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"
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 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);
153 * user categories support
155 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
156 simgrid::instr::Type* root)
158 if (root->get_name() == "HOST" || root->get_name() == "VM")
159 root->by_name_or_create(std::string("p") + new_typename, color);
161 if (root->get_name() == "LINK")
162 root->by_name_or_create(std::string("b") + new_typename, color);
164 for (auto const& elm : root->get_children()) {
165 recursiveNewVariableType(new_typename, color, elm.second.get());
169 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
171 recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
174 static void recursiveNewUserVariableType(const std::string& father_type, const std::string& new_typename,
175 const std::string& color, simgrid::instr::Type* root)
177 if (root->get_name() == father_type) {
178 root->by_name_or_create(new_typename, color);
180 for (auto const& elm : root->get_children())
181 recursiveNewUserVariableType(father_type, new_typename, color, elm.second.get());
184 void instr_new_user_variable_type(const std::string& father_type, const std::string& new_typename,
185 const std::string& color)
187 recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
190 static void recursiveNewUserStateType(const std::string& father_type, const std::string& new_typename,
191 simgrid::instr::Type* root)
193 if (root->get_name() == father_type)
194 root->by_name_or_create<simgrid::instr::StateType>(new_typename);
196 for (auto const& elm : root->get_children())
197 recursiveNewUserStateType(father_type, new_typename, elm.second.get());
200 void instr_new_user_state_type(const std::string& father_type, const std::string& new_typename)
202 recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
205 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
206 simgrid::instr::Type* root)
208 if (root->get_name() == type_name)
209 static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
211 for (auto const& elm : root->get_children())
212 recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
215 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
217 recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
223 void platform_graph_export_graphviz(const std::string& output_filename)
225 xbt_graph_t g = xbt_graph_new_graph(0, nullptr);
226 std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>();
227 std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>();
228 s4u::Engine::get_instance()->get_netzone_root()->extract_xbt_graph(g, nodes, edges);
231 fs.open(output_filename, std::ofstream::out);
232 xbt_assert(not fs.fail(), "Failed to open %s", output_filename.c_str());
235 fs << "digraph test {" << std::endl;
237 fs << "graph test {" << std::endl;
239 fs << " graph [overlap=scale]" << std::endl;
241 fs << " node [shape=box, style=filled]" << std::endl;
242 fs << " node [width=.3, height=.3, style=filled, color=skyblue]" << std::endl << std::endl;
244 for (auto const& elm : *nodes)
245 fs << " \"" << elm.first << "\";" << std::endl;
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);
251 fs << " \"" << src_s << "\" -> \"" << dst_s << "\";" << std::endl;
253 fs << " \"" << src_s << "\" -- \"" << dst_s << "\";" << std::endl;
255 fs << "}" << std::endl;
258 xbt_graph_free_graph(g, xbt_free_f, xbt_free_f, nullptr);
264 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
265 static void on_netzone_creation(s4u::NetZone const& netzone)
267 std::string id = netzone.get_name();
268 if (Container::get_root() == nullptr) {
269 NetZoneContainer* root = new NetZoneContainer(id, 0, nullptr);
270 xbt_assert(Container::get_root() == root);
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");
282 if (TRACE_needs_platform()) {
283 currentContainer.push_back(root);
288 if (TRACE_needs_platform()) {
289 NetZoneContainer* container = new NetZoneContainer(id, currentContainer.size(), currentContainer.back());
290 currentContainer.push_back(container);
294 static void on_link_creation(s4u::Link const& link)
296 if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
299 Container* container = new Container(link.get_name(), "LINK", currentContainer.back());
301 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
302 VariableType* bandwidth = container->type_->by_name_or_create("bandwidth", "");
303 bandwidth->set_calling_container(container);
304 bandwidth->set_event(0, link.get_bandwidth());
305 VariableType* latency = container->type_->by_name_or_create("latency", "");
306 latency->set_calling_container(container);
307 latency->set_event(0, link.get_latency());
310 if (TRACE_uncategorized()) {
311 container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
315 static void on_host_creation(s4u::Host const& host)
317 Container* container = new HostContainer(host, currentContainer.back());
318 const Container* root = Container::get_root();
320 if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
321 VariableType* speed = container->type_->by_name_or_create("speed", "");
322 speed->set_calling_container(container);
323 speed->set_event(0, host.get_speed());
325 VariableType* cores = container->type_->by_name_or_create("core_count", "");
326 cores->set_calling_container(container);
327 cores->set_event(0, host.get_core_count());
330 if (TRACE_uncategorized())
331 container->type_->by_name_or_create("speed_used", "0.5 0.5 0.5");
333 if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
334 ContainerType* mpi = container->type_->by_name_or_create<ContainerType>("MPI");
335 mpi->by_name_or_create<StateType>("MPI_STATE");
336 // TODO See if we can move this to the LoadBalancer plugin
337 root->type_->by_name_or_create("MIGRATE_LINK", mpi, mpi);
338 mpi->by_name_or_create<StateType>("MIGRATE_STATE");
342 static void on_action_state_change(kernel::resource::Action const& action,
343 kernel::resource::Action::State /* previous */)
345 int n = action.get_variable()->get_number_of_constraint();
347 for (int i = 0; i < n; i++) {
348 double value = action.get_variable()->get_value() * action.get_variable()->get_constraint_weight(i);
349 /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
350 kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
