Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
include cleanups (mostly surf/surf.hpp)
[simgrid.git] / src / instr / instr_platform.cpp
1 /* Copyright (c) 2010-2021. 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/surf/cpu_interface.hpp"
18 #include "src/surf/network_interface.hpp"
19 #include "src/surf/surf_interface.hpp"
20 #include "src/surf/xml/platf_private.hpp"
21
22 #include <fstream>
23
24 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
25
26 std::string instr_pid(simgrid::s4u::Actor const& proc)
27 {
28   return std::string(proc.get_name()) + "-" + std::to_string(proc.get_pid());
29 }
30
31 static simgrid::instr::Container* lowestCommonAncestor(const simgrid::instr::Container* a1,
32                                                        const simgrid::instr::Container* a2)
33 {
34   // this is only an optimization (since most of a1 and a2 share the same parent)
35   if (a1->get_parent() == a2->get_parent())
36     return a1->get_parent();
37
38   // create an array with all ancestors of a1
39   std::vector<simgrid::instr::Container*> ancestors_a1;
40   for (auto* p = a1->get_parent(); p != nullptr; p = p->get_parent())
41     ancestors_a1.push_back(p);
42
43   // create an array with all ancestors of a2
44   std::vector<simgrid::instr::Container*> ancestors_a2;
45   for (auto* p = a2->get_parent(); p != nullptr; p = p->get_parent())
46     ancestors_a2.push_back(p);
47
48   // find the lowest ancestor
49   simgrid::instr::Container* p = nullptr;
50   int i = static_cast<int>(ancestors_a1.size()) - 1;
51   int j = static_cast<int>(ancestors_a2.size()) - 1;
52   while (i >= 0 && j >= 0) {
53     simgrid::instr::Container* a1p       = ancestors_a1.at(i);
54     const simgrid::instr::Container* a2p = ancestors_a2.at(j);
55     if (a1p == a2p) {
56       p = a1p;
57     } else {
58       break;
59     }
60     i--;
61     j--;
62   }
63   return p;
64 }
65
66 static void linkContainers(simgrid::instr::Container* src, simgrid::instr::Container* dst,
67                            std::set<std::string, std::less<>>* filter)
68 {
69   // ignore loopback
70   if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
71     XBT_DEBUG("  linkContainers: ignoring loopback link");
72     return;
73   }
74
75   // find common parent
76   simgrid::instr::Container* parent = lowestCommonAncestor(src, dst);
77   xbt_assert(parent, "common parent unknown, this is a tracing problem");
78
79   // check if we already register this pair (we only need one direction)
80   std::string aux1 = src->get_name() + dst->get_name();
81   std::string aux2 = dst->get_name() + src->get_name();
82   if (filter->find(aux1) != filter->end()) {
83     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
84     return;
85   }
86   if (filter->find(aux2) != filter->end()) {
87     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
88     return;
89   }
90
91   // ok, not found, register it
92   filter->insert(aux1);
93   filter->insert(aux2);
94
95   // declare type
96   std::string link_typename = parent->get_type()->get_name() + "-" + src->get_type()->get_name() +
97                               std::to_string(src->get_type()->get_id()) + "-" + dst->get_type()->get_name() +
98                               std::to_string(dst->get_type()->get_id());
99   simgrid::instr::LinkType* link =
100       parent->get_type()->by_name_or_create(link_typename, src->get_type(), dst->get_type());
101   link->set_calling_container(parent);
102
103   // create the link
104   static long long counter = 0;
105
106   std::string key = std::to_string(counter);
107   counter++;
108
109   link->start_event(src, "topology", key);
110   link->end_event(dst, "topology", key);
111
112   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
113 }
114
115 static void recursiveGraphExtraction(const simgrid::s4u::NetZone* netzone, const simgrid::instr::Container* container,
116                                      std::set<std::string, std::less<>>* filter)
117 {
118   if (not TRACE_platform_topology()) {
119     XBT_DEBUG("Graph extraction disabled by user.");
120     return;
121   }
122   XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
123
124   // bottom-up recursion
125   for (auto const& nz_son : netzone->get_children()) {
126     const simgrid::instr::Container* child_container = container->get_child_by_name(nz_son->get_name());
