1 /* Copyright (c) 2010-2020. The SimGrid Team. All rights reserved. */
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. */
6 #include "src/instr/instr_private.hpp"
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"
23 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
25 static std::vector<simgrid::instr::NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
27 static const char* instr_node_name(const s_xbt_node_t* node)
29 return static_cast<char*>(xbt_graph_node_get_data(node));
32 static container_t lowestCommonAncestor(const simgrid::instr::Container* a1, const simgrid::instr::Container* a2)
34 // this is only an optimization (since most of a1 and a2 share the same parent)
35 if (a1->father_ == a2->father_)
38 // create an array with all ancestors of a1
39 std::vector<container_t> ancestors_a1;
40 container_t p = a1->father_;
42 ancestors_a1.push_back(p);
46 // create an array with all ancestors of a2
47 std::vector<container_t> ancestors_a2;
50 ancestors_a2.push_back(p);
54 // find the lowest ancestor
56 int i = ancestors_a1.size() - 1;
57 int j = 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);
72 static void linkContainers(container_t src, container_t dst, std::set<std::string>* filter)
75 if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
76 XBT_DEBUG(" linkContainers: ignoring loopback link");
81 container_t father = lowestCommonAncestor(src, dst);
83 xbt_die("common father unknown, this is a tracing problem");
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());
93 if (filter->find(aux2) != filter->end()) {
94 XBT_DEBUG(" linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
98 // ok, not found, register it
100 filter->insert(aux2);
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);
110 static long long counter = 0;
112 std::string key = std::to_string(counter);
115 link->start_event(src, "topology", key);
116 link->end_event(dst, "topology", key);
118 XBT_DEBUG(" linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
121 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, container_t container,
122 std::set<std::string>* filter)
124 if (not TRACE_platform_topology()) {
125 XBT_DEBUG("Graph extraction disabled by user.");
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);
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>();
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);
149 xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
155 static void instr_netzone_on_creation(simgrid::s4u::NetZone const& netzone)
157 std::string id = netzone.get_name();
158 if (simgrid::instr::Container::get_root() == nullptr) {
159 simgrid::instr::NetZoneContainer* root = new simgrid::instr::NetZoneContainer(id, 0, nullptr);
160 xbt_assert(simgrid::instr::Container::get_root() == root);
162 if (TRACE_smpi_is_enabled()) {
163 simgrid::instr::ContainerType* mpi = root->type_->by_name_or_create<simgrid::instr::ContainerType>("MPI");
164 if (not TRACE_smpi_is_grouped())
165 mpi->by_name_or_create<simgrid::instr::StateType>("MPI_STATE");
166 root->type_->by_name_or_create("MPI_LINK", mpi, mpi);
167 // TODO See if we can move this to the LoadBalancer plugin
168 root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
169 mpi->by_name_or_create<simgrid::instr::StateType>("MIGRATE_STATE");
172 if (TRACE_needs_platform()) {
173 currentContainer.push_back(root);
178 if (TRACE_needs_platform()) {
179 simgrid::instr::NetZoneContainer* container =
180 new simgrid::instr::NetZoneContainer(id, currentContainer.size(), currentContainer.back());
181 currentContainer.push_back(container);
185 static void instr_link_on_creation(simgrid::s4u::Link const& link)
187 if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
190 container_t container = new simgrid::instr::Container(link.get_name(), "LINK", currentContainer.back());
192 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
193 simgrid::instr::VariableType* bandwidth = container->type_->by_name_or_create("bandwidth", "");
194 bandwidth->set_calling_container(container);
195 bandwidth->set_event(0, link.get_bandwidth());
196 simgrid::instr::VariableType* latency = container->type_->by_name_or_create("latency", "");
197 latency->set_calling_container(container);
198 latency->set_event(0, link.get_latency());
200 if (TRACE_uncategorized()) {
201 container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
205 static void instr_host_on_creation(simgrid::s4u::Host const& host)
207 simgrid::instr::Container* container = new simgrid::instr::HostContainer(host, currentContainer.back());
