Fix some errors on Travis* Add oversubscribe option if needed* Clean exit on cman* oclint conflict on macOS

Author: barche

conf/travis-install-mpi.sh

@@ -14,6 +14,7 @@ case "$os" in
     Darwin)
         brew update
         brew upgrade cmake
+        brew cask uninstall oclint # Prevent conflict with gcc
         case "$MPI_IMPL" in
             mpich|mpich3)
                 brew install mpich

src/cman.jl

@@ -245,8 +245,12 @@ end
 
 # Kill a worker
 function kill(mgr::MPIManager, pid::Int, config::WorkerConfig)
-    # Do nothing, as the worker will self-terminate after calling MPI.Finalize
-    Base.set_worker_state(Base.Worker(pid), Base.W_TERMINATED)
+    # Exit the worker to avoid EOF errors on the workers
+    @spawnat pid begin
+        MPI.Finalize()
+        exit()
+    end
+    Distributed.set_worker_state(Distributed.Worker(pid), Distributed.W_TERMINATED)
 end
 
 # Set up a connection to a worker
@@ -418,14 +422,7 @@ end
 function stop_main_loop(mgr::MPIManager)
     if mgr.mode == TCP_TRANSPORT_ALL
         # Shut down all workers
-        for i in workers()
-            if i != myid()
-                @spawnat i begin
-                    MPI.Finalize()
-                    exit()
-                end
-            end
-        end
+        rmprocs(workers())
         # Poor man's flush of the send queue
         sleep(1)
         put!(mgr.initiate_shutdown, nothing)

test/runtests.jl

@@ -14,21 +14,34 @@ const coverage_opts =
                       JL_LOG_USER => "user",
                       JL_LOG_ALL => "all")
 
+# Files to run without mpiexec
+juliafiles = ["test_cman_julia.jl"]
+# Files to run with mpiexec -n 1
+singlefiles = []
+
 function runtests()
     nprocs = clamp(Sys.CPU_CORES, 2, 4)
     exename = joinpath(JULIA_HOME, Base.julia_exename())
     testdir = dirname(@__FILE__)
-    istest(f) = endswith(f, ".jl") && f != "runtests.jl"
+    istest(f) = endswith(f, ".jl") && startswith(f, "test_")
     testfiles = sort(filter(istest, readdir(testdir)))
+
+    extra_args = []
+    if contains(readlines(open(`mpiexec --version`)[1])[1], "OpenRTE")
+        push!(extra_args,"--oversubscribe")
+    end
+
     nfail = 0
     print_with_color(:white, "Running MPI.jl tests\n")
     for f in testfiles
         try
             coverage_opt = coverage_opts[Base.JLOptions().code_coverage]
-            if f == "test_cman_julia.jl"
+            if f ∈ singlefiles
+                run(`mpiexec $extra_args -n 1 $exename --code-coverage=$coverage_opt $(joinpath(testdir, f))`)
+            elseif f ∈ juliafiles
                 run(`$exename --code-coverage=$coverage_opt $(joinpath(testdir, f))`)
             else
-                run(`mpiexec -n $nprocs $exename --code-coverage=$coverage_opt $(joinpath(testdir, f))`)
+                run(`mpiexec $extra_args -n $nprocs $exename --code-coverage=$coverage_opt $(joinpath(testdir, f))`)
             end
             Base.with_output_color(:green,STDOUT) do io
                 println(io,"\tSUCCESS: $f")

test/test_cman_julia.jl

@@ -3,7 +3,11 @@ using MPI
 
 # Start workers via `mpiexec` that communicate among themselves via MPI;
 # communicate with the workers via TCP
-mgr = MPI.MPIManager(np=4)
+if contains(readlines(open(`mpiexec --version`)[1])[1], "OpenRTE")
+    mgr = MPI.MPIManager(np=4, mpirun_cmd=`mpiexec --oversubscribe -n 4`)
+else
+    mgr = MPI.MPIManager(np=4)
+end
 addprocs(mgr)
 
 refs = []