src/instr/instr_platform.cpp

   1 /* Copyright (c) 2010-2018. 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/Engine.hpp"
  12 #include "simgrid/s4u/Host.hpp"
  13 #include "simgrid/s4u/VirtualMachine.hpp"
  14 #include "src/surf/cpu_interface.hpp"
  15 #include "src/surf/network_interface.hpp"
  16 #include "src/surf/xml/platf_private.hpp"
  17 #include "surf/surf.hpp"
  18 #include "xbt/graph.h"
  19
  20 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
  21
  22 static std::vector<simgrid::instr::NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
  23
  24 static const char* instr_node_name(xbt_node_t node)
  25 {
  26   return static_cast<char*>(xbt_graph_node_get_data(node));
  27 }
  28
  29 static container_t lowestCommonAncestor(container_t a1, container_t a2)
  30 {
  31   // this is only an optimization (since most of a1 and a2 share the same parent)
  32   if (a1->father_ == a2->father_)
  33     return a1->father_;
  34
  35   // create an array with all ancestors of a1
  36   std::vector<container_t> ancestors_a1;
  37   container_t p = a1->father_;
  38   while (p) {
  39     ancestors_a1.push_back(p);
  40     p = p->father_;
  41   }
  42
  43   // create an array with all ancestors of a2
  44   std::vector<container_t> ancestors_a2;
  45   p = a2->father_;
  46   while (p) {
  47     ancestors_a2.push_back(p);
  48     p = p->father_;
  49   }
  50
  51   // find the lowest ancestor
  52   p     = nullptr;
  53   int i = ancestors_a1.size() - 1;
  54   int j = ancestors_a2.size() - 1;
  55   while (i >= 0 && j >= 0) {
  56     container_t a1p = ancestors_a1.at(i);
  57     container_t a2p = ancestors_a2.at(j);
  58     if (a1p == a2p) {
  59       p = a1p;
  60     } else {
  61       break;
  62     }
  63     i--;
  64     j--;
  65   }
  66   return p;
  67 }
  68
  69 static void linkContainers(container_t src, container_t dst, std::set<std::string>* filter)
  70 {
  71   // ignore loopback
  72   if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
  73     XBT_DEBUG("  linkContainers: ignoring loopback link");
  74     return;
  75   }
  76
  77   // find common father
  78   container_t father = lowestCommonAncestor(src, dst);
  79   if (not father) {
  80     xbt_die("common father unknown, this is a tracing problem");
  81   }
  82
  83   // check if we already register this pair (we only need one direction)
  84   std::string aux1 = src->get_name() + dst->get_name();
  85   std::string aux2 = dst->get_name() + src->get_name();
  86   if (filter->find(aux1) != filter->end()) {
  87     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
  88     return;
  89   }
  90   if (filter->find(aux2) != filter->end()) {
  91     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
  92     return;
  93   }
  94
  95   // ok, not found, register it
  96   filter->insert(aux1);
  97   filter->insert(aux2);
  98
  99   // declare type
 100   std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
 101                               std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
 102                               std::to_string(dst->type_->get_id());
 103   simgrid::instr::LinkType* link = father->type_->getOrCreateLinkType(link_typename, src->type_, dst->type_);
 104   link->setCallingContainer(father);
 105
 106   // register EDGE types for triva configuration
 107   trivaEdgeTypes.insert(link->get_name());
 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->startEvent(src, "topology", key);
 116   link->endEvent(dst, "topology", key);
 117
 118   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
 119 }
 120
 121 static void recursiveGraphExtraction(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->getChildren()->empty()) {
 130     // bottom-up recursion
 131     for (auto const& nz_son : *netzone->getChildren()) {
 132       container_t child_container = container->children_.at(nz_son->get_cname());
 133       recursiveGraphExtraction(nz_son, child_container, filter);
 134     }
 135   }
 136
