Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Use structured binding declarations (sonar, c++17).
[simgrid.git] / 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/NetworkModel.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& [_, edge] : edges) {
135     linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
136                    simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
137   }
138   xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
139 }
140
141 /*
142  * user categories support
143  */
144 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
145                                      simgrid::instr::Type* root)
146 {
147   if (root->get_name() == "HOST" || root->get_name() == "VM")
148     root->by_name_or_create(std::string("p") + new_typename, color);
149
150   if (root->get_name() == "LINK")
151     root->by_name_or_create(std::string("b") + new_typename, color);
152
153   for (auto const& [_, child] : root->get_children()) {
154     recursiveNewVariableType(new_typename, color, child.get());
155   }
156 }
157
158 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
159 {
160   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->get_type());
161 }
162
163 static void recursiveNewUserVariableType(const std::string& parent_type, const std::string& new_typename,
164                                          const std::string& color, simgrid::instr::Type* root)
165 {
166   if (root->get_name() == parent_type) {
167     root->by_name_or_create(new_typename, color);
168   }
169   for (auto const& [_, child] : root->get_children())
170     recursiveNewUserVariableType(parent_type, new_typename, color, child.get());
171 }
172
173 void instr_new_user_variable_type(const std::string& parent_type, const std::string& new_typename,
174                                   const std::string& color)
175 {
176   recursiveNewUserVariableType(parent_type, new_typename, color, simgrid::instr::Container::get_root()->get_type());
177 }
178
179 static void recursiveNewUserStateType(const std::string& parent_type, const std::string& new_typename,
180                                       simgrid::instr::Type* root)
181 {
182   if (root->get_name() == parent_type)
183     root->by_name_or_create<simgrid::instr::StateType>(new_typename);
184
185   for (auto const& [_, child] : root->get_children())
186     recursiveNewUserStateType(parent_type, new_typename, child.get());
187 }
188
189 void instr_new_user_state_type(const std::string& parent_type, const std::string& new_typename)
190 {
191   recursiveNewUserStateType(parent_type, new_typename, simgrid::instr::Container::get_root()->get_type());
192 }
193
194 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
195                                               simgrid::instr::Type* root)
196 {
197   if (root->get_name() == type_name)
198     static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
199
200   for (auto const& [_, child] : root->get_children())
201     recursiveNewValueForUserStateType(type_name, val, color, child.get());
202 }
203
204 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
205 {
206   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->get_type());
207 }
208
209 namespace simgrid {
210 namespace instr {
211
212 /** @brief Creates a file with the topology of the platform file used for the simulator.
213  *
214  *  The graph topology will have the following properties: all hosts, links and routers of the platform file are mapped
215  *  to graph nodes; routes are mapped to edges. The platform's zones are not represented in the output.
216  */
217 void platform_graph_export_graphviz(const std::string& output_filename)
218 {
219   auto* g     = xbt_graph_new_graph(0, nullptr);
220   std::map<std::string, xbt_node_t, std::less<>> nodes;
221   std::map<std::string, xbt_edge_t, std::less<>> edges;
222   s4u::Engine::get_instance()->get_netzone_root()->extract_xbt_graph(g, &nodes, &edges);
223
224   std::ofstream fs;
225   fs.open(output_filename, std::ofstream::out);
226   xbt_assert(not fs.fail(), "Failed to open %s", output_filename.c_str());
227
228   if (g->directed)
229     fs << "digraph test {" << std::endl;
230   else
231     fs << "graph test {" << std::endl;
232
233   fs << "  graph [overlap=scale]" << std::endl;
234
235   fs << "  node [shape=box, style=filled]" << std::endl;
236   fs << "  node [width=.3, height=.3, style=filled, color=skyblue]" << std::endl << std::endl;
237
238   for (auto const& [node, _] : nodes)
239     fs << "  \"" << node << "\";" << std::endl;
240
241   for (auto const& [_, edge] : edges) {
242     const char* src_s = static_cast<char*>(edge->src->data);
243     const char* dst_s = static_cast<char*>(edge->dst->data);
244     if (g->directed)
245       fs << "  \"" << src_s << "\" -> \"" << dst_s << "\";" << std::endl;
246     else
247       fs << "  \"" << src_s << "\" -- \"" << dst_s << "\";" << std::endl;
248   }
249   fs << "}" << std::endl;
250   fs.close();
251
252   xbt_graph_free_graph(g, xbt_free_f, xbt_free_f, nullptr);
253 }
254
255 /* Callbacks */
256 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
257 static void on_netzone_creation(s4u::NetZone const& netzone)
258 {
259   std::string id = netzone.get_name();
260   if (Container::get_root() == nullptr) {
261     auto* root = new NetZoneContainer(id, 0, nullptr);
262     xbt_assert(Container::get_root() == root);
263
