src/instr/instr_platform.cpp

   1 /* Copyright (c) 2010-2022. 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 <simgrid/kernel/routing/NetPoint.hpp>
   7 #include <simgrid/kernel/routing/NetZoneImpl.hpp>
   8 #include <simgrid/s4u/Actor.hpp>
   9 #include <simgrid/s4u/Comm.hpp>
  10 #include <simgrid/s4u/Engine.hpp>
  11 #include <simgrid/s4u/Exec.hpp>
  12 #include <simgrid/s4u/Host.hpp>
  13 #include <simgrid/s4u/VirtualMachine.hpp>
  14 #include <xbt/graph.h>
  15
  16 #include "src/instr/instr_private.hpp"
  17 #include "src/kernel/resource/CpuImpl.hpp"
  18 #include "src/kernel/resource/StandardLinkImpl.hpp"
  19 #include "src/surf/surf_interface.hpp"
  20
  21 #include <fstream>
  22
  23 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
  24
  25 std::string instr_pid(simgrid::s4u::Actor const& proc)
  26 {
  27   return std::string(proc.get_name()) + "-" + std::to_string(proc.get_pid());
  28 }
  29
  30 static simgrid::instr::Container* lowestCommonAncestor(const simgrid::instr::Container* a1,
  31                                                        const simgrid::instr::Container* a2)
  32 {
  33   // this is only an optimization (since most of a1 and a2 share the same parent)
  34   if (a1->get_parent() == a2->get_parent())
  35     return a1->get_parent();
  36
  37   // create an array with all ancestors of a1
  38   std::vector<simgrid::instr::Container*> ancestors_a1;
  39   for (auto* p = a1->get_parent(); p != nullptr; p = p->get_parent())
  40     ancestors_a1.push_back(p);
  41
  42   // create an array with all ancestors of a2
  43   std::vector<simgrid::instr::Container*> ancestors_a2;
  44   for (auto* p = a2->get_parent(); p != nullptr; p = p->get_parent())
  45     ancestors_a2.push_back(p);
  46
  47   // find the lowest ancestor
  48   simgrid::instr::Container* p = nullptr;
  49   int i = static_cast<int>(ancestors_a1.size()) - 1;
  50   int j = static_cast<int>(ancestors_a2.size()) - 1;
  51   while (i >= 0 && j >= 0) {
  52     simgrid::instr::Container* a1p       = ancestors_a1.at(i);
  53     const simgrid::instr::Container* a2p = ancestors_a2.at(j);
  54     if (a1p == a2p) {
  55       p = a1p;
  56     } else {
  57       break;
  58     }
  59     i--;
  60     j--;
  61   }
  62   return p;
  63 }
  64
  65 static void linkContainers(simgrid::instr::Container* src, simgrid::instr::Container* dst,
  66                            std::set<std::string, std::less<>>* filter)
  67 {
  68   // ignore loopback
  69   if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
  70     XBT_DEBUG("  linkContainers: ignoring loopback link");
  71     return;
  72   }
  73
  74   // find common parent
  75   simgrid::instr::Container* parent = lowestCommonAncestor(src, dst);
  76   xbt_assert(parent, "common parent unknown, this is a tracing problem");
  77
  78   // check if we already register this pair (we only need one direction)
  79   std::string aux1 = src->get_name() + dst->get_name();
  80   std::string aux2 = dst->get_name() + src->get_name();
  81   if (filter->find(aux1) != filter->end()) {
  82     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
  83     return;
  84   }
  85   if (filter->find(aux2) != filter->end()) {
  86     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
  87     return;
  88   }
  89
  90   // ok, not found, register it
  91   filter->insert(aux1);
  92   filter->insert(aux2);
  93
  94   // declare type
  95   std::string link_typename = parent->get_type()->get_name() + "-" + src->get_type()->get_name() +
  96                               std::to_string(src->get_type()->get_id()) + "-" + dst->get_type()->get_name() +
  97                               std::to_string(dst->get_type()->get_id());
  98   simgrid::instr::LinkType* link =
  99       parent->get_type()->by_name_or_create(link_typename, src->get_type(), dst->get_type());
 100   link->set_calling_container(parent);
 101
 102   // create the link
 103   static long long counter = 0;
 104
 105   std::string key = std::to_string(counter);
 106   counter++;
 107
 108   link->start_event(src, "topology", key);
 109   link->end_event(dst, "topology", key);
 110
 111   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
 112 }
 113
