Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
use signals to instrument VMs
[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/network_interface.hpp"
15 #include "src/surf/xml/platf_private.hpp"
16 #include "surf/surf.hpp"
17 #include "xbt/graph.h"
18
19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(instr_routing, instr, "Tracing platform hierarchy");
20
21 static std::vector<simgrid::instr::NetZoneContainer*> currentContainer; /* push and pop, used only in creation */
22
23 static const char* instr_node_name(xbt_node_t node)
24 {
25   return static_cast<char*>(xbt_graph_node_get_data(node));
26 }
27
28 static container_t lowestCommonAncestor(container_t a1, container_t a2)
29 {
30   // this is only an optimization (since most of a1 and a2 share the same parent)
31   if (a1->father_ == a2->father_)
32     return a1->father_;
33
34   // create an array with all ancestors of a1
35   std::vector<container_t> ancestors_a1;
36   container_t p = a1->father_;
37   while (p) {
38     ancestors_a1.push_back(p);
39     p = p->father_;
40   }
41
42   // create an array with all ancestors of a2
43   std::vector<container_t> ancestors_a2;
44   p = a2->father_;
45   while (p) {
46     ancestors_a2.push_back(p);
47     p = p->father_;
48   }
49
50   // find the lowest ancestor
51   p     = nullptr;
52   int i = ancestors_a1.size() - 1;
53   int j = ancestors_a2.size() - 1;
54   while (i >= 0 && j >= 0) {
55     container_t a1p = ancestors_a1.at(i);
56     container_t a2p = ancestors_a2.at(j);
57     if (a1p == a2p) {
58       p = a1p;
59     } else {
60       break;
61     }
62     i--;
63     j--;
64   }
65   return p;
66 }
67
68 static void linkContainers(container_t src, container_t dst, std::set<std::string>* filter)
69 {
70   // ignore loopback
71   if (src->get_name() == "__loopback__" || dst->get_name() == "__loopback__") {
72     XBT_DEBUG("  linkContainers: ignoring loopback link");
73     return;
74   }
75
76   // find common father
77   container_t father = lowestCommonAncestor(src, dst);
78   if (not father) {
79     xbt_die("common father unknown, this is a tracing problem");
80   }
81
82   // check if we already register this pair (we only need one direction)
83   std::string aux1 = src->get_name() + dst->get_name();
84   std::string aux2 = dst->get_name() + src->get_name();
85   if (filter->find(aux1) != filter->end()) {
86     XBT_DEBUG("  linkContainers: already registered %s <-> %s (1)", src->get_cname(), dst->get_cname());
87     return;
88   }
89   if (filter->find(aux2) != filter->end()) {
90     XBT_DEBUG("  linkContainers: already registered %s <-> %s (2)", dst->get_cname(), src->get_cname());
91     return;
92   }
93
94   // ok, not found, register it
95   filter->insert(aux1);
96   filter->insert(aux2);
97
98   // declare type
99   std::string link_typename = father->type_->get_name() + "-" + src->type_->get_name() +
100                               std::to_string(src->type_->get_id()) + "-" + dst->type_->get_name() +
101                               std::to_string(dst->type_->get_id());
102   simgrid::instr::LinkType* link = father->type_->getOrCreateLinkType(link_typename, src->type_, dst->type_);
103   link->setCallingContainer(father);
104
105   // register EDGE types for triva configuration
106   trivaEdgeTypes.insert(link->get_name());
107
108   // create the link
109   static long long counter = 0;
110
111   std::string key = std::to_string(counter);
112   counter++;
113
114   link->startEvent(src, "topology", key);
115   link->endEvent(dst, "topology", key);
116
117   XBT_DEBUG("  linkContainers %s <-> %s", src->get_cname(), dst->get_cname());
118 }
119
120 static void recursiveGraphExtraction(simgrid::s4u::NetZone* netzone, container_t container,
121                                      std::set<std::string>* filter)
122 {
123   if (not TRACE_platform_topology()) {
124     XBT_DEBUG("Graph extraction disabled by user.");
125     return;
126   }
127   XBT_DEBUG("Graph extraction for NetZone = %s", netzone->get_cname());