127     recursiveGraphExtraction(nz_son, child_container, filter);
128   }
129
130   auto* graph = xbt_graph_new_graph(0, nullptr);
131   std::map<std::string, xbt_node_t, std::less<>> nodes;
132   std::map<std::string, xbt_edge_t, std::less<>> edges;
133
134   netzone->get_impl()->get_graph(graph, &nodes, &edges);
135   for (auto const& elm : edges) {
136     const xbt_edge* edge = elm.second;
137     linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
138                    simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
139   }
140   xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
141 }
142
143 /*
144  * user categories support
145  */
146 static void recursiveNewVariableType(const std::string& new_typename, const std::string& color,
147                                      simgrid::instr::Type* root)
148 {
149   if (root->get_name() == "HOST" || root->get_name() == "VM")
150     root->by_name_or_create(std::string("p") + new_typename, color);
151
152   if (root->get_name() == "LINK")
153     root->by_name_or_create(std::string("b") + new_typename, color);
154
155   for (auto const& elm : root->get_children()) {
156     recursiveNewVariableType(new_typename, color, elm.second.get());
157   }
158 }
159
160 void instr_new_variable_type(const std::string& new_typename, const std::string& color)
161 {
162   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->get_type());
163 }
164
165 static void recursiveNewUserVariableType(const std::string& parent_type, const std::string& new_typename,
166                                          const std::string& color, simgrid::instr::Type* root)
167 {
168   if (root->get_name() == parent_type) {
169     root->by_name_or_create(new_typename, color);
170   }
171   for (auto const& elm : root->get_children())
172     recursiveNewUserVariableType(parent_type, new_typename, color, elm.second.get());
173 }
174
175 void instr_new_user_variable_type(const std::string& parent_type, const std::string& new_typename,
176                                   const std::string& color)
177 {
178   recursiveNewUserVariableType(parent_type, new_typename, color, simgrid::instr::Container::get_root()->get_type());
179 }
180
181 static void recursiveNewUserStateType(const std::string& parent_type, const std::string& new_typename,
182                                       simgrid::instr::Type* root)
183 {
184   if (root->get_name() == parent_type)
185     root->by_name_or_create<simgrid::instr::StateType>(new_typename);
186
187   for (auto const& elm : root->get_children())
188     recursiveNewUserStateType(parent_type, new_typename, elm.second.get());
189 }
190
191 void instr_new_user_state_type(const std::string& parent_type, const std::string& new_typename)
192 {
193   recursiveNewUserStateType(parent_type, new_typename, simgrid::instr::Container::get_root()->get_type());
194 }
195
196 static void recursiveNewValueForUserStateType(const std::string& type_name, const char* val, const std::string& color,
197                                               simgrid::instr::Type* root)
198 {
199   if (root->get_name() == type_name)
200     static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
201
202   for (auto const& elm : root->get_children())
203     recursiveNewValueForUserStateType(type_name, val, color, elm.second.get());
204 }
205
206 void instr_new_value_for_user_state_type(const std::string& type_name, const char* value, const std::string& color)
207 {
208   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->get_type());
209 }
210
211 namespace simgrid {
212 namespace instr {
213
214 void platform_graph_export_graphviz(const std::string& output_filename)
215 {
216   auto* g     = xbt_graph_new_graph(0, nullptr);
217   std::map<std::string, xbt_node_t, std::less<>> nodes;
218   std::map<std::string, xbt_edge_t, std::less<>> edges;
219   s4u::Engine::get_instance()->get_netzone_root()->extract_xbt_graph(g, &nodes, &edges);
220
221   std::ofstream fs;
222   fs.open(output_filename, std::ofstream::out);
223   xbt_assert(not fs.fail(), "Failed to open %s", output_filename.c_str());
224
225   if (g->directed)
226     fs << "digraph test {" << std::endl;
227   else
228     fs << "graph test {" << std::endl;
229
230   fs << "  graph [overlap=scale]" << std::endl;