 114 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, const simgrid::instr::Container* container,
 115                                      std::set<std::string, std::less<>>* filter)
 116 {
 117   if (not TRACE_platform_topology()) {
 118     XBT_DEBUG("Graph extraction disabled by user.");
 119     return;
 120   }
 121   XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
 122
 123   // bottom-up recursion
 124   for (auto const& nz_son : netzone->get_children()) {
 125     const simgrid::instr::Container* child_container = container->get_child_by_name(nz_son->get_name());
 126     recursiveGraphExtraction(nz_son, child_container, filter);
 127   }
 128
 129   auto* graph = xbt_graph_new_graph(0, nullptr);
 130   std::map<std::string, xbt_node_t, std::less<>> nodes;
 131   std::map<std::string, xbt_edge_t, std::less<>> edges;
 132
 133   netzone->get_impl()->get_graph(graph, &nodes, &edges);
 134   for (auto const& elm : edges) {
 135     const xbt_edge* edge = elm.second;
 136     linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
 137                    simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
 138   }
 139   xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
 140 }
 141
 142 /*
 143  * user categories support
 144  */
 145 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
 146                                      simgrid::instr::Type* root)
 147 {
 148   if (root->get_name() == "HOST" || root->get_name() == "VM")
 149     root->by_name_or_create(std::string("p") + new_typename, color);
 150
 151   if (root->get_name() == "LINK")
 152     root->by_name_or_create(std::string("b") + new_typename, color);
 153
 154   for (auto const& elm : root->get_children()) {
 155     recursiveNewVariableType(new_typename, color, elm.second.get());
 156   }
 157 }
 158
 159 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
 160 {
 161   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->get_type());
 162 }
 163
 164 static void recursiveNewUserVariableType(const std::string& parent_type, const std::string& new_typename,
 165                                          const std::string& color, simgrid::instr::Type* root)
 166 {
 167   if (root->get_name() == parent_type) {
 168     root->by_name_or_create(new_typename, color);
 169   }
 170   for (auto const& elm : root->get_children())
 171     recursiveNewUserVariableType(parent_type, new_typename, color, elm.second.get());
 172 }
 173
 174 void instr_new_user_variable_type(const std::string& parent_type, const std::string& new_typename,
 175                                   const std::string& color)
 176 {
 177   recursiveNewUserVariableType(parent_type, new_typename, color, simgrid::instr::Container::get_root()->get_type());
 178 }
 179
 180 static void recursiveNewUserStateType(const std::string& parent_type, const std::string& new_typename,
 181                                       simgrid::instr::Type* root)
 182 {
 183   if (root->get_name() == parent_type)
 184     root->by_name_or_create<simgrid::instr::StateType>(new_typename);
 185
 186   for (auto const& elm : root->get_children())
 187     recursiveNewUserStateType(parent_type, new_typename, elm.second.get());
 188 }
 189
 190 void instr_new_user_state_type(const std::string& parent_type, const std::string& new_typename)
 191 {
 192   recursiveNewUserStateType(parent_type, new_typename, simgrid::instr::Container::get_root()->get_type());
 193 }
 194
 195 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
 196                                               simgrid::instr::Type* root)
 197 {
 198   if (root->get_name() == type_name)
 199     static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
 200
 201   for (auto const& elm : root->get_children())
 202     recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
 203 }
 204
 205 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
 206 {
 207   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->get_type());
 208 }
 209
 210 namespace simgrid {
 211 namespace instr {
 212
 213 void platform_graph_export_graphviz(const std::string& output_filename)
 214 {
 215   auto* g     = xbt_graph_new_graph(0, nullptr);
 216   std::map<std::string, xbt_node_t, std::less<>> nodes;
 217   std::map<std::string, xbt_edge_t, std::less<>> edges;
