Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Tracing: rename power to speed, and add a core_amount variable
[simgrid.git] / src / instr / instr_platform.cpp
1 /* Copyright (c) 2010-2018. 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 "src/instr/instr_private.hpp"
7
8 #include "simgrid/kernel/routing/NetPoint.hpp"
9 #include "simgrid/kernel/routing/NetZoneImpl.hpp"
10 #include "simgrid/s4u/Actor.hpp"
11 #include "simgrid/s4u/Engine.hpp"
12 #include "simgrid/s4u/Host.hpp"
13 #include "simgrid/s4u/VirtualMachine.hpp"
14 #include "src/surf/cpu_interface.hpp"
15 #include "src/surf/network_interface.hpp"
16 #include "src/surf/surf_interface.hpp"
17 #include "src/surf/xml/platf_private.hpp"
18 #include "surf/surf.hpp"
19 #include "xbt/graph.h"
20
21 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
22
23 static std::vector<simgrid::instr::NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
24
25 static const char* instr_node_name(xbt_node_t node)
26 {
27   return static_cast<char*>(xbt_graph_node_get_data(node));
28 }
29
30 static container_t lowestCommonAncestor(container_t a1, container_t a2)
31 {
32   // this is only an optimization (since most of a1 and a2 share the same parent)
33   if (a1->father_ == a2->father_)
34     return a1->father_;
35
36   // create an array with all ancestors of a1
37   std::vector<container_t> ancestors_a1;
38   container_t p = a1->father_;
39   while (p) {
40     ancestors_a1.push_back(p);
41     p = p->father_;
42   }
43
44   // create an array with all ancestors of a2
45   std::vector<container_t> ancestors_a2;
46   p = a2->father_;
47   while (p) {
48     ancestors_a2.push_back(p);
49     p = p->father_;
50   }
51
52   // find the lowest ancestor
53   p     = nullptr;
54   int i = ancestors_a1.size() - 1;
55   int j = ancestors_a2.size() - 1;
56   while (i >= 0 && j >= 0) {
57     container_t a1p = ancestors_a1.at(i);
58     container_t a2p = ancestors_a2.at(j);
59     if (a1p == a2p) {
60       p = a1p;
61     } else {
62       break;
63     }
64     i--;
65     j--;
66   }
67   return p;
68 }
69
70 static void linkContainers(container_t src, container_t dst, std::set<std::string>* filter)
71 {
72   // ignore loopback
73   if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
74     XBT_DEBUG("  linkContainers: ignoring loopback link");
75     return;
76   }
77
78   // find common father
79   container_t father = lowestCommonAncestor(src, dst);
80   if (not father) {
81     xbt_die("common father unknown, this is a tracing problem");
82   }
83
84   // check if we already register this pair (we only need one direction)
85   std::string aux1 = src->get_name() + dst->get_name();
86   std::string aux2 = dst->get_name() + src->get_name();
87   if (filter->find(aux1) != filter->end()) {
88     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
89     return;
90   }
91   if (filter->find(aux2) != filter->end()) {
92     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
93     return;
94   }
95
96   // ok, not found, register it
97   filter->insert(aux1);
98   filter->insert(aux2);
99
100   // declare type
101   std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
102                               std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
103                               std::to_string(dst->type_->get_id());
104   simgrid::instr::LinkType* link = father->type_->by_name_or_create(link_typename, src->type_, dst->type_);
105   link->set_calling_container(father);
106
107   // register EDGE types for triva configuration
108   trivaEdgeTypes.insert(link->get_name());
109
110   // create the link
111   static long long counter = 0;
112
113   std::string key = std::to_string(counter);
114   counter++;
115
116   link->start_event(src, "topology", key);
117   link->end_event(dst, "topology", key);
118
119   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
120 }
121
122 static void recursiveGraphExtraction(simgrid::s4u::NetZone* netzone, container_t container,
123                                      std::set<std::string>* filter)
124 {
125   if (not TRACE_platform_topology()) {