 137   xbt_graph_t graph                        = xbt_graph_new_graph(0, nullptr);
 138   std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
 139   std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
 140
 141   static_cast<simgrid::kernel::routing::NetZoneImpl*>(netzone)->get_graph(graph, nodes, edges);
 142   for (auto elm : *edges) {
 143     xbt_edge_t edge = elm.second;
 144     linkContainers(simgrid::instr::Container::byName(static_cast<const char*>(edge->src->data)),
 145                    simgrid::instr::Container::byName(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  * Callbacks
 154  */
 155 static void instr_netzone_on_creation(simgrid::s4u::NetZone& netzone)
 156 {
 157   std::string id = netzone.get_name();
 158   if (simgrid::instr::Container::getRoot() == nullptr) {
 159     simgrid::instr::NetZoneContainer* root = new simgrid::instr::NetZoneContainer(id, 0, nullptr);
 160
 161     if (TRACE_smpi_is_enabled()) {
 162       simgrid::instr::Type* mpi = root->type_->getOrCreateContainerType("MPI");
 163       if (not TRACE_smpi_is_grouped())
 164         mpi->getOrCreateStateType("MPI_STATE");
 165       root->type_->getOrCreateLinkType("MPI_LINK", mpi, mpi);
 166       // TODO See if we can move this to the LoadBalancer plugin
 167       root->type_->getOrCreateLinkType("MIGRATE_LINK", mpi, mpi);
 168       mpi->getOrCreateStateType("MIGRATE_STATE");
 169     }
 170
 171     if (TRACE_needs_platform()) {
 172       currentContainer.push_back(root);
 173     }
 174     return;
 175   }
 176
 177   if (TRACE_needs_platform()) {
 178     simgrid::instr::NetZoneContainer* container =
 179         new simgrid::instr::NetZoneContainer(id, currentContainer.size(), currentContainer.back());
 180     currentContainer.push_back(container);
 181   }
 182 }
 183
 184 static void instr_netzone_on_seal(simgrid::s4u::NetZone& /*netzone*/)
 185 {
 186   if (TRACE_needs_platform()) {
 187     currentContainer.pop_back();
 188   }
 189 }
 190
 191 static void instr_link_on_creation(simgrid::s4u::Link& link)
 192 {
 193   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
 194     return;
 195
 196   container_t container = new simgrid::instr::Container(link.get_name(), "LINK", currentContainer.back());
 197
 198   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
 199     simgrid::instr::VariableType* bandwidth = container->type_->getOrCreateVariableType("bandwidth", "");
 200     bandwidth->setCallingContainer(container);
 201     bandwidth->setEvent(0, link.bandwidth());
 202     simgrid::instr::VariableType* latency = container->type_->getOrCreateVariableType("latency", "");
 203     latency->setCallingContainer(container);
 204     latency->setEvent(0, link.latency());
 205   }
 206   if (TRACE_uncategorized()) {
 207     container->type_->getOrCreateVariableType("bandwidth_used", "0.5 0.5 0.5");
 208   }
 209 }
 210
 211 static void instr_host_on_creation(simgrid::s4u::Host& host)
 212 {
 213   container_t container = new simgrid::instr::HostContainer(host, currentContainer.back());
 214   container_t root      = simgrid::instr::Container::getRoot();
 215
 216   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
 217     simgrid::instr::VariableType* power = container->type_->getOrCreateVariableType("power", "");
 218     power->setCallingContainer(container);
 219     power->setEvent(0, host.getSpeed());
 220   }
 221
 222   if (TRACE_uncategorized())
 223     container->type_->getOrCreateVariableType("power_used", "0.5 0.5 0.5");
 224
 225   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
 226     simgrid::instr::ContainerType* mpi = container->type_->getOrCreateContainerType("MPI");
 227     mpi->getOrCreateStateType("MPI_STATE");
 228     // TODO See if we can move this to the LoadBalancer plugin
 229     root->type_->getOrCreateLinkType("MIGRATE_LINK", mpi, mpi);
 230     mpi->getOrCreateStateType("MIGRATE_STATE");
 231   }
 232 }
 233
 234 static void instr_cpu_action_on_state_change(simgrid::surf::CpuAction* action,
 235                                              simgrid::kernel::resource::Action::State /* previous */)