264     if (TRACE_smpi_is_enabled()) {
265       auto* mpi = root->get_type()->by_name_or_create<ContainerType>("MPI");
266       if (not TRACE_smpi_is_grouped())
267         mpi->by_name_or_create<StateType>("MPI_STATE");
268       root->get_type()->by_name_or_create("MPI_LINK", mpi, mpi);
269       root->get_type()->by_name_or_create("MIGRATE_LINK", mpi, mpi);
270       mpi->by_name_or_create<StateType>("MIGRATE_STATE");
271     }
272
273     if (TRACE_needs_platform()) {
274       currentContainer.push_back(root);
275     }
276     return;
277   }
278
279   if (TRACE_needs_platform()) {
280     auto level      = static_cast<unsigned>(currentContainer.size());
281     auto* container = new NetZoneContainer(id, level, currentContainer.back());
282     currentContainer.push_back(container);
283   }
284 }
285
286 static void on_link_creation(s4u::Link const& link)
287 {
288   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
289     return;
290
291   auto* container = new Container(link.get_name(), "LINK", currentContainer.back());
292
293   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
294     VariableType* bandwidth = container->get_type()->by_name_or_create("bandwidth", "");
295     bandwidth->set_calling_container(container);
296     bandwidth->set_event(0, link.get_bandwidth());
297     VariableType* latency = container->get_type()->by_name_or_create("latency", "");
298     latency->set_calling_container(container);
299     latency->set_event(0, link.get_latency());
300   }
301
302   if (TRACE_uncategorized()) {
303     container->get_type()->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
304   }
305 }
306
307 static void on_host_creation(s4u::Host const& host)
308 {
309   if (Container::by_name_or_null(host.get_name())) // This host already exists, do nothing
310     return;
311
312   Container* container  = new HostContainer(host, currentContainer.back());
313   const Container* root = Container::get_root();
314
315   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
316     VariableType* speed = container->get_type()->by_name_or_create("speed", "");
317     speed->set_calling_container(container);
318     speed->set_event(0, host.get_speed());
319
320     VariableType* cores = container->get_type()->by_name_or_create("core_count", "");
321     cores->set_calling_container(container);
322     cores->set_event(0, host.get_core_count());
323   }
324
325   if (TRACE_uncategorized())
326     container->get_type()->by_name_or_create("speed_used", "0.5 0.5 0.5");
327
328   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
329     auto* mpi = container->get_type()->by_name_or_create<ContainerType>("MPI");
330     mpi->by_name_or_create<StateType>("MPI_STATE");
331     root->get_type()->by_name_or_create("MIGRATE_LINK", mpi, mpi);
332     mpi->by_name_or_create<StateType>("MIGRATE_STATE");
333   }
334 }
335
336 static void on_action_state_change(kernel::resource::Action const& action,
337                                    kernel::resource::Action::State /* previous */)
338 {
339   auto n = static_cast<unsigned>(action.get_variable()->get_number_of_constraint());
340
341   for (unsigned i = 0; i < n; i++) {
342     double value = action.get_rate() * action.get_variable()->get_constraint_weight(i);
343     /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
344     kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
345     const kernel::resource::CpuImpl* cpu = dynamic_cast<kernel::resource::CpuImpl*>(resource);
346
347     if (cpu != nullptr)
348       resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
349                                action.get_last_update(), simgrid_get_clock() - action.get_last_update());
350
351     const kernel::resource::StandardLinkImpl* link = dynamic_cast<kernel::resource::StandardLinkImpl*>(resource);
352
353     if (link != nullptr)
354       resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
355                                action.get_last_update(), simgrid_get_clock() - action.get_last_update());
356   }
357 }
358
359 static void on_platform_created()
360 {
361   currentContainer.clear();
362   std::set<std::string, std::less<>> filter;
363   XBT_DEBUG("Starting graph extraction.");
364   recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), &filter);
365   XBT_DEBUG("Graph extraction finished.");
366   dump_buffer(true);
367 }
368
369 static void on_actor_creation(s4u::Actor const& actor)
370 {
371   const Container* root      = Container::get_root();
372   Container* container       = Container::by_name(actor.get_host()->get_name());
373   std::string container_name = instr_pid(actor);
374
375   container->create_child(container_name, "ACTOR");
376   auto* actor_type = container->get_type()->by_name_or_create<ContainerType>("ACTOR");
377   auto* state      = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
378   state->add_entity_value("suspend", "1 0 1");
379   state->add_entity_value("sleep", "1 1 0");
380   state->add_entity_value("receive", "1 0 0");
381   state->add_entity_value("send", "0 0 1");
382   state->add_entity_value("execute", "0 1 1");
383   root->get_type()->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
384
385   actor.on_exit([container_name](bool failed) {
386     if (failed)
387       // kill means that this actor no longer exists, let's destroy it
388       Container::by_name(container_name)->remove_from_parent();