126     XBT_DEBUG("Graph extraction disabled by user.");
127     return;
128   }
129   XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
130   if (not netzone->get_children().empty()) {
131     // bottom-up recursion
132     for (auto const& nz_son : netzone->get_children()) {
133       container_t child_container = container->children_.at(nz_son->get_cname());
134       recursiveGraphExtraction(nz_son, child_container, filter);
135     }
136   }
137
138   xbt_graph_t graph                        = xbt_graph_new_graph(0, nullptr);
139   std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
140   std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
141
142   netzone->get_impl()->get_graph(graph, nodes, edges);
143   for (auto elm : *edges) {
144     xbt_edge_t edge = elm.second;
145     linkContainers(simgrid::instr::Container::by_name(static_cast<const char*>(edge->src->data)),
146                    simgrid::instr::Container::by_name(static_cast<const char*>(edge->dst->data)), filter);
147   }
148   delete nodes;
149   delete edges;
150   xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
151 }
152
153 /*
154  * Callbacks
155  */
156 static void instr_netzone_on_creation(simgrid::s4u::NetZone& netzone)
157 {
158   std::string id = netzone.get_name();
159   if (simgrid::instr::Container::get_root() == nullptr) {
160     simgrid::instr::NetZoneContainer* root = new simgrid::instr::NetZoneContainer(id, 0, nullptr);
161
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");
170     }
171
172     if (TRACE_needs_platform()) {
173       currentContainer.push_back(root);
174     }
175     return;
176   }
177
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);
182   }
183 }
184
185 static void instr_link_on_creation(simgrid::s4u::Link& link)
186 {
187   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
188     return;
189
190   container_t container = new simgrid::instr::Container(link.get_name(), "LINK", currentContainer.back());
191
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());
199   }
200   if (TRACE_uncategorized()) {
201     container->type_->by_name_or_create("bandwidth_used", "0.5 0.5 0.5");
202   }
203 }
204
205 static void instr_host_on_creation(simgrid::s4u::Host& host)
206 {
207   container_t container = new simgrid::instr::HostContainer(host, currentContainer.back());
208   container_t root      = simgrid::instr::Container::get_root();
209
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());
214
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());
218   }
219
220   if (TRACE_uncategorized())
221     container->type_->by_name_or_create("speed_used", "0.5 0.5 0.5");
222
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");
229   }
230 }
231
232 static void instr_host_on_speed_change(simgrid::s4u::Host& host)
233 {
234   simgrid::instr::Container::by_name(host.get_cname())
235       ->get_variable("speed")
236       ->set_event(surf_get_clock(), host.get_core_count() * host.get_available_speed());
237 }
238
239 static void instr_action_on_state_change(simgrid::kernel::resource::Action* action,
240                                          simgrid::kernel::resource::Action::State /* previous */)
241 {
242   int n = action->get_variable()->get_number_of_constraint();
243
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 =
248         static_cast<simgrid::kernel::resource::Resource*>(action->get_variable()->get_constraint(i)->get_id());
249     simgrid::surf::Cpu* cpu = dynamic_cast<simgrid::surf::Cpu*>(resource);
250
251     if (cpu != nullptr)
252       TRACE_surf_resource_set_utilization("HOST", "speed_used", cpu->get_cname(), action->get_category(), value,
253                                           action->get_last_update(), SIMIX_get_clock() - action->get_last_update());
254
255     simgrid::kernel::resource::LinkImpl* link = dynamic_cast<simgrid::kernel::resource::LinkImpl*>(resource);
256
257     if (link != nullptr)
258       TRACE_surf_resource_set_utilization("LINK", "bandwidth_used", link->get_cname(), action->get_category(), value,
