;
 286   }
 287
 288   if (TRACE_needs_platform()) {
 289     auto level      = static_cast<unsigned>(currentContainer.size());
 290     auto* container = new NetZoneContainer(id, level, currentContainer.back());
 291     currentContainer.push_back(container);
 292   }
 293 }
 294
 295 static void on_link_creation(s4u::Link const& link)
 296 {
 297   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
 298     return;
 299
 300   auto* container = new Container(link.get_name(), "LINK", currentContainer.back());
 301
 302   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
 303     VariableType* bandwidth = container->type_->by_name_or_create("bandwidth", "");
 304     bandwidth->set_calling_container(container);
 305     bandwidth->set_event(0, link.get_bandwidth());
 306     VariableType* latency = container->type_->by_name_or_create("latency", "");
 307     latency->set_calling_container(container);
 308     latency->set_event(0, link.get_latency());
 309   }
 310
 311   if (TRACE_uncategorized()) {
 312     container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
 313   }
 314 }
 315
 316 static void on_host_creation(s4u::Host const& host)
 317 {
 318   Container* container  = new HostContainer(host, currentContainer.back());
 319   const Container* root = Container::get_root();
 320
 321   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
 322     VariableType* speed = container->type_->by_name_or_create("speed", "");
 323     speed->set_calling_container(container);
 324     speed->set_event(0, host.get_speed());
 325
 326     VariableType* cores = container->type_->by_name_or_create("core_count", "");
 327     cores->set_calling_container(container);
 328     cores->set_event(0, host.get_core_count());
 329   }
 330
 331   if (TRACE_uncategorized())
 332     container->type_->by_name_or_create("speed_used", "0.5 0.5 0.5");
 333
 334   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
 335     ContainerType* mpi = container->type_->by_name_or_create<ContainerType>("MPI");
 336     mpi->by_name_or_create<StateType>("MPI_STATE");
 337     // TODO See if we can move this to the LoadBalancer plugin
 338     root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
 339     mpi->by_name_or_create<StateType>("MIGRATE_STATE");
 340   }
 341 }
 342
 343 static void on_action_state_change(kernel::resource::Action const& action,
 344                                    kernel::resource::Action::State /* previous */)
 345 {
 346   auto n = static_cast<unsigned>(action.get_variable()->get_number_of_constraint());
 347
 348   for (unsigned i = 0; i < n; i++) {
 349     double value = action.get_variable()->get_value() * action.get_variable()->get_constraint_weight(i);
 350     /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
 351     kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
 352     const kernel::resource::Cpu* cpu     = dynamic_cast<kernel::resource::Cpu*>(resource);
 353
 354     if (cpu != nullptr)
 355       resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
 356                                action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
 357
 358     const kernel::resource::LinkImpl* link = dynamic_cast<kernel::resource::LinkImpl*>(resource);
 359
 360     if (link != nullptr)
 361       resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
 362                                action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
 363   }
 364 }
 365
 366 static void on_platform_created()
 367 {
 368   currentContainer.clear();
 369   auto* filter = new std::set<std::string>();
 370   XBT_DEBUG("Starting graph extraction.");
 371   recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), filter);
 372   XBT_DEBUG("Graph extraction finished.");
 373   delete filter;
 374   dump_buffer(true);
 375 }
 376
 377 static void on_actor_creation(s4u::Actor const& actor)
 378 {
 379   const Container* root      = Container::get_root();
 380   Container* container       = Container::by_name(actor.get_host()->get_name());
 381   std::string container_name = instr_pid(actor);
 382
 383   container->create_child(container_name, "ACTOR");
 384   ContainerType* actor_type = container->type_->by_name_or_create<ContainerType>("ACTOR");
 385   StateType* state          = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
 386   state->add_entity_value("suspend", "1 0 1");
 387   state->add_entity_value("sleep", "1 1 0");
 388   state->add_entity_value("receive", "1 0 0");
 389   state->add_entity_value("send", "0 0 1");
 390   state->add_entity_value("execute", "0 1 1");
 391   root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
 392
 393   actor.on_exit([container_name](bool failed) {
 394     if (failed)
 395       // kill means that this actor no longer exists, let's destroy it
 396       Container::by_name(container_name)->remove_from_parent();
 397   });
 398 }
 399
 400 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
 401 {
 402   static long long int counter = 0;
 403   container_t container        = Container::by_name(instr_pid(actor));
 404   LinkType* link               = Container::get_root()->get_link("ACTOR_LINK");
 405
 406   // start link
 407   link->start_event(container, "M", std::to_string(counter));
 408   // destroy existing container of this process
 409   container->remove_from_parent();
 410   // create new container on the new_host location
 411   Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
