normal state running ok

test/Makefile

@@ -5,12 +5,12 @@ export LDFLAGS= -pthread -lm  -lrt
 export CFLAGS = -Wall -O3 -msse2  -Wno-unknown-pragmas -fPIC -I../src
 
 # specify tensor path
-BIN = test_allreduce test_recover test_model_recover speed_test
+BIN = test_allreduce test_recover test_model_recover speed_test test_local_recover
 # objectives that makes up rabit library
 RABIT_OBJ = allreduce_base.o allreduce_robust.o engine.o
 MPIOBJ = engine_mpi.o
 
-OBJ = $(RABIT_OBJ) test_allreduce.o test_recover.o test_model_recover.o speed_test.o
+OBJ = $(RABIT_OBJ) test_allreduce.o test_recover.o test_model_recover.o speed_test.o test_local_recover.o
 MPIBIN = test_allreduce.mpi speed_test.mpi
 .PHONY: clean all
 
@@ -24,6 +24,7 @@ test_allreduce.o: test_allreduce.cpp ../src/*.h
 speed_test.o: speed_test.cpp ../src/*.h
 test_recover.o: test_recover.cpp ../src/*.h 
 test_model_recover.o: test_model_recover.cpp ../src/*.h
+test_local_recover.o: test_local_recover.cpp ../src/*.h
 
 # we can link against MPI version to get use MPI
 test_allreduce: test_allreduce.o  $(RABIT_OBJ)
@@ -32,6 +33,7 @@ speed_test: speed_test.o  $(RABIT_OBJ)
 speed_test.mpi: speed_test.o $(MPIOBJ)
 test_recover: test_recover.o  $(RABIT_OBJ)
 test_model_recover: test_model_recover.o  $(RABIT_OBJ)
+test_local_recover: test_local_recover.o  $(RABIT_OBJ)
 
 $(BIN) : 
 	$(CXX) $(CFLAGS) -o $@ $(filter %.cpp %.o %.c %.cc, $^) $(LDFLAGS)

test/keepalive.sh

@@ -6,8 +6,8 @@ then
     exit -1
 fi
 nrep=0
-echo ./$@ job_id=$OMPI_COMM_WORLD_RANK
-until ./$@ job_id=$OMPI_COMM_WORLD_RANK repeat=$nrep; do
+echo ./$@ task_id=$OMPI_COMM_WORLD_RANK
+until ./$@ task_id=$OMPI_COMM_WORLD_RANK repeat=$nrep; do
     sleep 1
     nrep=$((nrep+1))
     echo ./$@ job_id=$OMPI_COMM_WORLD_RANK repeat=$nrep

