Merge pull request #1528 from hpcraink/pr/osx_sun_path

OSX tempdir too long for sun_path

Author: rhc54

opal/mca/pmix/pmix120/pmix/src/server/pmix_server.c

@@ -143,6 +143,7 @@ static pmix_status_t initialize_server_base(pmix_server_module_t *module)
 {
     int debug_level;
     char *tdir, *evar;
+    char * pmix_pid;
     pid_t pid;
 
     /* initialize the output system */
@@ -219,7 +220,14 @@ static pmix_status_t initialize_server_base(pmix_server_module_t *module)
     /* now set the address - we use the pid here to reduce collisions */
     memset(&myaddress, 0, sizeof(struct sockaddr_un));
     myaddress.sun_family = AF_UNIX;
-    snprintf(myaddress.sun_path, sizeof(myaddress.sun_path)-1, "%s/pmix-%d", tdir, pid);
+    asprintf(&pmix_pid, "pmix-%d", pid);
+    // If the above set temporary directory name plus the pmix-PID string
+    // plus the '/' separator are too long, just fail, so the caller
+    // may provide the user with a proper help... *Cough*, *Cough* OSX...
+    if ((strlen(tdir) + strlen(pmix_pid) + 1) > sizeof(myaddress.sun_path)-1) {
+        return PMIX_ERR_INVALID_LENGTH;
+    }
+    snprintf(myaddress.sun_path, sizeof(myaddress.sun_path)-1, "%s/%s", tdir, pmix_pid);
     asprintf(&myuri, "%s:%lu:%s", pmix_globals.myid.nspace, (unsigned long)pmix_globals.myid.rank, myaddress.sun_path);
 
 

opal/mca/pmix/pmix120/pmix/src/server/pmix_server_listener.c

@@ -72,19 +72,20 @@ pmix_status_t pmix_start_listening(struct sockaddr_un *address)
 {
     int flags;
     pmix_status_t rc;
-    unsigned int addrlen;
+    socklen_t addrlen;
     char *ptr;
 
     /* create a listen socket for incoming connection attempts */
     pmix_server_globals.listen_socket = socket(PF_UNIX, SOCK_STREAM, 0);
     if (pmix_server_globals.listen_socket < 0) {
-        printf("%s:%d socket() failed", __FILE__, __LINE__);
+        printf("%s:%d socket() failed\n", __FILE__, __LINE__);
         return PMIX_ERROR;
     }
 
     addrlen = sizeof(struct sockaddr_un);
     if (bind(pmix_server_globals.listen_socket, (struct sockaddr*)address, addrlen) < 0) {
-        printf("%s:%d bind() failed", __FILE__, __LINE__);
+        printf("%s:%d bind() failed error:%s\n", __FILE__, __LINE__,
+                strerror(errno));
         return PMIX_ERROR;
     }
     /* set the mode as required */
@@ -95,18 +96,18 @@ pmix_status_t pmix_start_listening(struct sockaddr_un *address)
 
     /* setup listen backlog to maximum allowed by kernel */
     if (listen(pmix_server_globals.listen_socket, SOMAXCONN) < 0) {
-        printf("%s:%d listen() failed", __FILE__, __LINE__);
+        printf("%s:%d listen() failed\n", __FILE__, __LINE__);
         return PMIX_ERROR;
     }
 
     /* set socket up to be non-blocking, otherwise accept could block */
     if ((flags = fcntl(pmix_server_globals.listen_socket, F_GETFL, 0)) < 0) {
-        printf("%s:%d fcntl(F_GETFL) failed", __FILE__, __LINE__);
+        printf("%s:%d fcntl(F_GETFL) failed\n", __FILE__, __LINE__);
         return PMIX_ERROR;
     }
     flags |= O_NONBLOCK;
     if (fcntl(pmix_server_globals.listen_socket, F_SETFL, flags) < 0) {
-        printf("%s:%d fcntl(F_SETFL) failed", __FILE__, __LINE__);
+        printf("%s:%d fcntl(F_SETFL) failed\n", __FILE__, __LINE__);
         return PMIX_ERROR;
     }
 

orte/mca/ess/base/ess_base_std_orted.c

@@ -515,7 +515,7 @@ int orte_ess_base_orted_setup(char **hosts)
     /* setup the PMIx server */
     if (ORTE_SUCCESS != (ret = pmix_server_init())) {
         ORTE_ERROR_LOG(ret);
-        error = "pmix server init";
+        error = "Try a shorter TMPDIR var. or change your computer's name (see uname -n), since pmix_server_init";
         goto error;
     }
 

orte/mca/ess/hnp/ess_hnp_module.c

@@ -634,7 +634,7 @@ static int rte_init(void)
     /* setup the PMIx server */
     if (ORTE_SUCCESS != (ret = pmix_server_init())) {
         ORTE_ERROR_LOG(ret);
-        error = "pmix server init";
+        error = "Try a shorter TMPDIR var. or change your computer's name (see uname -n), since pmix_server_init";
         goto error;
     }
 

orte/orted/pmix/pmix_server.c

@@ -246,6 +246,9 @@ int pmix_server_init(void)
     if (ORTE_SUCCESS != (rc = opal_pmix.server_init(&pmix_server, &info))) {
         ORTE_ERROR_LOG(rc);
         /* memory cleanup will occur when finalize is called */
+        orte_show_help("help-orterun.txt", "orterun:pmix-failed", true,
+                       orte_process_info.proc_session_dir);
+        return rc;
     }
     OPAL_LIST_DESTRUCT(&info);
 

orte/tools/orterun/help-orterun.txt

@@ -660,3 +660,12 @@ method and try launching your job again.
 
 Your job will now abort.
 #
+[orterun:pmix-failed]
+The call to pmix_init_server() failed. This may be due to your
+system's restriction for Unix's socket's path-length.
+
+   orte_proc_session_dir: %s
+
+Please try to set TMPDIR to something short (like /tmp) or change
+Your computer's name (see uname -n).
+#