Merge pull request #310 from edoapra/too-many-open-files

MPI-PR: check for number of open files
GlobalArrays · Jul 13, 2023 · ab9af9c · ab9af9c
2 parents 3cb97b6 + 06c48eb
commit ab9af9c
Show file tree

Hide file tree

Showing 2 changed files with 26 additions and 5 deletions.
diff --git a/.github/workflows/github_actions.yml b/.github/workflows/github_actions.yml
@@ -67,6 +67,7 @@ jobs:
             armci_network: mpi-pr
             f77: ifort
             cc: icc
+            config_opts: LIBS=-lifcore
             oneapi: /Users/runner/apps/oneapi
           - os: ubuntu-20.04
             experimental: true 
@@ -116,6 +117,7 @@ jobs:
             armci_network: mpi-ts
             f77: ifort
             cc: gcc
+            oneapi: /opt/intel/oneapi
         exclude:
           - armci_network: mpi-pr
             mpi_impl: openmpi

diff --git a/comex/src-mpi-pr/comex.c b/comex/src-mpi-pr/comex.c
@@ -179,7 +179,9 @@ static char *static_server_buffer = NULL;
 static int static_server_buffer_size = 0;
 static int eager_threshold = -1;
 static int max_message_size = -1;
+#if ENABLE_SYSV
 static int use_dev_shm = 1;
+#endif
 static int token_counter = 0;
 static int init_from_comm = 0;
 
@@ -375,6 +377,7 @@ STATIC void check_devshm(int fd, size_t size);
 static int devshm_initialized = 0;
 static long devshm_fs_left = 0;
 static long devshm_fs_initial = 0;
+STATIC void count_open_fds(void);
 
 int _comex_init(MPI_Comm comm)
 {
@@ -7561,13 +7564,9 @@ STATIC void check_devshm(int fd, size_t size){
 	    g_state.rank, g_state.node_size, devshm_fs_initial/CONVERT_TO_M, (long) ufs_statfs.f_bsize, (long)  g_state.node_size);
 #endif
   }
-  //  if (size > 0) {
+  count_open_fds();
     newspace = (long) ( size*(g_state.node_size -1));
-    //  }else{
-    //    newspace = (long) ( size);
-    //  }
     if(newspace>0){
-      // noo fd for space<0
     fstatfs(fd, &ufs_statfs);
 #ifdef DEBUGSHM
     fprintf(stderr, "[%d] /dev/shm filesize %ld filesize*np %ld initial devshm space %ld current /dev/shm space %ld \n",
@@ -7594,3 +7593,23 @@ STATIC void check_devshm(int fd, size_t size){
 #endif
 #endif
 }
+
+STATIC void count_open_fds(void) {
+  FILE *f = fopen("/proc/sys/fs/file-nr", "r");
+
+  long nfiles, unused, maxfiles;
+  fscanf(f, "%ld %ld %ld", &nfiles, &unused, &maxfiles);
+#ifdef DEBUGSHM
+  if(nfiles % 1000 == 0) fprintf(stderr," %d: no. open files = %ld maxfiles = %ld\n", g_state.rank, nfiles, maxfiles);
+#endif
+    if(nfiles > (maxfiles/100)*60) {
+      printf(" %d: running out of files; files = %ld  maxfiles = %ld\n", g_state.rank, nfiles, maxfiles);
+#if PAUSE_ON_ERROR
+    fprintf(stderr,"%d(%d): too many open files\n",
+            g_state.rank,  getpid());
+    pause();
+#endif
+    comex_error("count_open_fds: too many open files", -1);
+  }
+  fclose(f);
+}