259                                           action->get_last_update(), SIMIX_get_clock() - action->get_last_update());
260   }
261 }
262
263 static void instr_link_on_bandwidth_change(simgrid::s4u::Link& link)
264 {
265   simgrid::instr::Container::by_name(link.get_cname())
266       ->get_variable("bandwidth")
267       ->set_event(surf_get_clock(), sg_bandwidth_factor * link.get_bandwidth());
268 }
269
270 static void instr_netpoint_on_creation(simgrid::kernel::routing::NetPoint* netpoint)
271 {
272   if (netpoint->is_router())
273     new simgrid::instr::RouterContainer(netpoint->get_cname(), currentContainer.back());
274 }
275
276 static void instr_on_platform_created()
277 {
278   currentContainer.clear();
279   std::set<std::string>* filter = new std::set<std::string>;
280   XBT_DEBUG("Starting graph extraction.");
281   recursiveGraphExtraction(simgrid::s4u::Engine::get_instance()->get_netzone_root(),
282                            simgrid::instr::Container::get_root(), filter);
283   XBT_DEBUG("Graph extraction finished.");
284   delete filter;
285   TRACE_paje_dump_buffer(true);
286 }
287
288 static void instr_actor_on_creation(simgrid::s4u::ActorPtr actor)
289 {
290   container_t root      = simgrid::instr::Container::get_root();
291   container_t container = simgrid::instr::Container::by_name(actor->get_host()->get_name());
292
293   container->create_child(instr_pid(actor.get()), "ACTOR");
294   simgrid::instr::ContainerType* actor_type =
295       container->type_->by_name_or_create<simgrid::instr::ContainerType>("ACTOR");
296   simgrid::instr::StateType* state = actor_type->by_name_or_create<simgrid::instr::StateType>("ACTOR_STATE");
297   state->add_entity_value("suspend", "1 0 1");
298   state->add_entity_value("sleep", "1 1 0");
299   state->add_entity_value("receive", "1 0 0");
300   state->add_entity_value("send", "0 0 1");
301   state->add_entity_value("task_execute", "0 1 1");
302   root->type_->by_name_or_create("ACTOR_LINK", actor_type, actor_type);
303   root->type_->by_name_or_create("ACTOR_TASK_LINK", actor_type, actor_type);
304
305   std::string container_name = instr_pid(actor.get());
306   actor->on_exit(
307       [container_name](int status, void* actor) {
308         if (status == SMX_EXIT_FAILURE)
309           // kill means that this actor no longer exists, let's destroy it
310           simgrid::instr::Container::by_name(container_name)->remove_from_parent();
311       },
312       actor->get_impl());
313 }
314
315 static long long int counter = 0;
316
317 static void instr_actor_on_migration_start(simgrid::s4u::ActorPtr actor)
318 {
319   // start link
320   container_t container = simgrid::instr::Container::by_name(instr_pid(actor.get()));
321   simgrid::instr::Container::get_root()->get_link("ACTOR_LINK")->start_event(container, "M", std::to_string(counter));
322
323   // destroy existing container of this process
324   container->remove_from_parent();
325 }
326
327 static void instr_actor_on_migration_end(simgrid::s4u::ActorPtr actor)
328 {
329   // create new container on the new_host location
330   simgrid::instr::Container::by_name(actor->get_host()->get_name())->create_child(instr_pid(actor.get()), "ACTOR");
331   // end link
332   simgrid::instr::Container::get_root()
333       ->get_link("ACTOR_LINK")
334       ->end_event(simgrid::instr::Container::by_name(instr_pid(actor.get())), "M", std::to_string(counter));
335   counter++;
336 }
337
338 static void instr_vm_on_creation(simgrid::s4u::Host& host)
339 {
340   container_t container             = new simgrid::instr::HostContainer(host, currentContainer.back());
341   container_t root                  = simgrid::instr::Container::get_root();
342   simgrid::instr::ContainerType* vm = container->type_->by_name_or_create<simgrid::instr::ContainerType>("VM");
343   simgrid::instr::StateType* state  = vm->by_name_or_create<simgrid::instr::StateType>("VM_STATE");
344   state->add_entity_value("suspend", "1 0 1");
345   state->add_entity_value("sleep", "1 1 0");
346   state->add_entity_value("receive", "1 0 0");
347   state->add_entity_value("send", "0 0 1");
348   state->add_entity_value("task_execute", "0 1 1");