test/test_local_recover.cpp

@@ -0,0 +1,154 @@
+// this is a test case to test whether rabit can recover model when 
+// facing an exception
+#include <rabit.h>
+#include <utils.h>
+#include <cstdio>
+#include <cstdlib>
+#include <cmath>
+#include <mock.h>
+
+using namespace rabit;
+
+struct MockException {
+};
+
+// dummy model
+class Model : public rabit::utils::ISerializable {
+ public:
+  // iterations
+  std::vector<float> data;
+  // load from stream
+  virtual void Load(rabit::utils::IStream &fi) {
+    fi.Read(&data);
+  }
+  /*! \brief save the model to the stream */
+  virtual void Save(rabit::utils::IStream &fo) const {
+    fo.Write(data);
+  }
+  virtual void InitModel(size_t n, float v) {
+    data.resize(n, v);
+  }
+};
+
+inline void TestMax(test::Mock &mock, Model *model, Model *local, int ntrial, int iter) {
+  int rank = rabit::GetRank();
+  int nproc = rabit::GetWorldSize();
+  const int z = iter + 111;
+
+  std::vector<float> ndata(model->data.size());
+  for (size_t i = 0; i < ndata.size(); ++i) {
+    ndata[i] = (i * (rank+1)) % z  + local->data[i];
+  }
+  mock.Allreduce<op::Max>(&ndata[0], ndata.size());  
+  if (ntrial == iter && rank == 3) {
+    //exit(-1);
+  }
+  for (size_t i = 0; i < ndata.size(); ++i) {
+    float rmax = (i * 1) % z + model->data[i];
+    for (int r = 0; r < nproc; ++r) {
+      rmax = std::max(rmax, (float)((i * (r+1)) % z) + model->data[i] + r);
+    }
+    utils::Check(rmax == ndata[i], "[%d] TestMax check failure", rank);
+  }
+  model->data = ndata;
+  local->data = ndata;
+  for (size_t i = 0; i < ndata.size(); ++i) {
+    local->data[i] = ndata[i] + rank;
+  }
+}
+
+inline void TestSum(test::Mock &mock, Model *model, Model *local, int ntrial, int iter) {
+  int rank = rabit::GetRank();
+  int nproc = rabit::GetWorldSize();
+  const int z = 131 + iter;
+
+  std::vector<float> ndata(model->data.size());
+  for (size_t i = 0; i < ndata.size(); ++i) {
+    ndata[i] = (i * (rank+1)) % z + local->data[i];
+  }
+  mock.Allreduce<op::Sum>(&ndata[0], ndata.size());
+
+  if (ntrial == iter && rank == 0) {
+    exit(-1);
+  }
+
+  for (size_t i = 0; i < ndata.size(); ++i) {
+    float rsum = 0.0f;
+    for (int r = 0; r < nproc; ++r) {
+      rsum += (float)((i * (r+1)) % z) + model->data[i] + r;
+    }
+    utils::Check(fabsf(rsum - ndata[i]) < 1e-5 ,
+                 "[%d] TestSum check failure, local=%g, allreduce=%g", rank, rsum, ndata[i]);
+  }
+  model->data = ndata;
+  for (size_t i = 0; i < ndata.size(); ++i) {
+    local->data[i] = ndata[i] + rank;
+  }
+}
+
+inline void TestBcast(test::Mock &mock, size_t n, int root, int ntrial) {
+  int rank = rabit::GetRank();
+  std::string s; s.resize(n);      
+  for (size_t i = 0; i < n; ++i) {
+    s[i] = char(i % 126 + 1);
+  }
+  std::string res;
+  if (root == rank) {
+    res = s;
+    mock.Broadcast(&res, root);
+  } else {
+    mock.Broadcast(&res, root);
+  }
+  utils::Check(res == s, "[%d] TestBcast fail", rank);
+}
+
+int main(int argc, char *argv[]) {
+  if (argc < 3) {
+    printf("Usage: <ndata> <config>\n");
+    return 0;
+  }
+  int n = atoi(argv[1]);
+  rabit::Init(argc, argv);
+  int rank = rabit::GetRank();
+  int nproc = rabit::GetWorldSize();
+  std::string name = rabit::GetProcessorName();
+  test::Mock mock(rank, argv[2], argv[3]);
+  Model model, local;  
+  srand(0);
+  int ntrial = 0;
+  for (int i = 1; i < argc; ++i) {
+    int n;
+    if (sscanf(argv[i], "repeat=%d", &n) == 1) ntrial = n; 
+  } 
+  while (true) {
+    try {
+      int iter = rabit::LoadCheckPoint(&model, &local);
+      if (iter == 0) {
+        model.InitModel(n, 1.0f);
+        local.InitModel(n, 1.0f + rank);
+        utils::LogPrintf("[%d] reload-trail=%d, init iter=%d\n", rank, ntrial, iter);
+      } else {
+        utils::LogPrintf("[%d] reload-trail=%d, init iter=%d\n", rank, ntrial, iter);
+      }
+      for (int r = iter; r < 3; ++r) { 
+        TestMax(mock, &model, &local, ntrial, r);
+        utils::LogPrintf("[%d] !!!TestMax pass, iter=%d\n",  rank, r);  
+        int step = std::max(nproc / 3, 1);
+        for (int i = 0; i < nproc; i += step) {
+          TestBcast(mock, n, i, ntrial);
+        }
+        utils::LogPrintf("[%d] !!!TestBcast pass, iter=%d\n", rank, r);
+        TestSum(mock, &model, &local, ntrial, r);
+        utils::LogPrintf("[%d] !!!TestSum pass, iter=%d\n", rank, r);
+        rabit::CheckPoint(&model, &local);
+        utils::LogPrintf("[%d] !!!CheckPont pass, iter=%d\n", rank, r);
+      }
+      break;
+    } catch (MockException &e) {
+      rabit::engine::GetEngine()->InitAfterException();
+      ++ntrial;
+    }
+  }
+  rabit::Finalize();
+  return 0;
+}

test/test_model_recover.cpp

@@ -132,7 +132,7 @@ int main(int argc, char *argv[]) {
         utils::LogPrintf("[%d] !!!TestBcast pass, iter=%d\n", rank, r);
         TestSum(mock, &model, ntrial, r);
         utils::LogPrintf("[%d] !!!TestSum pass, iter=%d\n", rank, r);
-        rabit::CheckPoint(model);
+        rabit::CheckPoint(&model);
         utils::LogPrintf("[%d] !!!CheckPont pass, iter=%d\n", rank, r);
       }
       break;