 236 {
 237   simgrid::surf::Cpu* cpu = static_cast<simgrid::surf::Cpu*>(action->get_variable()->get_constraint(0)->get_id());
 238   TRACE_surf_resource_set_utilization("HOST", "power_used", cpu->get_cname(), action->get_category(),
 239                                       action->get_variable()->get_value(), action->get_last_update(),
 240                                       SIMIX_get_clock() - action->get_last_update());
 241 }
 242
 243 static void instr_netpoint_on_creation(simgrid::kernel::routing::NetPoint* netpoint)
 244 {
 245   if (netpoint->is_router() && TRACE_needs_platform() && TRACE_is_enabled())
 246     new simgrid::instr::RouterContainer(netpoint->get_cname(), currentContainer.back());
 247 }
 248
 249 static void instr_on_platform_created()
 250 {
 251   currentContainer.clear();
 252   std::set<std::string>* filter = new std::set<std::string>;
 253   XBT_DEBUG("Starting graph extraction.");
 254   recursiveGraphExtraction(simgrid::s4u::Engine::get_instance()->get_netzone_root(),
 255                            simgrid::instr::Container::getRoot(), filter);
 256   XBT_DEBUG("Graph extraction finished.");
 257   delete filter;
 258   TRACE_paje_dump_buffer(true);
 259 }
 260
 261 static void instr_actor_on_creation(simgrid::s4u::ActorPtr actor)
 262 {
 263   container_t root      = simgrid::instr::Container::getRoot();
 264   container_t container = simgrid::instr::Container::byName(actor->get_host()->get_name());
 265
 266   container->createChild(instr_pid(actor.get()), "ACTOR");
 267   simgrid::instr::ContainerType* actor_type = container->type_->getOrCreateContainerType("ACTOR");
 268   simgrid::instr::StateType* state          = actor_type->getOrCreateStateType("ACTOR_STATE");
 269   state->addEntityValue("suspend", "1 0 1");
 270   state->addEntityValue("sleep", "1 1 0");
 271   state->addEntityValue("receive", "1 0 0");
 272   state->addEntityValue("send", "0 0 1");
 273   state->addEntityValue("task_execute", "0 1 1");
 274   root->type_->getOrCreateLinkType("ACTOR_LINK", actor_type, actor_type);
 275   root->type_->getOrCreateLinkType("ACTOR_TASK_LINK", actor_type, actor_type);
 276 }
 277
 278 static void instr_actor_on_suspend(simgrid::s4u::ActorPtr actor)
 279 {
 280   simgrid::instr::Container::byName(instr_pid(actor.get()))->getState("ACTOR_STATE")->pushEvent("suspend");
 281 }
 282
 283 static void instr_actor_on_resume(simgrid::s4u::ActorPtr actor)
 284 {
 285   simgrid::instr::Container::byName(instr_pid(actor.get()))->getState("ACTOR_STATE")->popEvent();
 286 }
 287
 288 static long long int counter = 0;
 289
 290 static void instr_actor_on_migration_start(simgrid::s4u::ActorPtr actor)
 291 {
 292   // start link
 293   container_t container = simgrid::instr::Container::byName(instr_pid(actor.get()));
 294   simgrid::instr::Container::getRoot()->getLink("ACTOR_LINK")->startEvent(container, "M", std::to_string(counter));
 295
 296   // destroy existing container of this process
 297   container->removeFromParent();
 298 }
 299
 300 static void instr_actor_on_migration_end(simgrid::s4u::ActorPtr actor)
 301 {
 302   // create new container on the new_host location
 303   simgrid::instr::Container::byName(actor->get_host()->get_name())->createChild(instr_pid(actor.get()), "ACTOR");
 304   // end link
 305   simgrid::instr::Container::getRoot()
 306       ->getLink("ACTOR_LINK")
 307       ->endEvent(simgrid::instr::Container::byName(instr_pid(actor.get())), "M", std::to_string(counter));
 308   counter++;
 309 }
 310
 311 static void instr_vm_on_creation(simgrid::s4u::Host& host)
 312 {
 313   container_t container                 = new simgrid::instr::HostContainer(host, currentContainer.back());
 314   container_t root                      = simgrid::instr::Container::getRoot();
 315   simgrid::instr::ContainerType* msg_vm = container->type_->getOrCreateContainerType("VM");
 316   simgrid::instr::StateType* state      = msg_vm->getOrCreateStateType("VM_STATE");
 317   state->addEntityValue("suspend", "1 0 1");