231
232   fs << "  node [shape=box, style=filled]" << std::endl;
233   fs << "  node [width=.3, height=.3, style=filled, color=skyblue]" << std::endl << std::endl;
234
235   for (auto const& elm : nodes)
236     fs << "  \"" << elm.first << "\";" << std::endl;
237
238   for (auto const& elm : edges) {
239     const char* src_s = static_cast<char*>(elm.second->src->data);
240     const char* dst_s = static_cast<char*>(elm.second->dst->data);
241     if (g->directed)
242       fs << "  \"" << src_s << "\" -> \"" << dst_s << "\";" << std::endl;
243     else
244       fs << "  \"" << src_s << "\" -- \"" << dst_s << "\";" << std::endl;
245   }
246   fs << "}" << std::endl;
247   fs.close();
248
249   xbt_graph_free_graph(g, xbt_free_f, xbt_free_f, nullptr);
250 }
251
252 /* Callbacks */
253 static std::vector<NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
254 static void on_netzone_creation(s4u::NetZone const& netzone)
255 {
256   std::string id = netzone.get_name();
257   if (Container::get_root() == nullptr) {
258     auto* root = new NetZoneContainer(id, 0, nullptr);
259     xbt_assert(Container::get_root() == root);
260
261     if (TRACE_smpi_is_enabled()) {
262       auto* mpi = root->get_type()->by_name_or_create<ContainerType>("MPI");
263       if (not TRACE_smpi_is_grouped())
264         mpi->by_name_or_create<StateType>("MPI_STATE");
265       root->get_type()->by_name_or_create("MPI_LINK", mpi, mpi);
266       root->get_type()->by_name_or_create("MIGRATE_LINK", mpi, mpi);
267       mpi->by_name_or_create<StateType>("MIGRATE_STATE");
268     }
269
270     if (TRACE_needs_platform()) {
271       currentContainer.push_back(root);
272     }
273     return;
274   }
275
276   if (TRACE_needs_platform()) {
277     auto level      = static_cast<unsigned>(currentContainer.size());
278     auto* container = new NetZoneContainer(id, level, currentContainer.back());
279     currentContainer.push_back(container);
280   }
281 }
282
283 static void on_link_creation(s4u::Link const& link)
284 {
285   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
286     return;
287
288   auto* container = new Container(link.get_name(), "LINK", currentContainer.back());
289
290   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
291     VariableType* bandwidth = container->get_type()->by_name_or_create("bandwidth", "");
292     bandwidth->set_calling_container(container);
293     bandwidth->set_event(0, link.get_bandwidth());
294     VariableType* latency = container->get_type()->by_name_or_create("latency", "");
295     latency->set_calling_container(container);
296     latency->set_event(0, link.get_latency());
297   }
298
299   if (TRACE_uncategorized()) {
300     container->get_type()->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
301   }
302 }
303
304 static void on_host_creation(s4u::Host const& host)
305 {
306   if (Container::by_name_or_null(host.get_name())) // This host already exists, do nothing
307     return;
308
309   Container* container  = new HostContainer(host, currentContainer.back());
310   const Container* root = Container::get_root();
311
312   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
313     VariableType* speed = container->get_type()->by_name_or_create("speed", "");
314     speed->set_calling_container(container);
315     speed->set_event(0, host.get_speed());
316
317     VariableType* cores = container->get_type()->by_name_or_create("core_count", "");
318     cores->set_calling_container(container);
319     cores->set_event(0, host.get_core_count());
320   }
321
322   if (TRACE_uncategorized())
323     container->get_type()->by_name_or_create("speed_used", "0.5 0.5 0.5");
324
325   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
326     auto* mpi = container->get_type()->by_name_or_create<ContainerType>("MPI");
327     mpi->by_name_or_create<StateType>("MPI_STATE");
328     root->get_type()->by_name_or_create("MIGRATE_LINK", mpi, mpi);
329     mpi->by_name_or_create<StateType>("MIGRATE_STATE");
330   }
331 }
332
333 static void on_action_state_change(kernel::resource::Action const& action,
334                                    kernel::resource::Action::State /* previous */)
335 {
336   auto n = static_cast<unsigned>(action.get_variable()->get_number_of_constraint());
337
338   for (unsigned i = 0; i < n; i++) {
