1 /* Copyright (c) 2010-2021. 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"
25 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
27 std::string instr_pid(simgrid::s4u::Actor const& proc)
29 return std::string(proc.get_name()) + "-" + std::to_string(proc.get_pid());
32 static simgrid::instr::Container* lowestCommonAncestor(const simgrid::instr::Container* a1,
33 const simgrid::instr::Container* a2)
35 // this is only an optimization (since most of a1 and a2 share the same parent)
36 if (a1->father_ == a2->father_)
39 // create an array with all ancestors of a1
40 std::vector<simgrid::instr::Container*> ancestors_a1;
41 for (auto* p = a1->father_; p != nullptr; p = p->father_)
42 ancestors_a1.push_back(p);
44 // create an array with all ancestors of a2
45 std::vector<simgrid::instr::Container*> ancestors_a2;
46 for (auto* p = a2->father_; p != nullptr; p = p->father_)
47 ancestors_a2.push_back(p);
49 // find the lowest ancestor
50 simgrid::instr::Container* p = nullptr;
51 int i = static_cast<int>(ancestors_a1.size()) - 1;
52 int j = static_cast<int>(ancestors_a2.size()) - 1;
53 while (i >= 0 && j >= 0) {
54 simgrid::instr::Container* a1p = ancestors_a1.at(i);
55 const simgrid::instr::Container* a2p = ancestors_a2.at(j);
67 static void linkContainers(simgrid::instr::Container* src, simgrid::instr::Container* dst,
68 std::set<std::string, std::less<>>* filter)
71 if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
72 XBT_DEBUG(" linkContainers: ignoring loopback link");
77 simgrid::instr::Container* father = lowestCommonAncestor(src, dst);
78 xbt_assert(father, "common father unknown, this is a tracing problem");
80 // check if we already register this pair (we only need one direction)
81 std::string aux1 = src->get_name() + dst->get_name();
82 std::string aux2 = dst->get_name() + src->get_name();
83 if (filter->find(aux1) != filter->end()) {
84 XBT_DEBUG(" linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
87 if (filter->find(aux2) != filter->end()) {
88 XBT_DEBUG(" linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
92 // ok, not found, register it
97 std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
98 std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
99 std::to_string(dst->type_->get_id());
100 simgrid::instr::LinkType* link = father->type_->by_name_or_create(link_typename, src->type_, dst->type_);
101 link->set_calling_container(father);
104 static long long counter = 0;
106 std::string key = std::to_string(counter);
109 link->start_event(src, "topology", key);
110 link->end_event(dst, "topology", key);
112 XBT_DEBUG(" linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
115 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, simgrid::instr::Container* container,
116 std::set<std::string, std::less<>>* filter)
118 if (not TRACE_platform_topology()) {
119 XBT_DEBUG("Graph extraction disabled by user.");
122 XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
123 if (not netzone->get_children().empty()) {
124 // bottom-up recursion
125 for (auto const& nz_son : netzone->get_children()) {
126 simgrid::instr::Container* child_container = container->children_.at(nz_son->get_name());
127 recursiveGraphExtraction(nz_son, child_container, filter);
131 auto* graph = xbt_graph_new_graph(0, nullptr);
132 std::map<std::string, xbt_node_t, std::less<>> nodes;
133 std::map<std::string, xbt_edge_t, std::less<>> edges;
135 netzone->get_impl()->get_graph(graph, &nodes, &edges);
136 for (auto const& elm : edges) {
137 const xbt_edge* edge = elm.second;
138 linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
139 simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
141 xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
145 * user categories support
147 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
148 simgrid::instr::Type* root)
150 if (root->get_name() == "HOST" || root->get_name() == "VM")
151 root->by_name_or_create(std::string("p") + new_typename, color);
153 if (root->get_name() == "LINK")
154 root->by_name_or_create(std::string("b") + new_typename, color);
156 for (auto const& elm : root->get_children()) {
157 recursiveNewVariableType(new_typename, color, elm.second.get());
161 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
163 recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
166 static void recursiveNewUserVariableType(const std::string& father_type, const std::string& new_typename,
167 const std::string& color, simgrid::instr::Type* root)
169 if (root->get_name() == father_type) {
170 root->by_name_or_create(new_typename, color);
172 for (auto const& elm : root->get_children())
173 recursiveNewUserVariableType(father_type, new_typename, color, elm.second.get());
176 void instr_new_user_variable_type(const std::string& father_type, const std::string& new_typename,
177 const std::string& color)
179 recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
182 static void recursiveNewUserStateType(const std::string& father_type, const std::string& new_typename,
183 simgrid::instr::Type* root)
185 if (root->get_name() == father_type)
186 root->by_name_or_create<simgrid::instr::StateType>(new_typename);
188 for (auto const& elm : root->get_children())
189 recursiveNewUserStateType(father_type, new_typename, elm.second.get());
192 void instr_new_user_state_type(const std::string& father_type, const std::string& new_typename)
194 recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
197 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
198 simgrid::instr::Type* root)
200 if (root->get_name() == type_name)
201 static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