 218   s4u::Engine::get_instance()->get_netzone_root()->extract_xbt_graph(g, &nodes, &edges);
 219
 220   std::ofstream fs;
 221   fs.open(output_filename, std::ofstream::out);
 222   xbt_assert(not fs.fail(), "Failed to open %s", output_filename.c_str());
 223
 224   if (g->directed)
 225     fs << "digraph test {" << std::endl;
 226   else
 227     fs << "graph test {" << std::endl;
 228
 229   fs << "  graph [overlap=scale]" << std::endl;
 230
 231   fs << "  node [shape=box, style=filled]" << std::endl;
 232   fs << "  node [width=.3, height=.3, style=filled, color=skyblue]" << std::endl << std::endl;
 233
 234   for (auto const& elm : nodes)
 235     fs << "  \"" << elm.first << "\";" << std::endl;
 236
 237   for (auto const& elm : edges) {
 238     const char* src_s = static_cast<char*>(elm.second->src->data);
 239     const char* dst_s = static_cast<char*>(elm.second->dst->data);
 240     if (g->directed)
 241       fs << "  \"" << src_s << "\" -> \"" << dst_s << "\";" << std::endl;
 242     else
 243       fs << "  \"" << src_s << "\" -- \"" << dst_s << "\";" << std::endl;
 244   }
 245   fs << "}" << std::endl;
 246   fs.close();
 247
 248   xbt_graph_free_graph(g, xbt_free_f, xbt_free_f, nullptr);
 249 }
 250
 251 /* Callbacks */
 252 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
 253 static void on_netzone_creation(s4u::NetZone const& netzone)
 254 {
 255   std::string id = netzone.get_name();
 256   if (Container::get_root() == nullptr) {
 257     auto* root = new NetZoneContainer(id, 0, nullptr);
 258     xbt_assert(Container::get_root() == root);
 259
 260     if (TRACE_smpi_is_enabled()) {
 261       auto* mpi = root->get_type()->by_name_or_create<ContainerType>("MPI");
 262       if (not TRACE_smpi_is_grouped())
 263         mpi->by_name_or_create<StateType>("MPI_STATE");
 264       root->get_type()->by_name_or_create("MPI_LINK", mpi, mpi);
 265       root->get_type()->by_name_or_create("MIGRATE_LINK", mpi, mpi);
 266       mpi->by_name_or_create<StateType>("MIGRATE_STATE");
 267     }
 268
 269     if (TRACE_needs_platform()) {
 270       currentContainer.push_back(root);
 271     }
 272     return;
 273   }
 274
 275   if (TRACE_needs_platform()) {
 276     auto level      = static_cast<unsigned>(currentContainer.size());
 277     auto* container = new NetZoneContainer(id, level, currentContainer.back());
 278     currentContainer.push_back(container);
 279   }
 280 }
 281
 282 static void on_link_creation(s4u::Link const& link)
 283 {
 284   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
 285     return;
 286
 287   auto* container = new Container(link.get_name(), "LINK", currentContainer.back());
 288
 289   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
 290     VariableType* bandwidth = container->get_type()->by_name_or_create("bandwidth", "");
 291     bandwidth->set_calling_container(container);
 292     bandwidth->set_event(0, link.get_bandwidth());
 293     VariableType* latency = container->get_type()->by_name_or_create("latency", "");
 294     latency->set_calling_container(container);
 295     latency->set_event(0, link.get_latency());
 296   }
 297
 298   if (TRACE_uncategorized()) {
 299     container->get_type()->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
 300   }
 301 }
 302
 303 static void on_host_creation(s4u::Host const& host)
 304 {
 305   if (Container::by_name_or_null(host.get_name())) // This host already exists, do nothing
 306     return;
 307
 308   Container* container  = new HostContainer(host, currentContainer.back());
 309   const Container* root = Container::get_root();
 310
 311   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
 312     VariableType* speed = container->get_type()->by_name_or_create("speed", "");
 313     speed->set_calling_container(container);
 314     speed->set_event(0, host.get_speed());
 315
 316     VariableType* cores = container->get_type()->by_name_or_create("core_count", "");
 317     cores->set_calling_container(container);
 318     cores->set_event(0, host.get_core_count());
 319   }
 320
 321   if (TRACE_uncategorized())
 322     container->get_type()->by_name_or_create("speed_used", "0.5 0.5 0.5");
 323
 324   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