389   });
390 }
391
392 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
393 {
394   static long long int counter = 0;
395   Container* container         = Container::by_name(instr_pid(actor));
396   LinkType* link               = Container::get_root()->get_link("ACTOR_LINK");
397
398   // start link
399   link->start_event(container, "M", std::to_string(counter));
400   // destroy existing container of this process
401   container->remove_from_parent();
402   // create new container on the new_host location
403   Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
404   // end link
405   link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
406   counter++;
407 }
408
409 static void on_vm_creation(s4u::Host const& host)
410 {
411   const Container* container = new HostContainer(host, currentContainer.back());
412   const Container* root      = Container::get_root();
413   auto* vm                   = container->get_type()->by_name_or_create<ContainerType>("VM");
414   auto* state                = vm->by_name_or_create<StateType>("VM_STATE");
415   state->add_entity_value("suspend", "1 0 1");
416   state->add_entity_value("sleep", "1 1 0");
417   state->add_entity_value("receive", "1 0 0");
418   state->add_entity_value("send", "0 0 1");
419   state->add_entity_value("execute", "0 1 1");
420   root->get_type()->by_name_or_create("VM_LINK", vm, vm);
421   root->get_type()->by_name_or_create("VM_ACTOR_LINK", vm, vm);
422 }
423
424 void define_callbacks()
425 {
426   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
427   // properly
428   if (TRACE_needs_platform()) {
429     s4u::Engine::on_platform_created_cb(on_platform_created);
430     s4u::Host::on_creation_cb(on_host_creation);
431     s4u::Host::on_speed_change_cb([](s4u::Host const& host) {
432       Container::by_name(host.get_name())
433           ->get_variable("speed")
434           ->set_event(simgrid_get_clock(), host.get_core_count() * host.get_available_speed());
435     });
436     s4u::Link::on_creation_cb(on_link_creation);
437     s4u::Link::on_bandwidth_change_cb([](s4u::Link const& link) {
438       Container::by_name(link.get_name())
439           ->get_variable("bandwidth")
440           ->set_event(simgrid_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
441     });
442     s4u::NetZone::on_seal_cb([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
443     kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
444       if (netpoint.is_router())
445         new RouterContainer(netpoint.get_name(), currentContainer.back());
446     });
447   }
448
449   s4u::NetZone::on_creation_cb(on_netzone_creation);
450
451   kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
452   s4u::Link::on_communication_state_change_cb(on_action_state_change);
453
454   if (TRACE_actor_is_enabled()) {
455     s4u::Actor::on_creation_cb(on_actor_creation);
456     s4u::Actor::on_destruction_cb([](s4u::Actor const& actor) {
457       auto container = Container::by_name_or_null(instr_pid(actor));
458       if (container != nullptr)
459         container->remove_from_parent();
460     });
461     s4u::Actor::on_suspend_cb([](s4u::Actor const& actor) {
462       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
463     });
464     s4u::Actor::on_resume_cb(
465         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
466     s4u::Actor::on_sleep_cb([](s4u::Actor const& actor) {
467       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
468     });
469     s4u::Actor::on_wake_up_cb(
470         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
471     s4u::Exec::on_start_cb([](s4u::Exec const&) {
472       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("execute");
473     });
474     s4u::Activity::on_completion_cb([](const s4u::Activity&) {
475       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->pop_event();
476     });
477     s4u::Comm::on_send_cb([](s4u::Comm const&) {
478       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("send");
479     });
480     s4u::Comm::on_recv_cb([](s4u::Comm const&) {
481       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("receive");
482     });
483     s4u::Actor::on_host_change_cb(on_actor_host_change);
484   }
485
486   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing()) {
487     s4u::Exec::on_start_cb([](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()));
491     });
492     s4u::Activity::on_completion_cb([](const s4u::Activity&) {
493       Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
494           ->get_state("MPI_STATE")
495           ->pop_event();
496     });
497   }
498
499   if (TRACE_vm_is_enabled()) {
500     s4u::Host::on_creation_cb(on_vm_creation);
501     s4u::VirtualMachine::on_start_cb([](s4u::VirtualMachine const& vm) {
502       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
503     });
504     s4u::VirtualMachine::on_started_cb(
505         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
506     s4u::VirtualMachine::on_suspend_cb([](s4u::VirtualMachine const& vm) {
507       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
508     });
509     s4u::VirtualMachine::on_resume_cb(
510         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
511     s4u::Host::on_destruction_cb(
512         [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
513   }
514 }
515 } // namespace instr
516 } // namespace simgrid