203 for (auto const& elm : root->get_children())
204 recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
207 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
209 recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
215 void platform_graph_export_graphviz(const std::string& output_filename)
217 auto* g = xbt_graph_new_graph(0, nullptr);
218 std::map<std::string, xbt_node_t, std::less<>> nodes;
219 std::map<std::string, xbt_edge_t, std::less<>> edges;
220 s4u::Engine::get_instance()->get_netzone_root()->extract_xbt_graph(g, &nodes, &edges);
223 fs.open(output_filename, std::ofstream::out);
224 xbt_assert(not fs.fail(), "Failed to open %s", output_filename.c_str());
227 fs << "digraph test {" << std::endl;
229 fs << "graph test {" << std::endl;
231 fs << " graph [overlap=scale]" << std::endl;
233 fs << " node [shape=box, style=filled]" << std::endl;
234 fs << " node [width=.3, height=.3, style=filled, color=skyblue]" << std::endl << std::endl;
236 for (auto const& elm : nodes)
237 fs << " \"" << elm.first << "\";" << std::endl;
239 for (auto const& elm : edges) {
240 const char* src_s = static_cast<char*>(elm.second->src->data);
241 const char* dst_s = static_cast<char*>(elm.second->dst->data);
243 fs << " \"" << src_s << "\" -> \"" << dst_s << "\";" << std::endl;
245 fs << " \"" << src_s << "\" -- \"" << dst_s << "\";" << std::endl;
247 fs << "}" << std::endl;
250 xbt_graph_free_graph(g, xbt_free_f, xbt_free_f, nullptr);
254 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
255 static void on_netzone_creation(s4u::NetZone const& netzone)
257 std::string id = netzone.get_name();
258 if (Container::get_root() == nullptr) {
259 auto* root = new NetZoneContainer(id, 0, nullptr);
260 xbt_assert(Container::get_root() == root);
262 if (TRACE_smpi_is_enabled()) {
263 auto* mpi = root->type_->by_name_or_create<ContainerType>("MPI");
264 if (not TRACE_smpi_is_grouped())
265 mpi->by_name_or_create<StateType>("MPI_STATE");
266 root->type_->by_name_or_create("MPI_LINK", mpi, mpi);
267 root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
268 mpi->by_name_or_create<StateType>("MIGRATE_STATE");
271 if (TRACE_needs_platform()) {
272 currentContainer.push_back(root);
277 if (TRACE_needs_platform()) {
278 auto level = static_cast<unsigned>(currentContainer.size());
279 auto* container = new NetZoneContainer(id, level, currentContainer.back());
280 currentContainer.push_back(container);
284 static void on_link_creation(s4u::Link const& link)
286 if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
289 auto* container = new Container(link.get_name(), "LINK", currentContainer.back());
291 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
292 VariableType* bandwidth = container->type_->by_name_or_create("bandwidth", "");
293 bandwidth->set_calling_container(container);
294 bandwidth->set_event(0, link.get_bandwidth());
295 VariableType* latency = container->type_->by_name_or_create("latency", "");
296 latency->set_calling_container(container);
297 latency->set_event(0, link.get_latency());
300 if (TRACE_uncategorized()) {
301 container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
305 static void on_host_creation(s4u::Host const& host)
307 if (Container::by_name_or_null(host.get_name())) // This host already exists, do nothing
310 Container* container = new HostContainer(host, currentContainer.back());
311 const Container* root = Container::get_root();
313 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
314 VariableType* speed = container->type_->by_name_or_create("speed", "");
315 speed->set_calling_container(container);
316 speed->set_event(0, host.get_speed());
318 VariableType* cores = container->type_->by_name_or_create("core_count", "");
319 cores->set_calling_container(container);
320 cores->set_event(0, host.get_core_count());
323 if (TRACE_uncategorized())
324 container->type_->by_name_or_create("speed_used", "0.5 0.5 0.5");
326 if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
327 auto* mpi = container->type_->by_name_or_create<ContainerType>("MPI");
328 mpi->by_name_or_create<StateType>("MPI_STATE");
329 root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
330 mpi->by_name_or_create<StateType>("MIGRATE_STATE");
334 static void on_action_state_change(kernel::resource::Action const& action,
335 kernel::resource::Action::State /* previous */)
337 auto n = static_cast<unsigned>(action.get_variable()->get_number_of_constraint());
339 for (unsigned i = 0; i < n; i++) {
340 double value = action.get_variable()->get_value() * action.get_variable()->get_constraint_weight(i);
341 /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
342 kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
343 const kernel::resource::CpuImpl* cpu = dynamic_cast<kernel::resource::CpuImpl*>(resource);
346 resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
347 action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
349 const kernel::resource::LinkImpl* link = dynamic_cast<kernel::resource::LinkImpl*>(resource);
352 resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
353 action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
357 static void on_platform_created()
359 currentContainer.clear();
360 std::set<std::string, std::less<>> filter;
361 XBT_DEBUG("Starting graph extraction.");
362 recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), &filter);
363 XBT_DEBUG("Graph extraction finished.");
367 static void on_actor_creation(s4u::Actor const& actor)
369 const Container* root = Container::get_root();
370 Container* container = Container::by_name(actor.get_host()->get_name());