 325     auto* mpi = container->get_type()->by_name_or_create<ContainerType>("MPI");
 326     mpi->by_name_or_create<StateType>("MPI_STATE");
 327     root->get_type()->by_name_or_create("MIGRATE_LINK", mpi, mpi);
 328     mpi->by_name_or_create<StateType>("MIGRATE_STATE");
 329   }
 330 }
 331
 332 static void on_action_state_change(kernel::resource::Action const& action,
 333                                    kernel::resource::Action::State /* previous */)
 334 {
 335   auto n = static_cast<unsigned>(action.get_variable()->get_number_of_constraint());
 336
 337   for (unsigned i = 0; i < n; i++) {
 338     double value = action.get_rate() * action.get_variable()->get_constraint_weight(i);
 339     /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
 340     kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
 341     const kernel::resource::CpuImpl* cpu = dynamic_cast<kernel::resource::CpuImpl*>(resource);
 342
 343     if (cpu != nullptr)
 344       resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
 345                                action.get_last_update(), simgrid_get_clock() - action.get_last_update());
 346
 347     const kernel::resource::StandardLinkImpl* link = dynamic_cast<kernel::resource::StandardLinkImpl*>(resource);
 348
 349     if (link != nullptr)
 350       resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
 351                                action.get_last_update(), simgrid_get_clock() - action.get_last_update());
 352   }
 353 }
 354
 355 static void on_platform_created()
 356 {
 357   currentContainer.clear();
 358   std::set<std::string, std::less<>> filter;
 359   XBT_DEBUG("Starting graph extraction.");
 360   recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), &filter);
 361   XBT_DEBUG("Graph extraction finished.");
 362   dump_buffer(true);
 363 }
 364
 365 static void on_actor_creation(s4u::Actor const& actor)
 366 {
 367   const Container* root      = Container::get_root();
 368   Container* container       = Container::by_name(actor.get_host()->get_name());
 369   std::string container_name = instr_pid(actor);
 370
 371   container->create_child(container_name, "ACTOR");
 372   auto* actor_type = container->get_type()->by_name_or_create<ContainerType>("ACTOR");
 373   auto* state      = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
 374   state->add_entity_value("suspend", "1 0 1");
 375   state->add_entity_value("sleep", "1 1 0");
 376   state->add_entity_value("receive", "1 0 0");
 377   state->add_entity_value("send", "0 0 1");
 378   state->add_entity_value("execute", "0 1 1");
 379   root->get_type()->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
 380
 381   actor.on_exit([container_name](bool failed) {
 382     if (failed)
 383       // kill means that this actor no longer exists, let's destroy it
 384       Container::by_name(container_name)->remove_from_parent();
 385   });
 386 }
 387
 388 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
 389 {
 390   static long long int counter = 0;
 391   Container* container         = Container::by_name(instr_pid(actor));
 392   LinkType* link               = Container::get_root()->get_link("ACTOR_LINK");
 393
 394   // start link
 395   link->start_event(container, "M", std::to_string(counter));
 396   // destroy existing container of this process
 397   container->remove_from_parent();
 398   // create new container on the new_host location
 399   Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
 400   // end link
 401   link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
 402   counter++;
 403 }
 404
 405 static void on_vm_creation(s4u::Host const& host)
 406 {
 407   const Container* container = new HostContainer(host, currentContainer.back());
 408   const Container* root      = Container::get_root();
 409   auto* vm                   = container->get_type()->by_name_or_create<ContainerType>("VM");
 410   auto* state                = vm->by_name_or_create<StateType>("VM_STATE");
 411   state->add_entity_value("suspend", "1 0 1");
 412   state->add_entity_value("sleep", "1 1 0");
 413   state->add_entity_value("receive", "1 0 0");
 414   state->add_entity_value("send", "0 0 1");
 415   state->add_entity_value("execute", "0 1 1");
 416   root->get_type()->by_name_or_create("VM_LINK", vm, vm);