128   if (not netzone->getChildren()->empty()) {
129     // bottom-up recursion
130     for (auto const& nz_son : *netzone->getChildren()) {
131       container_t child_container = container->children_.at(nz_son->get_cname());
132       recursiveGraphExtraction(nz_son, child_container, filter);
133     }
134   }
135
136   xbt_graph_t graph                        = xbt_graph_new_graph(0, nullptr);
137   std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
138   std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
139
140   static_cast<simgrid::kernel::routing::NetZoneImpl*>(netzone)->get_graph(graph, nodes, edges);
141   for (auto elm : *edges) {
142     xbt_edge_t edge = elm.second;
143     linkContainers(simgrid::instr::Container::byName(static_cast<const char*>(edge->src->data)),
144                    simgrid::instr::Container::byName(static_cast<const char*>(edge->dst->data)), filter);
145   }
146   delete nodes;
147   delete edges;
148   xbt_graph_free_graph(graph, xbt_free_f, xbt_free_f, nullptr);
149 }
150
151 /*
152  * Callbacks
153  */
154 static void instr_netzone_on_creation(simgrid::s4u::NetZone& netzone)
155 {
156   std::string id = netzone.get_name();
157   if (simgrid::instr::Container::getRoot() == nullptr) {
158     simgrid::instr::NetZoneContainer* root = new simgrid::instr::NetZoneContainer(id, 0, nullptr);
159
160     if (TRACE_smpi_is_enabled()) {
161       simgrid::instr::Type* mpi = root->type_->getOrCreateContainerType("MPI");
162       if (not TRACE_smpi_is_grouped())
163         mpi->getOrCreateStateType("MPI_STATE");
164       root->type_->getOrCreateLinkType("MPI_LINK", mpi, mpi);
165       // TODO See if we can move this to the LoadBalancer plugin
166       root->type_->getOrCreateLinkType("MIGRATE_LINK", mpi, mpi);
167       mpi->getOrCreateStateType("MIGRATE_STATE");
168     }
169
170     if (TRACE_needs_platform()) {
171       currentContainer.push_back(root);
172     }
173     return;
174   }
175
176   if (TRACE_needs_platform()) {
177     simgrid::instr::NetZoneContainer* container =
178         new simgrid::instr::NetZoneContainer(id, currentContainer.size(), currentContainer.back());
179     currentContainer.push_back(container);
180   }
181 }
182
183 static void instr_netzone_on_seal(simgrid::s4u::NetZone& /*netzone*/)
184 {
185   if (TRACE_needs_platform()) {
186     currentContainer.pop_back();
187   }
188 }
189
190 static void instr_link_on_creation(simgrid::s4u::Link& link)
191 {
192   if (currentContainer.empty()) // No ongoing parsing. Are you creating the loopback?
193     return;
194
195   container_t father    = currentContainer.back();
196   container_t container = new simgrid::instr::Container(link.get_name(), "LINK", father);
197
198   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_link())) {
199     simgrid::instr::VariableType* bandwidth = container->type_->getOrCreateVariableType("bandwidth", "");
200     bandwidth->setCallingContainer(container);
201     bandwidth->setEvent(0, link.bandwidth());
202     simgrid::instr::VariableType* latency = container->type_->getOrCreateVariableType("latency", "");
203     latency->setCallingContainer(container);
204     latency->setEvent(0, link.latency());
205   }
206   if (TRACE_uncategorized()) {
207     container->type_->getOrCreateVariableType("bandwidth_used", "0.5 0.5 0.5");
208   }
209 }
210
211 static void instr_host_on_creation(simgrid::s4u::Host& host)
212 {
213   container_t container = new simgrid::instr::HostContainer(host, currentContainer.back());
214   container_t root      = simgrid::instr::Container::getRoot();
215
216   if ((TRACE_categorized() || TRACE_uncategorized() || TRACE_platform()) && (not TRACE_disable_speed())) {
217     simgrid::instr::VariableType* power = container->type_->getOrCreateVariableType("power", "");
218     power->setCallingContainer(container);
219     power->setEvent(0, host.getSpeed());
220   }
221
222   if (TRACE_uncategorized())