349   root->type_->by_name_or_create("VM_LINK", vm, vm);
350   root->type_->by_name_or_create("VM_ACTOR_LINK", vm, vm);
351 }
352
353 void instr_define_callbacks()
354 {
355   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
356   // properly
357   if (TRACE_needs_platform()) {
358     simgrid::s4u::on_platform_created.connect(instr_on_platform_created);
359     simgrid::s4u::Host::on_creation.connect(instr_host_on_creation);
360     simgrid::s4u::Host::on_speed_change.connect(instr_host_on_speed_change);
361     simgrid::s4u::Link::on_creation.connect(instr_link_on_creation);
362     simgrid::s4u::Link::on_bandwidth_change.connect(instr_link_on_bandwidth_change);
363     simgrid::s4u::NetZone::on_seal.connect([](simgrid::s4u::NetZone& /*netzone*/) { currentContainer.pop_back(); });
364     simgrid::kernel::routing::NetPoint::on_creation.connect(instr_netpoint_on_creation);
365   }
366   simgrid::s4u::NetZone::on_creation.connect(instr_netzone_on_creation);
367
368   simgrid::surf::CpuAction::on_state_change.connect(instr_action_on_state_change);
369   simgrid::s4u::Link::on_communication_state_change.connect(instr_action_on_state_change);
370
371   if (TRACE_actor_is_enabled()) {
372     simgrid::s4u::Actor::on_creation.connect(instr_actor_on_creation);
373     simgrid::s4u::Actor::on_suspend.connect([](simgrid::s4u::ActorPtr actor) {
374       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->push_event("suspend");
375     });
376     simgrid::s4u::Actor::on_resume.connect([](simgrid::s4u::ActorPtr actor) {
377       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->pop_event();
378     });
379     simgrid::s4u::Actor::on_sleep.connect([](simgrid::s4u::ActorPtr actor) {
380       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->push_event("sleep");
381     });
382     simgrid::s4u::Actor::on_wake_up.connect([](simgrid::s4u::ActorPtr actor) {
383       simgrid::instr::Container::by_name(instr_pid(actor.get()))->get_state("ACTOR_STATE")->pop_event();
384     });
385     simgrid::s4u::Actor::on_migration_start.connect(instr_actor_on_migration_start);
386     simgrid::s4u::Actor::on_migration_end.connect(instr_actor_on_migration_end);
387   }
388
389   if (TRACE_vm_is_enabled()) {
390     simgrid::s4u::Host::on_creation.connect(instr_vm_on_creation);
391     simgrid::s4u::VirtualMachine::on_start.connect([](simgrid::s4u::VirtualMachine& vm) {
392       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("start");
393     });
394     simgrid::s4u::VirtualMachine::on_started.connect([](simgrid::s4u::VirtualMachine& vm) {
395       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
396     });
397     simgrid::s4u::VirtualMachine::on_suspend.connect([](simgrid::s4u::VirtualMachine& vm) {
398       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->push_event("suspend");
399     });
400     simgrid::s4u::VirtualMachine::on_resume.connect([](simgrid::s4u::VirtualMachine& vm) {
401       simgrid::instr::Container::by_name(vm.get_name())->get_state("VM_STATE")->pop_event();
402     });
403     simgrid::s4u::Host::on_destruction.connect(
404         [](simgrid::s4u::Host& host) { simgrid::instr::Container::by_name(host.get_name())->remove_from_parent(); });
405   }
406 }
407 /*
408  * user categories support
409  */
410 static void recursiveNewVariableType(std::string new_typename, std::string color, simgrid::instr::Type* root)
411 {
412   if (root->get_name() == "HOST" || root->get_name() == "VM")
413     root->by_name_or_create(std::string("p") + new_typename, color);
414
415   if (root->get_name() == "LINK")
416     root->by_name_or_create(std::string("b") + new_typename, color);
417
418   for (auto elm : root->children_) {
419     recursiveNewVariableType(new_typename, color, elm.second);
420   }
421 }
422
423 void instr_new_variable_type(std::string new_typename, std::string color)
424 {
425   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::get_root()->type_);
426 }
427
428 static void recursiveNewUserVariableType(std::string father_type, std::string new_typename, std::string color,