208 const simgrid::instr::Container* root = simgrid::instr::Container::get_root();
210 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
211 simgrid::instr::VariableType* speed = container->type_->by_name_or_create("speed", "");
212 speed->set_calling_container(container);
213 speed->set_event(0, host.get_speed());
215 simgrid::instr::VariableType* cores = container->type_->by_name_or_create("core_count", "");
216 cores->set_calling_container(container);
217 cores->set_event(0, host.get_core_count());
220 if (TRACE_uncategorized())
221 container->type_->by_name_or_create("speed_used", "0.5 0.5 0.5");
223 if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
224 simgrid::instr::ContainerType* mpi = container->type_->by_name_or_create<simgrid::instr::ContainerType>("MPI");
225 mpi->by_name_or_create<simgrid::instr::StateType>("MPI_STATE");
226 // TODO See if we can move this to the LoadBalancer plugin
227 root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
228 mpi->by_name_or_create<simgrid::instr::StateType>("MIGRATE_STATE");
232 static void instr_host_on_speed_change(simgrid::s4u::Host const& host)
234 simgrid::instr::Container::by_name(host.get_name())
235 ->get_variable("speed")
236 ->set_event(surf_get_clock(), host.get_core_count() * host.get_available_speed());
239 static void instr_action_on_state_change(simgrid::kernel::resource::Action const& action,
240 simgrid::kernel::resource::Action::State /* previous */)
242 int n = action.get_variable()->get_number_of_constraint();
244 for (int i = 0; i < n; i++) {
245 double value = action.get_variable()->get_value() * action.get_variable()->get_constraint_weight(i);
246 /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
247 simgrid::kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
248 const simgrid::kernel::resource::Cpu* cpu = dynamic_cast<simgrid::kernel::resource::Cpu*>(resource);
251 TRACE_surf_resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
252 action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
254 const simgrid::kernel::resource::LinkImpl* link = dynamic_cast<simgrid::kernel::resource::LinkImpl*>(resource);
257 TRACE_surf_resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
258 action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
262 static void instr_link_on_bandwidth_change(simgrid::s4u::Link const& link)
264 simgrid::instr::Container::by_name(link.get_name())
265 ->get_variable("bandwidth")
266 ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
269 static void instr_netpoint_on_creation(simgrid::kernel::routing::NetPoint const& netpoint)
271 if (netpoint.is_router())
272 new simgrid::instr::RouterContainer(netpoint.get_name(), currentContainer.back());
275 static void instr_on_platform_created()
277 currentContainer.clear();
278 std::set<std::string>* filter = new std::set<std::string>();
279 XBT_DEBUG("Starting graph extraction.");
280 recursiveGraphExtraction(simgrid::s4u::Engine::get_instance()->get_netzone_root(),
281 simgrid::instr::Container::get_root(), filter);
282 XBT_DEBUG("Graph extraction finished.");
284 TRACE_paje_dump_buffer(true);
287 static void instr_actor_on_creation(simgrid::s4u::Actor const& actor)
289 const simgrid::instr::Container* root = simgrid::instr::Container::get_root();
290 simgrid::instr::Container* container = simgrid::instr::Container::by_name(actor.get_host()->get_name());
292 container->create_child(instr_pid(actor), "ACTOR");
293 simgrid::instr::ContainerType* actor_type =
294 container->type_->by_name_or_create<simgrid::instr::ContainerType>("ACTOR");
295 simgrid::instr::StateType* state = actor_type->by_name_or_create<simgrid::instr::StateType>("ACTOR_STATE");
296 state->add_entity_value("suspend", "1 0 1");
297 state->add_entity_value("sleep", "1 1 0");
298 state->add_entity_value("receive", "1 0 0");
299 state->add_entity_value("send", "0 0 1");
300 state->add_entity_value("execute", "0 1 1");
301 root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
302 root->type_->by_name_or_create("ACTOR_TASK_LINK", actor_type, actor_type);
304 std::string container_name = instr_pid(actor);
305 actor.on_exit([container_name](bool failed) {
307 // kill means that this actor no longer exists, let's destroy it
308 simgrid::instr::Container::by_name(container_name)->remove_from_parent();
312 static void instr_actor_on_host_change(simgrid::s4u::Actor const& actor,
313 simgrid::s4u::Host const& /*previous_location*/)
315 static long long int counter = 0;
316 container_t container = simgrid::instr::Container::by_name(instr_pid(actor));