 412   // end link
 413   link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
 414   counter++;
 415 }
 416
 417 static void on_vm_creation(s4u::Host const& host)
 418 {
 419   const Container* container = new HostContainer(host, currentContainer.back());
 420   const Container* root      = Container::get_root();
 421   ContainerType* vm          = container->type_->by_name_or_create<ContainerType>("VM");
 422   StateType* state           = vm->by_name_or_create<StateType>("VM_STATE");
 423   state->add_entity_value("suspend", "1 0 1");
 424   state->add_entity_value("sleep", "1 1 0");
 425   state->add_entity_value("receive", "1 0 0");
 426   state->add_entity_value("send", "0 0 1");
 427   state->add_entity_value("execute", "0 1 1");
 428   root->type_->by_name_or_create("VM_LINK", vm, vm);
 429   root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
 430 }
 431
 432 void define_callbacks()
 433 {
 434   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
 435   // properly
 436   if (TRACE_needs_platform()) {
 437     s4u::Engine::on_platform_created.connect(on_platform_created);
 438     s4u::Host::on_creation.connect(on_host_creation);
 439     s4u::Host::on_speed_change.connect([](s4u::Host const& host) {
 440       Container::by_name(host.get_name())
 441           ->get_variable("speed")
 442           ->set_event(surf_get_clock(), host.get_core_count() * host.get_available_speed());
 443     });
 444     s4u::Link::on_creation.connect(on_link_creation);
 445     s4u::Link::on_bandwidth_change.connect([](s4u::Link const& link) {
 446       Container::by_name(link.get_name())
 447           ->get_variable("bandwidth")
 448           ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
 449     });
 450     s4u::NetZone::on_seal.connect([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
 451     kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
 452       if (netpoint.is_router())
 453         new RouterContainer(netpoint.get_name(), currentContainer.back());
 454     });
 455   }
 456
 457   s4u::NetZone::on_creation.connect(on_netzone_creation);
 458
 459   kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
 460   s4u::Link::on_communication_state_change.connect(on_action_state_change);
 461
 462   if (TRACE_actor_is_enabled()) {
 463     s4u::Actor::on_creation.connect(on_actor_creation);
 464     s4u::Actor::on_destruction.connect([](s4u::Actor const& actor) {
 465       auto container = Container::by_name_or_null(instr_pid(actor));
 466       if (container != nullptr)
 467         container->remove_from_parent();
 468     });
 469     s4u::Actor::on_suspend.connect([](s4u::Actor const& actor) {
 470       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
 471     });
 472     s4u::Actor::on_resume.connect(
 473         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 474     s4u::Actor::on_sleep.connect([](s4u::Actor const& actor) {
 475       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
 476     });
 477     s4u::Actor::on_wake_up.connect(
 478         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 479     s4u::Exec::on_start.connect([](simgrid::s4u::Actor const& actor, s4u::Exec const&) {
 480       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("execute");
 481     });
 482     s4u::Exec::on_completion.connect([](s4u::Actor const& actor, s4u::Exec const&) {
 483       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
 484     });
 485     s4u::Comm::on_sender_start.connect([](s4u::Actor const& actor) {
 486       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("send");
 487     });
 488     s4u::Comm::on_receiver_start.connect([](s4u::Actor const& actor) {
 489       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("receive");
 490     });
 491     s4u::Comm::on_completion.connect(
 492         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 493     s4u::Actor::on_host_change.connect(on_actor_host_change);
 494   }
 495
 496   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing()) {
 497     s4u::Exec::on_start.connect([](simgrid::s4u::Actor const& actor, s4u::Exec const& exec) {
 498       Container::by_name(std::string("rank-") + std::to_string(actor.get_pid()))
 499           ->get_state("MPI_STATE")
 500           ->push_event("computing", new CpuTIData("compute", exec.get_cost()));
 501     });
 502     s4u::Exec::on_completion.connect([](s4u::Actor const& actor, s4u::Exec const&) {
 503       Container::by_name(std::string("rank-") + std::to_string(actor.get_pid()))->get_state("MPI_STATE")->pop_event();
 504     });
 505   }
 506
 507   if (TRACE_vm_is_enabled()) {
 508     s4u::Host::on_creation.connect(on_vm_creation);
 509     s4u::VirtualMachine::on_start.connect([](s4u::VirtualMachine const& vm) {
 510       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
 511     });
 512     s4u::VirtualMachine::on_started.connect(
 513         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
 514     s4u::VirtualMachine::on_suspend.connect([](s4u::VirtualMachine const& vm) {
 515       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
 516     });
 517     s4u::VirtualMachine::on_resume.connect(
 518         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
 519     s4u::Host::on_destruction.connect(
 520         [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
 521   }
 522 }
 523 } // namespace instr
 524 } // namespace simgrid