339     double value = action.get_rate() * action.get_variable()->get_constraint_weight(i);
340     /* Beware of composite actions: ptasks put links and cpus together. Extra pb: we cannot dynamic_cast from void* */
341     kernel::resource::Resource* resource = action.get_variable()->get_constraint(i)->get_id();
342     const kernel::resource::CpuImpl* cpu = dynamic_cast<kernel::resource::CpuImpl*>(resource);
343
344     if (cpu != nullptr)
345       resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action.get_category(), value,
346                                action.get_last_update(), simgrid_get_clock() - action.get_last_update());
347
348     const kernel::resource::LinkImpl* link = dynamic_cast<kernel::resource::LinkImpl*>(resource);
349
350     if (link != nullptr)
351       resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action.get_category(), value,
352                                action.get_last_update(), simgrid_get_clock() - action.get_last_update());
353   }
354 }
355
356 static void on_platform_created()
357 {
358   currentContainer.clear();
359   std::set<std::string, std::less<>> filter;
360   XBT_DEBUG("Starting graph extraction.");
361   recursiveGraphExtraction(s4u::Engine::get_instance()->get_netzone_root(), Container::get_root(), &filter);
362   XBT_DEBUG("Graph extraction finished.");
363   dump_buffer(true);
364 }
365
366 static void on_actor_creation(s4u::Actor const& actor)
367 {
368   const Container* root      = Container::get_root();
369   Container* container       = Container::by_name(actor.get_host()->get_name());
370   std::string container_name = instr_pid(actor);
371
372   container->create_child(container_name, "ACTOR");
373   auto* actor_type = container->get_type()->by_name_or_create<ContainerType>("ACTOR");
374   auto* state      = actor_type->by_name_or_create<StateType>("ACTOR_STATE");
375   state->add_entity_value("suspend", "1 0 1");
376   state->add_entity_value("sleep", "1 1 0");
377   state->add_entity_value("receive", "1 0 0");
378   state->add_entity_value("send", "0 0 1");
379   state->add_entity_value("execute", "0 1 1");
380   root->get_type()->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
381
382   actor.on_exit([container_name](bool failed) {
383     if (failed)
384       // kill means that this actor no longer exists, let's destroy it
385       Container::by_name(container_name)->remove_from_parent();
386   });
387 }
388
389 static void on_actor_host_change(s4u::Actor const& actor, s4u::Host const& /*previous_location*/)
390 {
391   static long long int counter = 0;
392   Container* container         = Container::by_name(instr_pid(actor));
393   LinkType* link               = Container::get_root()->get_link("ACTOR_LINK");
394
395   // start link
396   link->start_event(container, "M", std::to_string(counter));
397   // destroy existing container of this process
398   container->remove_from_parent();
399   // create new container on the new_host location
400   Container::by_name(actor.get_host()->get_name())->create_child(instr_pid(actor), "ACTOR");
401   // end link
402   link->end_event(Container::by_name(instr_pid(actor)), "M", std::to_string(counter));
403   counter++;
404 }
405
406 static void on_vm_creation(s4u::Host const& host)
407 {
408   const Container* container = new HostContainer(host, currentContainer.back());
409   const Container* root      = Container::get_root();
410   auto* vm                   = container->get_type()->by_name_or_create<ContainerType>("VM");
411   auto* state                = vm->by_name_or_create<StateType>("VM_STATE");
412   state->add_entity_value("suspend", "1 0 1");
413   state->add_entity_value("sleep", "1 1 0");
414   state->add_entity_value("receive", "1 0 0");
415   state->add_entity_value("send", "0 0 1");
416   state->add_entity_value("execute", "0 1 1");
417   root->get_type()->by_name_or_create("VM_LINK", vm, vm);
418   root->get_type()->by_name_or_create("VM_ACTOR_LINK", vm, vm);
419 }
420
421 void define_callbacks()
422 {
423   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
424   // properly
425   if (TRACE_needs_platform()) {
426     s4u::Engine::on_platform_created.connect(on_platform_created);
427     s4u::Host::on_creation.connect(on_host_creation);
428     s4u::Host::on_speed_change.connect([](s4u::Host const& host) {
