Merge branch 'smpi-replay-mm-robustness'

[simgrid.git] / examples / smpi / replay_multiple_manual_deploy / replay_multiple_manual.cpp

diff --git a/examples/smpi/replay_multiple_manual_deploy/replay_multiple_manual.cpp b/examples/smpi/replay_multiple_manual_deploy/replay_multiple_manual.cpp
index ae6b7f6..23c53d0 100644 (file)
--- a/examples/smpi/replay_multiple_manual_deploy/replay_multiple_manual.cpp
+++ b/examples/smpi/replay_multiple_manual_deploy/replay_multiple_manual.cpp
@@ -65,9 +65,6 @@ static void smpi_replay_process(Job* job, simgrid::s4u::BarrierPtr barrier, int
   argv[3]     = xbt_strdup(job->traces_filenames[rank].c_str()); // smpi trace file for this rank
   argv[4]     = xbt_strdup("0");                                 // ?
 
-  // Ugly double storage used for memory deallocation, as SMPI changes argc/argv without cleaning memory.
-  char* to_free[2] = {argv[0], argv[1]}; // <-- This ugly array should disappear.
-
   XBT_INFO("Replaying rank %d of job %d (smpi_app '%s')", rank, job->unique_job_number, job->smpi_app_name.c_str());
   smpi_replay_run(&argc, &argv);
   XBT_INFO("Finished replaying rank %d of job %d (smpi_app '%s')", rank, job->unique_job_number,
@@ -75,9 +72,7 @@ static void smpi_replay_process(Job* job, simgrid::s4u::BarrierPtr barrier, int
 
   barrier->wait();
 
-  // Memory clean-up
-  for (int i = 0; i < 2; ++i) // <-- This ugly loop should disappear.
-    xbt_free(to_free[i]);
+  // Memory clean-up — leaks can come from argc/argv modifications from smpi_replay_run
   for (int i = 0; i < argc; ++i)
     xbt_free(argv[i]);
   xbt_free(argv);
@@ -111,13 +106,14 @@ static int job_executor_process(Job* job)
   simgrid::s4u::BarrierPtr barrier = simgrid::s4u::Barrier::create(job->app_size + 1);
 
   for (int i = 0; i < job->app_size; ++i) {
-    char* str_pname = bprintf("%d_%d", job->unique_job_number, i);
+    char* str_pname = bprintf("rank_%d_%d", job->unique_job_number, i);
     simgrid::s4u::Actor::create(str_pname, hosts[job->allocation[i]], smpi_replay_process, job, barrier, i);
     xbt_free(str_pname);
   }
 
   barrier->wait();
 
+  simgrid::s4u::this_actor::sleep_for(1);
   XBT_INFO("Finished job %d (smpi_app '%s')", job->unique_job_number, job->smpi_app_name.c_str());
 
   return 0;
@@ -144,9 +140,10 @@ static int workload_executor_process(std::vector<Job*>* workload)
     }
 
     // Let's finally run the job executor
-    std::string job_process_name = "job_" + job->smpi_app_name;
+    char* str_pname = bprintf("job_%04d", job->unique_job_number);
     XBT_INFO("Launching the job executor of job %d (app '%s')", job->unique_job_number, job->smpi_app_name.c_str());
-    simgrid::s4u::Actor::create(job_process_name.c_str(), hosts[job->allocation[0]], job_executor_process, job);
+    simgrid::s4u::Actor::create(str_pname, hosts[job->allocation[0]], job_executor_process, job);
+    free(str_pname);
   }
 
   return 0;
@@ -260,7 +257,7 @@ int main(int argc, char* argv[])
   }
 
   // Let's execute the workload
-  simgrid::s4u::Actor::create("workload_executor", hosts[0], workload_executor_process, &jobs);
+  simgrid::s4u::Actor::create("workload", hosts[0], workload_executor_process, &jobs);
 
   e.run();
   XBT_INFO("Simulation finished! Final time: %g", e.get_clock());