351 const kernel::resource::Cpu* cpu = dynamic_cast<kernel::resource::Cpu*>(resource);
354 resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
355 action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
357 const kernel::resource::LinkImpl* link = dynamic_cast<kernel::resource::LinkImpl*>(resource);
360 resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
361 action.get_last_update(), SIMIX_get_clock() - action.get_last_update());
365 static void on_platform_created()
367 currentContainer.clear();
368 std::set<std::string>* filter = new std::set<std::string>();
369 XBT_DEBUG("Starting graph extraction.");
370 recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), filter);
371 XBT_DEBUG("Graph extraction finished.");
376 static void on_actor_creation(s4u::Actor const& actor)
378 const Container* root = Container::get_root();
379 Container* container = Container::by_name(actor.get_host()->get_name());
380 std::string container_name = instr_pid(actor);
382 container->create_child(container_name, "ACTOR");
383 ContainerType* actor_type = container->type_->by_name_or_create<ContainerType>("ACTOR");
384 StateType* state = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
385 state->add_entity_value("suspend", "1 0 1");
386 state->add_entity_value("sleep", "1 1 0");
387 state->add_entity_value("receive", "1 0 0");
388 state->add_entity_value("send", "0 0 1");
389 state->add_entity_value("execute", "0 1 1");
390 root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
392 actor.on_exit([container_name](bool failed) {
394 // kill means that this actor no longer exists, let's destroy it
395 Container::by_name(container_name)->remove_from_parent();
399 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
401 static long long int counter = 0;
402 container_t container = Container::by_name(instr_pid(actor));
403 LinkType* link = Container::get_root()->get_link("ACTOR_LINK");
406 link->start_event(container, "M", std::to_string(counter));
407 // destroy existing container of this process
408 container->remove_from_parent();
409 // create new container on the new_host location
410 Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
412 link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
416 static void on_vm_creation(s4u::Host const& host)
418 const Container* container = new HostContainer(host, currentContainer.back());
419 const Container* root = Container::get_root();
420 ContainerType* vm = container->type_->by_name_or_create<ContainerType>("VM");
421 StateType* state = vm->by_name_or_create<StateType>("VM_STATE");
422 state->add_entity_value("suspend", "1 0 1");
423 state->add_entity_value("sleep", "1 1 0");
424 state->add_entity_value("receive", "1 0 0");
425 state->add_entity_value("send", "0 0 1");
426 state->add_entity_value("execute", "0 1 1");
427 root->type_->by_name_or_create("VM_LINK", vm, vm);
428 root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
431 void define_callbacks()
433 // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
435 if (TRACE_needs_platform()) {
436 s4u::Engine::on_platform_created.connect(on_platform_created);
437 s4u::Host::on_creation.connect(on_host_creation);
438 s4u::Host::on_speed_change.connect([](s4u::Host const& host) {
439 Container::by_name(host.get_name())
440 ->get_variable("speed")
441 ->set_event(surf_get_clock(), host.get_core_count() * host.get_available_speed());
443 s4u::Link::on_creation.connect(on_link_creation);
444 s4u::Link::on_bandwidth_change.connect([](s4u::Link const& link) {
445 Container::by_name(link.get_name())
446 ->get_variable("bandwidth")
447 ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
449 s4u::NetZone::on_seal.connect([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
450 kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
451 if (netpoint.is_router())
452 new RouterContainer(netpoint.get_name(), currentContainer.back());
456 s4u::NetZone::on_creation.connect(on_netzone_creation);
458 kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
459 s4u::Link::on_communication_state_change.connect(on_action_state_change);
461 if (TRACE_actor_is_enabled()) {
462 s4u::Actor::on_creation.connect(on_actor_creation);
463 s4u::Actor::on_destruction.connect([](s4u::Actor const& actor) {
464 auto container = Container::by_name_or_null(instr_pid(actor));
465 if (container != nullptr)
466 container->remove_from_parent();
468 s4u::Actor::on_suspend.connect([](s4u::Actor const& actor) {
469 Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
471 s4u::Actor::on_resume.connect(
472 [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
473 s4u::Actor::on_sleep.connect([](s4u::Actor const& actor) {
474 Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
476 s4u::Actor::on_wake_up.connect(
477 [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
478 s4u::Exec::on_start.connect([](simgrid::s4u::Actor const& actor, s4u::Exec const&) {
479 Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("execute");
481 s4u::Exec::on_completion.connect([](s4u::Actor const& actor, s4u::Exec const&) {
482 Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event();
484 s4u::Comm::on_sender_start.connect([](s4u::Actor const& actor) {
485 Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("send");
487 s4u::Comm::on_receiver_start.connect([](s4u::Actor const& actor) {
488 Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("receive");
490 s4u::Comm::on_completion.connect(
491 [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
492 s4u::Actor::on_host_change.connect(on_actor_host_change);
495 if (TRACE_vm_is_enabled()) {
496 s4u::Host::on_creation.connect(on_vm_creation);
497 s4u::VirtualMachine::on_start.connect([](s4u::VirtualMachine const& vm) {
498 Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
500 s4u::VirtualMachine::on_started.connect(
501 [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
502 s4u::VirtualMachine::on_suspend.connect([](s4u::VirtualMachine const& vm) {
503 Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
505 s4u::VirtualMachine::on_resume.connect(
506 [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
507 s4u::Host::on_destruction.connect(
508 [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
512 } // namespace simgrid