223     container->type_->getOrCreateVariableType("power_used", "0.5 0.5 0.5");
224
225   if (TRACE_smpi_is_enabled() && TRACE_smpi_is_grouped()) {
226     simgrid::instr::ContainerType* mpi = container->type_->getOrCreateContainerType("MPI");
227     mpi->getOrCreateStateType("MPI_STATE");
228     // TODO See if we can move this to the LoadBalancer plugin
229     root->type_->getOrCreateLinkType("MIGRATE_LINK", mpi, mpi);
230     mpi->getOrCreateStateType("MIGRATE_STATE");
231   }
232 }
233
234 static void instr_netpoint_on_creation(simgrid::kernel::routing::NetPoint* netpoint)
235 {
236   if (netpoint->is_router() && TRACE_needs_platform() && TRACE_is_enabled())
237     new simgrid::instr::RouterContainer(netpoint->get_cname(), currentContainer.back());
238 }
239
240 static void instr_on_platform_created()
241 {
242   currentContainer.clear();
243   std::set<std::string>* filter = new std::set<std::string>;
244   XBT_DEBUG("Starting graph extraction.");
245   recursiveGraphExtraction(simgrid::s4u::Engine::get_instance()->getNetRoot(), simgrid::instr::Container::getRoot(),
246                            filter);
247   XBT_DEBUG("Graph extraction finished.");
248   delete filter;
249   TRACE_paje_dump_buffer(true);
250 }
251
252 static void instr_actor_on_creation(simgrid::s4u::ActorPtr actor)
253 {
254   container_t root      = simgrid::instr::Container::getRoot();
255   container_t container = simgrid::instr::Container::byName(actor->get_host()->get_name());
256
257   container->createChild(instr_pid(actor.get()), "ACTOR");
258   simgrid::instr::ContainerType* actor_type = container->type_->getOrCreateContainerType("ACTOR");
259   simgrid::instr::StateType* state          = actor_type->getOrCreateStateType("ACTOR_STATE");
260   state->addEntityValue("suspend", "1 0 1");
261   state->addEntityValue("sleep", "1 1 0");
262   state->addEntityValue("receive", "1 0 0");
263   state->addEntityValue("send", "0 0 1");
264   state->addEntityValue("task_execute", "0 1 1");
265   root->type_->getOrCreateLinkType("ACTOR_LINK", actor_type, actor_type);
266   root->type_->getOrCreateLinkType("ACTOR_TASK_LINK", actor_type, actor_type);
267 }
268
269 static void instr_actor_on_suspend(simgrid::s4u::ActorPtr actor)
270 {
271   simgrid::instr::Container::byName(instr_pid(actor.get()))->getState("ACTOR_STATE")->pushEvent("suspend");
272 }
273
274 static void instr_actor_on_resume(simgrid::s4u::ActorPtr actor)
275 {
276   simgrid::instr::Container::byName(instr_pid(actor.get()))->getState("ACTOR_STATE")->popEvent();
277 }
278
279 static long long int counter = 0;
280
281 static void instr_actor_on_migration_start(simgrid::s4u::ActorPtr actor)
282 {
283   // start link
284   container_t container = simgrid::instr::Container::byName(instr_pid(actor.get()));
285   simgrid::instr::Container::getRoot()->getLink("ACTOR_LINK")->startEvent(container, "M", std::to_string(counter));
286
287   // destroy existing container of this process
288   container->removeFromParent();
289 }
290
291 static void instr_actor_on_migration_end(simgrid::s4u::ActorPtr actor)
292 {
293   // create new container on the new_host location
294   simgrid::instr::Container::byName(actor->get_host()->get_name())->createChild(instr_pid(actor.get()), "ACTOR");
295   // end link
296   simgrid::instr::Container::getRoot()
297       ->getLink("ACTOR_LINK")
298       ->endEvent(simgrid::instr::Container::byName(instr_pid(actor.get())), "M", std::to_string(counter));
299   counter++;
300 }
301
302 static void instr_vm_on_creation(simgrid::s4u::Host& host)
303 {
304   container_t container                 = new simgrid::instr::HostContainer(host, currentContainer.back());
305   container_t root                      = simgrid::instr::Container::getRoot();
306   simgrid::instr::ContainerType* msg_vm = container->type_->getOrCreateContainerType("MSG_VM");
307   simgrid::instr::StateType* state      = msg_vm->getOrCreateStateType("MSG_VM_STATE");
308   state->addEntityValue("suspend", "1 0 1");
309   state->addEntityValue("sleep", "1 1 0");