317 simgrid::instr::LinkType* link = simgrid::instr::Container::get_root()->get_link("ACTOR_LINK");
320 link->start_event(container, "M", std::to_string(counter));
321 // destroy existing container of this process
322 container->remove_from_parent();
323 // create new container on the new_host location
324 simgrid::instr::Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
326 link->end_event(simgrid::instr::Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
330 static void instr_vm_on_creation(simgrid::s4u::Host const& host)
332 const simgrid::instr::Container* container = new simgrid::instr::HostContainer(host, currentContainer.back());
333 const simgrid::instr::Container* root = simgrid::instr::Container::get_root();
334 simgrid::instr::ContainerType* vm = container->type_->by_name_or_create<simgrid::instr::ContainerType>("VM");
335 simgrid::instr::StateType* state = vm->by_name_or_create<simgrid::instr::StateType>("VM_STATE");
336 state->add_entity_value("suspend", "1 0 1");
337 state->add_entity_value("sleep", "1 1 0");
338 state->add_entity_value("receive", "1 0 0");
339 state->add_entity_value("send", "0 0 1");
340 state->add_entity_value("execute", "0 1 1");
341 root->type_->by_name_or_create("VM_LINK", vm, vm);
342 root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
345 void instr_define_callbacks()
347 // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
349 if (TRACE_needs_platform()) {
350 simgrid::s4u::Engine::on_platform_created.connect(instr_on_platform_created);
351 simgrid::s4u::Host::on_creation.connect(instr_host_on_creation);
352 simgrid::s4u::Host::on_speed_change.connect(instr_host_on_speed_change);
353 simgrid::s4u::Link::on_creation.connect(instr_link_on_creation);
354 simgrid::s4u::Link::on_bandwidth_change.connect(instr_link_on_bandwidth_change);
355 simgrid::s4u::NetZone::on_seal.connect(
356 [](simgrid::s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
357 simgrid::kernel::routing::NetPoint::on_creation.connect(instr_netpoint_on_creation);
359 simgrid::s4u::NetZone::on_creation.connect(instr_netzone_on_creation);
361 simgrid::kernel::resource::CpuAction::on_state_change.connect(instr_action_on_state_change);
362 simgrid::s4u::Link::on_communication_state_change.connect(instr_action_on_state_change);
364 if (TRACE_actor_is_enabled()) {
365 simgrid::s4u::Actor::on_creation.connect(instr_actor_on_creation);
366 simgrid::s4u::Actor::on_destruction.connect([](simgrid::s4u::Actor const& actor) {
367 auto container = simgrid::instr::Container::by_name_or_null(instr_pid(actor));
368 if (container != nullptr)
369 container->remove_from_parent();
371 simgrid::s4u::Actor::on_suspend.connect([](simgrid::s4u::Actor const& actor) {
372 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
374 simgrid::s4u::Actor::on_resume.connect([](simgrid::s4u::Actor const& actor) {
375 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
377 simgrid::s4u::Actor::on_sleep.connect([](simgrid::s4u::Actor const& actor) {
378 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
380 simgrid::s4u::Actor::on_wake_up.connect([](simgrid::s4u::Actor const& actor) {
381 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
383 simgrid::s4u::Exec::on_start.connect([](simgrid::s4u::Actor const& actor, simgrid::s4u::Exec const&) {
384 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("execute");
386 simgrid::s4u::Exec::on_completion.connect([](simgrid::s4u::Actor const& actor, simgrid::s4u::Exec const&) {
387 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
389 simgrid::s4u::Comm::on_sender_start.connect([](simgrid::s4u::Actor const& actor) {
390 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("send");
392 simgrid::s4u::Comm::on_receiver_start.connect([](simgrid::s4u::Actor const& actor) {
393 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("receive");
395 simgrid::s4u::Comm::on_completion.connect([](simgrid::s4u::Actor const& actor) {
396 simgrid::instr::Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
398 simgrid::s4u::Actor::on_host_change.connect(instr_actor_on_host_change);
401 if (TRACE_vm_is_enabled()) {
402 simgrid::s4u::Host::on_creation.connect(instr_vm_on_creation);
403 simgrid::s4u::VirtualMachine::on_start.connect([](simgrid::s4u::VirtualMachine const& vm) {
404 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
406 simgrid::s4u::VirtualMachine::on_started.connect([](simgrid::s4u::VirtualMachine const& vm) {
407 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