429       Container::by_name(host.get_name())
430           ->get_variable("speed")
431           ->set_event(simgrid_get_clock(), host.get_core_count() * host.get_available_speed());
432     });
433     s4u::Link::on_creation.connect(on_link_creation);
434     s4u::Link::on_bandwidth_change.connect([](s4u::Link const& link) {
435       Container::by_name(link.get_name())
436           ->get_variable("bandwidth")
437           ->set_event(simgrid_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
438     });
439     s4u::NetZone::on_seal.connect([](s4u::NetZone const& /*netzone*/) { currentContainer.pop_back(); });
440     kernel::routing::NetPoint::on_creation.connect([](kernel::routing::NetPoint const& netpoint) {
441       if (netpoint.is_router())
442         new RouterContainer(netpoint.get_name(), currentContainer.back());
443     });
444   }
445
446   s4u::NetZone::on_creation.connect(on_netzone_creation);
447
448   kernel::resource::CpuAction::on_state_change.connect(on_action_state_change);
449   s4u::Link::on_communication_state_change.connect(on_action_state_change);
450
451   if (TRACE_actor_is_enabled()) {
452     s4u::Actor::on_creation.connect(on_actor_creation);
453     s4u::Actor::on_destruction.connect([](s4u::Actor const& actor) {
454       auto container = Container::by_name_or_null(instr_pid(actor));
455       if (container != nullptr)
456         container->remove_from_parent();
457     });
458     s4u::Actor::on_suspend.connect([](s4u::Actor const& actor) {
459       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("suspend");
460     });
461     s4u::Actor::on_resume.connect(
462         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
463     s4u::Actor::on_sleep.connect([](s4u::Actor const& actor) {
464       Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->push_event("sleep");
465     });
466     s4u::Actor::on_wake_up.connect(
467         [](s4u::Actor const& actor) { Container::by_name(instr_pid(actor))->get_state("ACTOR_STATE")->pop_event(); });
468     s4u::Exec::on_start.connect([](s4u::Exec const&) {
469       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->push_event("execute");
470     });
471     s4u::Exec::on_completion.connect([](s4u::Exec const&) {
472       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->pop_event();
473     });
474     s4u::Comm::on_start.connect([](s4u::Comm const&, bool is_sender) {
475       Container::by_name(instr_pid(*s4u::Actor::self()))
476           ->get_state("ACTOR_STATE")
477           ->push_event(is_sender ? "send" : "receive");
478     });
479     s4u::Comm::on_completion.connect([](s4u::Comm const&) {
480       Container::by_name(instr_pid(*s4u::Actor::self()))->get_state("ACTOR_STATE")->pop_event();
481     });
482     s4u::Actor::on_host_change.connect(on_actor_host_change);
483   }
484
485   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_computing()) {
486     s4u::Exec::on_start.connect([](s4u::Exec const& exec) {
487       Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
488           ->get_state("MPI_STATE")
489           ->push_event("computing", new CpuTIData("compute", exec.get_cost()));
490     });
491     s4u::Exec::on_completion.connect([](s4u::Exec const&) {
492       Container::by_name(std::string("rank-") + std::to_string(s4u::Actor::self()->get_pid()))
493           ->get_state("MPI_STATE")
494           ->pop_event();
495     });
496   }
497
498   if (TRACE_vm_is_enabled()) {
499     s4u::Host::on_creation.connect(on_vm_creation);
500     s4u::VirtualMachine::on_start.connect([](s4u::VirtualMachine const& vm) {
501       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
502     });
503     s4u::VirtualMachine::on_started.connect(
504         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
505     s4u::VirtualMachine::on_suspend.connect([](s4u::VirtualMachine const& vm) {
506       Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
507     });
508     s4u::VirtualMachine::on_resume.connect(
509         [](s4u::VirtualMachine const& vm) { Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event(); });
510     s4u::Host::on_destruction.connect(
511         [](s4u::Host const& host) { Container::by_name(host.get_name())->remove_from_parent(); });
512   }
513 }
514 } // namespace instr
515 } // namespace simgrid