429                                          simgrid::instr::Type* root)
430 {
431   if (root->get_name() == father_type) {
432     root->by_name_or_create(new_typename, color);
433   }
434   for (auto elm : root->children_)
435     recursiveNewUserVariableType(father_type, new_typename, color, elm.second);
436 }
437
438 void instr_new_user_variable_type(std::string father_type, std::string new_typename, std::string color)
439 {
440   recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::get_root()->type_);
441 }
442
443 static void recursiveNewUserStateType(std::string father_type, std::string new_typename, simgrid::instr::Type* root)
444 {
445   if (root->get_name() == father_type)
446     root->by_name_or_create<simgrid::instr::StateType>(new_typename);
447
448   for (auto elm : root->children_)
449     recursiveNewUserStateType(father_type, new_typename, elm.second);
450 }
451
452 void instr_new_user_state_type(std::string father_type, std::string new_typename)
453 {
454   recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::get_root()->type_);
455 }
456
457 static void recursiveNewValueForUserStateType(std::string type_name, const char* val, std::string color,
458                                               simgrid::instr::Type* root)
459 {
460   if (root->get_name() == type_name)
461     static_cast<simgrid::instr::StateType*>(root)->add_entity_value(val, color);
462
463   for (auto elm : root->children_)
464     recursiveNewValueForUserStateType(type_name, val, color, elm.second);
465 }
466
467 void instr_new_value_for_user_state_type(std::string type_name, const char* value, std::string color)
468 {
469   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::get_root()->type_);
470 }
471
472 #define GRAPHICATOR_SUPPORT_FUNCTIONS
473
474 static void recursiveXBTGraphExtraction(xbt_graph_t graph, std::map<std::string, xbt_node_t>* nodes,
475                                         std::map<std::string, xbt_edge_t>* edges, sg_netzone_t netzone,
476                                         container_t container)
477 {
478   if (not netzone->get_children().empty()) {
479     // bottom-up recursion
480     for (auto const& netzone_child : netzone->get_children()) {
481       container_t child_container = container->children_.at(netzone_child->get_cname());
482       recursiveXBTGraphExtraction(graph, nodes, edges, netzone_child, child_container);
483     }
484   }
485
486   netzone->get_impl()->get_graph(graph, nodes, edges);
487 }
488
489 xbt_graph_t instr_routing_platform_graph()
490 {
491   xbt_graph_t ret                          = xbt_graph_new_graph(0, nullptr);
492   std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
493   std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
494   recursiveXBTGraphExtraction(ret, nodes, edges, simgrid::s4u::Engine::get_instance()->get_netzone_root(),
495                               simgrid::instr::Container::get_root());
496   delete nodes;
497   delete edges;
498   return ret;
499 }
500
501 void instr_routing_platform_graph_export_graphviz(xbt_graph_t g, const char* filename)
502 {
503   unsigned int cursor = 0;
504   xbt_node_t node     = nullptr;
505   xbt_edge_t edge     = nullptr;
506
507   FILE* file = fopen(filename, "w");
508   xbt_assert(file, "Failed to open %s \n", filename);
509
510   if (g->directed)
511     fprintf(file, "digraph test {\n");
512   else
513     fprintf(file, "graph test {\n");
514
515   fprintf(file, "  graph [overlap=scale]\n");
516
517   fprintf(file, "  node [shape=box, style=filled]\n");
518   fprintf(file, "  node [width=.3, height=.3, style=filled, color=skyblue]\n\n");
519
520   xbt_dynar_foreach (g->nodes, cursor, node) {
521     fprintf(file, "  \"%s\";\n", instr_node_name(node));
522   }
523   xbt_dynar_foreach (g->edges, cursor, edge) {
524     const char* src_s = instr_node_name(edge->src);
525     const char* dst_s = instr_node_name(edge->dst);
526     if (g->directed)
527       fprintf(file, "  \"%s\" -> \"%s\";\n", src_s, dst_s);
528     else
529       fprintf(file, "  \"%s\" -- \"%s\";\n", src_s, dst_s);
530   }
531   fprintf(file, "}\n");
532   fclose(file);
533 }