 318   state->addEntityValue("sleep", "1 1 0");
 319   state->addEntityValue("receive", "1 0 0");
 320   state->addEntityValue("send", "0 0 1");
 321   state->addEntityValue("task_execute", "0 1 1");
 322   root->type_->getOrCreateLinkType("VM_LINK", msg_vm, msg_vm);
 323   root->type_->getOrCreateLinkType("VM_ACTOR_LINK", msg_vm, msg_vm);
 324 }
 325
 326 static void instr_vm_on_start(simgrid::s4u::VirtualMachine& vm)
 327 {
 328   simgrid::instr::Container::byName(vm.get_name())->getState("VM_STATE")->pushEvent("start");
 329 }
 330
 331 static void instr_vm_on_started(simgrid::s4u::VirtualMachine& vm)
 332 {
 333   simgrid::instr::Container::byName(vm.get_name())->getState("VM_STATE")->popEvent();
 334 }
 335
 336 static void instr_vm_on_suspend(simgrid::s4u::VirtualMachine& vm)
 337 {
 338   simgrid::instr::Container::byName(vm.get_name())->getState("VM_STATE")->pushEvent("suspend");
 339 }
 340
 341 static void instr_vm_on_resume(simgrid::s4u::VirtualMachine& vm)
 342 {
 343   simgrid::instr::Container::byName(vm.get_name())->getState("VM_STATE")->popEvent();
 344 }
 345
 346 static void instr_vm_on_destruction(simgrid::s4u::Host& host)
 347 {
 348   simgrid::instr::Container::byName(host.get_name())->removeFromParent();
 349 }
 350
 351 void instr_define_callbacks()
 352 {
 353   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
 354   // properly
 355   if (TRACE_needs_platform()) {
 356     simgrid::s4u::on_platform_created.connect(instr_on_platform_created);
 357     simgrid::s4u::Host::onCreation.connect(instr_host_on_creation);
 358     simgrid::s4u::Link::onCreation.connect(instr_link_on_creation);
 359   }
 360   simgrid::s4u::NetZone::onCreation.connect(instr_netzone_on_creation);
 361   simgrid::s4u::NetZone::onSeal.connect(instr_netzone_on_seal);
 362   simgrid::kernel::routing::NetPoint::onCreation.connect(instr_netpoint_on_creation);
 363
 364   simgrid::surf::CpuAction::onStateChange.connect(instr_cpu_action_on_state_change);
 365
 366   if (TRACE_actor_is_enabled()) {
 367     simgrid::s4u::Actor::on_creation.connect(instr_actor_on_creation);
 368     simgrid::s4u::Actor::on_suspend.connect(instr_actor_on_suspend);
 369     simgrid::s4u::Actor::on_resume.connect(instr_actor_on_resume);
 370     simgrid::s4u::Actor::on_migration_start.connect(instr_actor_on_migration_start);
 371     simgrid::s4u::Actor::on_migration_end.connect(instr_actor_on_migration_end);
 372   }
 373
 374   if (TRACE_vm_is_enabled()) {
 375     simgrid::s4u::Host::onCreation.connect(instr_vm_on_creation);
 376     simgrid::s4u::VirtualMachine::on_start.connect(instr_vm_on_start);
 377     simgrid::s4u::VirtualMachine::on_started.connect(instr_vm_on_started);
 378     simgrid::s4u::VirtualMachine::on_suspend.connect(instr_vm_on_suspend);
 379     simgrid::s4u::VirtualMachine::on_resume.connect(instr_vm_on_resume);
 380     simgrid::s4u::Host::onDestruction.connect(instr_vm_on_destruction);
 381   }
 382 }
 383 /*
 384  * user categories support
 385  */
 386 static void recursiveNewVariableType(std::string new_typename, std::string color, simgrid::instr::Type* root)
 387 {
 388   if (root->get_name() == "HOST" || root->get_name() == "VM")
 389     root->getOrCreateVariableType(std::string("p") + new_typename, color);
 390
 391   if (root->get_name() == "LINK")
 392     root->getOrCreateVariableType(std::string("b") + new_typename, color);
 393
 394   for (auto elm : root->children_) {
 395     recursiveNewVariableType(new_typename, color, elm.second);
 396   }
 397 }
 398
 399 void instr_new_variable_type(std::string new_typename, std::string color)
 400 {
 401   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::getRoot()->type_);
 402 }
 403
 404 static void recursiveNewUserVariableType(std::string father_type, std::string new_typename, std::string color,
 405                                          simgrid::instr::Type* root)
 406 {
 407   if (root->get_name() == father_type) {
 408     root->getOrCreateVariableType(new_typename, color);
 409   }