310   state->addEntityValue("receive", "1 0 0");
311   state->addEntityValue("send", "0 0 1");
312   state->addEntityValue("task_execute", "0 1 1");
313   root->type_->getOrCreateLinkType("MSG_VM_LINK", msg_vm, msg_vm);
314   root->type_->getOrCreateLinkType("MSG_VM_ACTOR_LINK", msg_vm, msg_vm);
315 }
316
317 static void instr_vm_on_start(simgrid::s4u::VirtualMachine& vm)
318 {
319   simgrid::instr::Container::byName(vm.get_name())->getState("MSG_VM_STATE")->pushEvent("start");
320 }
321
322 static void instr_vm_on_started(simgrid::s4u::VirtualMachine& vm)
323 {
324   simgrid::instr::Container::byName(vm.get_name())->getState("MSG_VM_STATE")->popEvent();
325 }
326
327 static void instr_vm_on_suspend(simgrid::s4u::VirtualMachine& vm)
328 {
329   simgrid::instr::Container::byName(vm.get_name())->getState("MSG_VM_STATE")->pushEvent("suspend");
330 }
331
332 static void instr_vm_on_resume(simgrid::s4u::VirtualMachine& vm)
333 {
334   simgrid::instr::Container::byName(vm.get_name())->getState("MSG_VM_STATE")->popEvent();
335 }
336
337 static void instr_vm_on_destruction(simgrid::s4u::Host& host)
338 {
339   simgrid::instr::Container::byName(host.get_name())->removeFromParent();
340 }
341
342 void instr_define_callbacks()
343 {
344   // always need the callbacks to zones (we need only the root zone), to create the rootContainer and the rootType
345   // properly
346   if (TRACE_needs_platform()) {
347     simgrid::s4u::on_platform_created.connect(instr_on_platform_created);
348     simgrid::s4u::Host::onCreation.connect(instr_host_on_creation);
349     simgrid::s4u::Link::onCreation.connect(instr_link_on_creation);
350   }
351   simgrid::s4u::NetZone::onCreation.connect(instr_netzone_on_creation);
352   simgrid::s4u::NetZone::onSeal.connect(instr_netzone_on_seal);
353   simgrid::kernel::routing::NetPoint::onCreation.connect(instr_netpoint_on_creation);
354
355   if (TRACE_actor_is_enabled()) {
356     simgrid::s4u::Actor::on_creation.connect(instr_actor_on_creation);
357     simgrid::s4u::Actor::on_suspend.connect(instr_actor_on_suspend);
358     simgrid::s4u::Actor::on_resume.connect(instr_actor_on_resume);
359     simgrid::s4u::Actor::on_migration_start.connect(instr_actor_on_migration_start);
360     simgrid::s4u::Actor::on_migration_end.connect(instr_actor_on_migration_end);
361   }
362
363   if (TRACE_vm_is_enabled()) {
364     simgrid::s4u::Host::onCreation.connect(instr_vm_on_creation);
365     simgrid::s4u::VirtualMachine::on_start.connect(instr_vm_on_start);
366     simgrid::s4u::VirtualMachine::on_started.connect(instr_vm_on_started);
367     simgrid::s4u::VirtualMachine::on_suspend.connect(instr_vm_on_suspend);
368     simgrid::s4u::VirtualMachine::on_resume.connect(instr_vm_on_resume);
369     simgrid::s4u::Host::onDestruction.connect(instr_vm_on_destruction);
370   }
371 }
372 /*
373  * user categories support
374  */
375 static void recursiveNewVariableType(std::string new_typename, std::string color, simgrid::instr::Type* root)
376 {
377   if (root->get_name() == "HOST" || root->get_name() == "MSG_VM")
378     root->getOrCreateVariableType(std::string("p") + new_typename, color);
379
380   if (root->get_name() == "LINK")
381     root->getOrCreateVariableType(std::string("b") + new_typename, color);
382
383   for (auto elm : root->children_) {
384     recursiveNewVariableType(new_typename, color, elm.second);
385   }
386 }
387
388 void instr_new_variable_type(std::string new_typename, std::string color)
389 {
390   recursiveNewVariableType(new_typename, color, simgrid::instr::Container::getRoot()->type_);
391 }
392
393 static void recursiveNewUserVariableType(std::string father_type, std::string new_typename, std::string color,
394                                          simgrid::instr::Type* root)
395 {
396   if (root->get_name() == father_type) {
397     root->getOrCreateVariableType(new_typename, color);
398   }
399   for (auto elm : root->children_)
400     recursiveNewUserVariableType(father_type, new_typename, color, elm.second);