409 simgrid::s4u::VirtualMachine::on_suspend.connect([](simgrid::s4u::VirtualMachine const& vm) {
410 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
412 simgrid::s4u::VirtualMachine::on_resume.connect([](simgrid::s4u::VirtualMachine const& vm) {
413 simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
415 simgrid::s4u::Host::on_destruction.connect([](simgrid::s4u::Host const& host) {
416 simgrid::instr::Container::by_name(host.get_name())->remove_from_parent();
421 * user categories support
423 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
424 simgrid::instr::Type* root)
426 if (root->get_name() == "HOST" || root->get_name() == "VM")
427 root->by_name_or_create(std::string("p") + new_typename, color);
429 if (root->get_name() == "LINK")
430 root->by_name_or_create(std::string("b") + new_typename, color);
432 for (auto const& elm : root->children_) {
433 recursiveNewVariableType(new_typename, color, elm.second.get());
437 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
439 recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
442 static void recursiveNewUserVariableType(const std::string& father_type, const std::string& new_typename,
443 const std::string& color, simgrid::instr::Type* root)
445 if (root->get_name() == father_type) {
446 root->by_name_or_create(new_typename, color);
448 for (auto const& elm : root->children_)
449 recursiveNewUserVariableType(father_type, new_typename, color, elm.second.get());
452 void instr_new_user_variable_type(const std::string& father_type, const std::string& new_typename,
453 const std::string& color)
455 recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
458 static void recursiveNewUserStateType(const std::string& father_type, const std::string& new_typename,
459 simgrid::instr::Type* root)
461 if (root->get_name() == father_type)
462 root->by_name_or_create<simgrid::instr::StateType>(new_typename);
464 for (auto const& elm : root->children_)
465 recursiveNewUserStateType(father_type, new_typename, elm.second.get());
468 void instr_new_user_state_type(const std::string& father_type, const std::string& new_typename)
470 recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
473 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
474 simgrid::instr::Type* root)
476 if (root->get_name() == type_name)
477 static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
479 for (auto const& elm : root->children_)
480 recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
483 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
485 recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
488 #define GRAPHICATOR_SUPPORT_FUNCTIONS
490 static void recursiveXBTGraphExtraction(const s_xbt_graph_t* graph, std::map<std::string, xbt_node_t>* nodes,
491 std::map<std::string, xbt_edge_t>* edges, const_sg_netzone_t netzone,
492 container_t container)
494 if (not netzone->get_children().empty()) {
495 // bottom-up recursion
496 for (auto const& netzone_child : netzone->get_children()) {
497 container_t child_container = container->children_.at(netzone_child->get_name());
498 recursiveXBTGraphExtraction(graph, nodes, edges, netzone_child, child_container);
502 netzone->get_impl()->get_graph(graph, nodes, edges);
505 xbt_graph_t instr_routing_platform_graph()
507 xbt_graph_t ret = xbt_graph_new_graph(0, nullptr);
508 std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>();
509 std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>();
510 recursiveXBTGraphExtraction(ret, nodes, edges, simgrid::s4u::Engine::get_instance()->get_netzone_root(),
511 simgrid::instr::Container::get_root());
517 void instr_routing_platform_graph_export_graphviz(const s_xbt_graph_t* g, const char* filename)
519 unsigned int cursor = 0;
520 xbt_node_t node = nullptr;
521 xbt_edge_t edge = nullptr;
523 FILE* file = fopen(filename, "w");
524 xbt_assert(file, "Failed to open %s \n", filename);
527 fprintf(file, "digraph test {\n");
529 fprintf(file, "graph test {\n");
531 fprintf(file, " graph [overlap=scale]\n");
533 fprintf(file, " node [shape=box, style=filled]\n");
534 fprintf(file, " node [width=.3, height=.3, style=filled, color=skyblue]\n\n");
536 xbt_dynar_foreach (g->nodes, cursor, node) {
537 fprintf(file, " \"%s\";\n", instr_node_name(node));
539 xbt_dynar_foreach (g->edges, cursor, edge) {
540 const char* src_s = instr_node_name(edge->src);
541 const char* dst_s = instr_node_name(edge->dst);
543 fprintf(file, " \"%s\" -> \"%s\";\n", src_s, dst_s);
545 fprintf(file, " \"%s\" -- \"%s\";\n", src_s, dst_s);
547 fprintf(file, "}\n");