 417   root->get_type()->by_name_or_create("VM_ACTOR_LINK", vm, vm);
 418 }
 419
 420 void define_callbacks()
 421 {
 422   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
 423   // properly
 424   if (TRACE_needs_platform()) {
 425     s4u::Engine::on_platform_created_cb(on_platform_created);
 426     s4u::Host::on_creation_cb(on_host_creation);
 427     s4u::Host::on_speed_change_cb([](s4u::Host const& host) {
 428       Container::by_name(host.get_name())
 429           ->get_variable("speed")
 430           ->set_event(simgrid_get_clock(), host.get_core_count() * host.get_available_speed());
 431     });
 432     s4u::Link::on_creation_cb(on_link_creation);
 433     s4u::Link::on_bandwidth_change_cb([](s4u::Link const& link) {
 434       Container::by_name(link.get_name())
 435           ->get_variable("bandwidth")
 436           ->set_event(simgrid_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
 437     });
 438     s4u::NetZone::on_seal_cb([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
 439     kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
 440       if (netpoint.is_router())
 441         new RouterContainer(netpoint.get_name(), currentContainer.back());
 442     });
 443   }
 444
 445   s4u::NetZone::on_creation_cb(on_netzone_creation);
 446
 447   kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
 448   s4u::Link::on_communication_state_change_cb(on_action_state_change);
 449
 450   if (TRACE_actor_is_enabled()) {
 451     s4u::Actor::on_creation_cb(on_actor_creation);
 452     s4u::Actor::on_destruction_cb([](s4u::Actor const& actor) {
 453       auto container = Container::by_name_or_null(instr_pid(actor));
 454       if (container != nullptr)
 455         container->remove_from_parent();
 456     });
 457     s4u::Actor::on_suspend_cb([](s4u::Actor const& actor) {
 458       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
 459     });
 460     s4u::Actor::on_resume_cb(
 461         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 462     s4u::Actor::on_sleep_cb([](s4u::Actor const& actor) {
 463       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
 464     });
 465     s4u::Actor::on_wake_up_cb(
 466         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
 467     s4u::Exec::on_start_cb([](s4u::Exec const&) {
 468       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("execute");
 469     });
 470     s4u::Activity::on_completion_cb([](const s4u::Activity&) {
 471       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->pop_event();
 472     });
 473     s4u::Comm::on_send_cb([](s4u::Comm const&) {
 474       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("send");
 475     });
 476     s4u::Comm::on_recv_cb([](s4u::Comm const&) {
 477       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("receive");
 478     });
 479     s4u::Actor::on_host_change_cb(on_actor_host_change);
 480   }
 481
 482   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing()) {
 483     s4u::Exec::on_start_cb([](s4u::Exec const& exec) {
 484       Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
 485           ->get_state("MPI_STATE")
 486           ->push_event("computing", new CpuTIData("compute", exec.get_cost()));
 487     });
 488     s4u::Activity::on_completion_cb([](const s4u::Activity&) {
 489       Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
 490           ->get_state("MPI_STATE")
 491           ->pop_event();
 492     });
 493   }
 494
 495   if (TRACE_vm_is_enabled()) {
 496     s4u::Host::on_creation_cb(on_vm_creation);
 497     s4u::VirtualMachine::on_start_cb([](s4u::VirtualMachine const& vm) {
 498       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
 499     });
 500     s4u::VirtualMachine::on_started_cb(
 501         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
 502     s4u::VirtualMachine::on_suspend_cb([](s4u::VirtualMachine const& vm) {
 503       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
 504     });
 505     s4u::VirtualMachine::on_resume_cb(
 506         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
 507     s4u::Host::on_destruction_cb(
 508         [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
 509   }
 510 }
 511 } // namespace instr
 512 } // namespace simgrid