401 }
402
403 void instr_new_user_variable_type(std::string father_type, std::string new_typename, std::string color)
404 {
405   recursiveNewUserVariableType(father_type, new_typename, color, simgrid::instr::Container::getRoot()->type_);
406 }
407
408 static void recursiveNewUserStateType(std::string father_type, std::string new_typename, simgrid::instr::Type* root)
409 {
410   if (root->get_name() == father_type)
411     root->getOrCreateStateType(new_typename);
412
413   for (auto elm : root->children_)
414     recursiveNewUserStateType(father_type, new_typename, elm.second);
415 }
416
417 void instr_new_user_state_type(std::string father_type, std::string new_typename)
418 {
419   recursiveNewUserStateType(father_type, new_typename, simgrid::instr::Container::getRoot()->type_);
420 }
421
422 static void recursiveNewValueForUserStateType(std::string type_name, const char* val, std::string color,
423                                               simgrid::instr::Type* root)
424 {
425   if (root->get_name() == type_name)
426     static_cast<simgrid::instr::StateType*>(root)->addEntityValue(val, color);
427
428   for (auto elm : root->children_)
429     recursiveNewValueForUserStateType(type_name, val, color, elm.second);
430 }
431
432 void instr_new_value_for_user_state_type(std::string type_name, const char* value, std::string color)
433 {
434   recursiveNewValueForUserStateType(type_name, value, color, simgrid::instr::Container::getRoot()->type_);
435 }
436
437 #define GRAPHICATOR_SUPPORT_FUNCTIONS
438
439 static void recursiveXBTGraphExtraction(xbt_graph_t graph, std::map<std::string, xbt_node_t>* nodes,
440                                         std::map<std::string, xbt_edge_t>* edges, sg_netzone_t netzone,
441                                         container_t container)
442 {
443   if (not netzone->getChildren()->empty()) {
444     // bottom-up recursion
445     for (auto const& netzone_child : *netzone->getChildren()) {
446       container_t child_container = container->children_.at(netzone_child->get_cname());
447       recursiveXBTGraphExtraction(graph, nodes, edges, netzone_child, child_container);
448     }
449   }
450
451   static_cast<simgrid::kernel::routing::NetZoneImpl*>(netzone)->get_graph(graph, nodes, edges);
452 }
453
454 xbt_graph_t instr_routing_platform_graph()
455 {
456   xbt_graph_t ret                          = xbt_graph_new_graph(0, nullptr);
457   std::map<std::string, xbt_node_t>* nodes = new std::map<std::string, xbt_node_t>;
458   std::map<std::string, xbt_edge_t>* edges = new std::map<std::string, xbt_edge_t>;
459   recursiveXBTGraphExtraction(ret, nodes, edges, simgrid::s4u::Engine::get_instance()->getNetRoot(),
460                               simgrid::instr::Container::getRoot());
461   delete nodes;
462   delete edges;
463   return ret;
464 }
465
466 void instr_routing_platform_graph_export_graphviz(xbt_graph_t g, const char* filename)
467 {
468   unsigned int cursor = 0;
469   xbt_node_t node     = nullptr;
470   xbt_edge_t edge     = nullptr;
471
472   FILE* file = fopen(filename, "w");
473   xbt_assert(file, "Failed to open %s \n", filename);
474
475   if (g->directed)
476     fprintf(file, "digraph test {\n");
477   else
478     fprintf(file, "graph test {\n");
479
480   fprintf(file, "  graph [overlap=scale]\n");
481
482   fprintf(file, "  node [shape=box, style=filled]\n");
483   fprintf(file, "  node [width=.3, height=.3, style=filled, color=skyblue]\n\n");
484
485   xbt_dynar_foreach (g->nodes, cursor, node) {
486     fprintf(file, "  \"%s\";\n", instr_node_name(node));
487   }
488   xbt_dynar_foreach (g->edges, cursor, edge) {
489     const char* src_s = instr_node_name(edge->src);
490     const char* dst_s = instr_node_name(edge->dst);
491     if (g->directed)
492       fprintf(file, "  \"%s\" -> \"%s\";\n", src_s, dst_s);
493     else
494       fprintf(file, "  \"%s\" -- \"%s\";\n", src_s, dst_s);
495   }
496   fprintf(file, "}\n");
497   fclose(file);
498 }