 410   for (auto elm : root->children_)
 411     recursiveNewUserVariableType(father_type, new_typename, color, elm.second);
 412 }
 413
 414 void instr_new_user_variable_type(std::string father_type, std::string new_typename, std::string color)
 415 {
 416   recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::getRoot()->type_);
 417 }
 418
 419 static void recursiveNewUserStateType(std::string father_type, std::string new_typename, simgrid::instr::Type* root)
 420 {
 421   if (root->get_name() == father_type)
 422     root->getOrCreateStateType(new_typename);
 423
 424   for (auto elm : root->children_)
 425     recursiveNewUserStateType(father_type, new_typename, elm.second);
 426 }
 427
 428 void instr_new_user_state_type(std::string father_type, std::string new_typename)
 429 {
 430   recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::getRoot()->type_);
 431 }
 432
 433 static void recursiveNewValueForUserStateType(std::string type_name, const char* val, std::string color,
 434                                               simgrid::instr::Type* root)
 435 {
 436   if (root->get_name() == type_name)
 437     static_cast<simgrid::instr::StateType*>(root)->addEntityValue(val, color);
 438
 439   for (auto elm : root->children_)
 440     recursiveNewValueForUserStateType(type_name, val, color, elm.second);
 441 }
 442
 443 void instr_new_value_for_user_state_type(std::string type_name, const char* value, std::string color)
 444 {
 445   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::getRoot()->type_);
 446 }
 447
 448 #define GRAPHICATOR_SUPPORT_FUNCTIONS
 449
 450 static void recursiveXBTGraphExtraction(xbt_graph_t graph, std::map<std::string, xbt_node_t>* nodes,
 451                                         std::map<std::string, xbt_edge_t>* edges, sg_netzone_t netzone,
 452                                         container_t container)
 453 {
 454   if (not netzone->getChildren()->empty()) {
 455     // bottom-up recursion
 456     for (auto const& netzone_child : *netzone->getChildren()) {
 457       container_t child_container = container->children_.at(netzone_child->get_cname());
 458       recursiveXBTGraphExtraction(graph, nodes, edges, netzone_child, child_container);
 459     }
 460   }
 461
 462   static_cast<simgrid::kernel::routing::NetZoneImpl*>(netzone)->get_graph(graph, nodes, edges);
 463 }
 464
 465 xbt_graph_t instr_routing_platform_graph()
 466 {
 467   xbt_graph_t ret                          = xbt_graph_new_graph(0, nullptr);
 468   std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
 469   std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
 470   recursiveXBTGraphExtraction(ret, nodes, edges, simgrid::s4u::Engine::get_instance()->get_netzone_root(),
 471                               simgrid::instr::Container::getRoot());
 472   delete nodes;
 473   delete edges;
 474   return ret;
 475 }
 476
 477 void instr_routing_platform_graph_export_graphviz(xbt_graph_t g, const char* filename)
 478 {
 479   unsigned int cursor = 0;
 480   xbt_node_t node     = nullptr;
 481   xbt_edge_t edge     = nullptr;
 482
 483   FILE* file = fopen(filename, "w");
 484   xbt_assert(file, "Failed to open %s \n", filename);
 485
 486   if (g->directed)
 487     fprintf(file, "digraph test {\n");
 488   else
 489     fprintf(file, "graph test {\n");
 490
 491   fprintf(file, "  graph [overlap=scale]\n");
 492
 493   fprintf(file, "  node [shape=box, style=filled]\n");
 494   fprintf(file, "  node [width=.3, height=.3, style=filled, color=skyblue]\n\n");
 495
 496   xbt_dynar_foreach (g->nodes, cursor, node) {
 497     fprintf(file, "  \"%s\";\n", instr_node_name(node));
 498   }
 499   xbt_dynar_foreach (g->edges, cursor, edge) {
 500     const char* src_s = instr_node_name(edge->src);
 501     const char* dst_s = instr_node_name(edge->dst);
 502     if (g->directed)
 503       fprintf(file, "  \"%s\" -> \"%s\";\n", src_s, dst_s);
 504     else
 505       fprintf(file, "  \"%s\" -- \"%s\";\n", src_s, dst_s);
 506   }
 507   fprintf(file, "}\n");
 508   fclose(file);
 509 }