371 std::string container_name = instr_pid(actor);
373 container->create_child(container_name, "ACTOR");
374 auto* actor_type = container->type_->by_name_or_create<ContainerType>("ACTOR");
375 auto* state = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
376 state->add_entity_value("suspend", "1 0 1");
377 state->add_entity_value("sleep", "1 1 0");
378 state->add_entity_value("receive", "1 0 0");
379 state->add_entity_value("send", "0 0 1");
380 state->add_entity_value("execute", "0 1 1");
381 root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
383 actor.on_exit([container_name](bool failed) {
385 // kill means that this actor no longer exists, let's destroy it
386 Container::by_name(container_name)->remove_from_parent();
390 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
392 static long long int counter = 0;
393 Container* container = Container::by_name(instr_pid(actor));
394 LinkType* link = Container::get_root()->get_link("ACTOR_LINK");
397 link->start_event(container, "M", std::to_string(counter));
398 // destroy existing container of this process
399 container->remove_from_parent();
400 // create new container on the new_host location
401 Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
403 link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
407 static void on_vm_creation(s4u::Host const& host)
409 const Container* container = new HostContainer(host, currentContainer.back());
410 const Container* root = Container::get_root();
411 auto* vm = container->type_->by_name_or_create<ContainerType>("VM");
412 auto* state = vm->by_name_or_create<StateType>("VM_STATE");
413 state->add_entity_value("suspend", "1 0 1");
414 state->add_entity_value("sleep", "1 1 0");
415 state->add_entity_value("receive", "1 0 0");
416 state->add_entity_value("send", "0 0 1");
417 state->add_entity_value("execute", "0 1 1");
418 root->type_->by_name_or_create("VM_LINK", vm, vm);
419 root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
422 void define_callbacks()
424 // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
426 if (TRACE_needs_platform()) {
427 s4u::Engine::on_platform_created.connect(on_platform_created);
428 s4u::Host::on_creation.connect(on_host_creation);
429 s4u::Host::on_speed_change.connect([](s4u::Host const& host) {
430 Container::by_name(host.get_name())
431 ->get_variable("speed")
432 ->set_event(surf_get_clock(), host.get_core_count() * host.get_available_speed());
434 s4u::Link::on_creation.connect(on_link_creation);
435 s4u::Link::on_bandwidth_change.connect([](s4u::Link const& link) {
436 Container::by_name(link.get_name())
437 ->get_variable("bandwidth")
438 ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
440 s4u::NetZone::on_seal.connect([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
441 kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
442 if (netpoint.is_router())
443 new RouterContainer(netpoint.get_name(), currentContainer.back());
447 s4u::NetZone::on_creation.connect(on_netzone_creation);
449 kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
450 s4u::Link::on_communication_state_change.connect(on_action_state_change);
452 if (TRACE_actor_is_enabled()) {
453 s4u::Actor::on_creation.connect(on_actor_creation);
454 s4u::Actor::on_destruction.connect([](s4u::Actor const& actor) {
455 auto container = Container::by_name_or_null(instr_pid(actor));
456 if (container != nullptr)
457 container->remove_from_parent();
459 s4u::Actor::on_suspend.connect([](s4u::Actor const& actor) {
460 Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
462 s4u::Actor::on_resume.connect(
463 [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
464 s4u::Actor::on_sleep.connect([](s4u::Actor const& actor) {
465 Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
467 s4u::Actor::on_wake_up.connect(
468 [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
469 s4u::Exec::on_start.connect([](s4u::Exec const&) {
470 Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("execute");
472 s4u::Exec::on_completion.connect([](s4u::Exec const&) {
473 Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->pop_event();
475 s4u::Comm::on_start.connect([](s4u::Comm const&, bool is_sender) {
476 Container::by_name(instr_pid(*s4u::Actor::self()))
477 ->get_state("ACTOR_STATE")
478 ->push_event(is_sender ? "send" : "receive");
480 s4u::Comm::on_completion.connect([](s4u::Comm const&) {
481 Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->pop_event();
483 s4u::Actor::on_host_change.connect(on_actor_host_change);
486 if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing()) {
487 s4u::Exec::on_start.connect([](s4u::Exec const& exec) {
488 Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
489 ->get_state("MPI_STATE")
490 ->push_event("computing", new CpuTIData("compute", exec.get_cost()));
492 s4u::Exec::on_completion.connect([](s4u::Exec const&) {
493 Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
494 ->get_state("MPI_STATE")
499 if (TRACE_vm_is_enabled()) {
500 s4u::Host::on_creation.connect(on_vm_creation);
501 s4u::VirtualMachine::on_start.connect([](s4u::VirtualMachine const& vm) {
502 Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
504 s4u::VirtualMachine::on_started.connect(
505 [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
506 s4u::VirtualMachine::on_suspend.connect([](s4u::VirtualMachine const& vm) {
507 Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
509 s4u::VirtualMachine::on_resume.connect(
510 [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
511 s4u::Host::on_destruction.connect(
512 [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
516 } // namespace simgrid