write server cluster info to file

include/mobject-server.h

@@ -13,7 +13,7 @@
 
 #define MOBJECT_SERVER_GROUP_NAME "mobject-store-servers"
 
-int mobject_server_init(margo_instance_id mid);
+int mobject_server_init(margo_instance_id mid, const char *cluster_file);
 
 /**
  * Start a mobject server instance

src/mobject-server-daemon.c

@@ -12,16 +12,16 @@
 
 void usage(void)
 {
-    fprintf(stderr, "Usage: mobject-server-daemon <listen_addr> <gid_file>\n");
+    fprintf(stderr, "Usage: mobject-server-daemon <listen_addr> <cluster_file>\n");
     fprintf(stderr, "  <listen_addr>    the Mercury address to listen on\n");
-    fprintf(stderr, "  <gid_file>       the file to write the server SSG group ID to\n");
+    fprintf(stderr, "  <cluster_file>   the file to write mobject cluster connect info to\n");
     exit(-1);
 }
 
 int main(int argc, char *argv[])
 {
     char *listen_addr;
-    char *gid_file;
+    char *cluster_file;
     margo_instance_id mid;
     int ret;
 
@@ -29,7 +29,7 @@ int main(int argc, char *argv[])
     if (argc != 3)
         usage();
     listen_addr = argv[1];
-    gid_file = argv[2];
+    cluster_file = argv[2];
 
     /* XXX: MPI required for SSG bootstrapping */
     MPI_Init(&argc, &argv);
@@ -41,7 +41,7 @@ int main(int argc, char *argv[])
         return -1;
     }
 
-    ret = mobject_server_init(mid);
+    ret = mobject_server_init(mid, cluster_file);
     if (ret != 0)
     {
         fprintf(stderr, "Error: Unable to initialize mobject server\n");
@@ -49,12 +49,10 @@ int main(int argc, char *argv[])
         return -1;
     }
 
-    /* XXX write GID to file... where does gid come from? */
-
     /* shutdown */
+    //margo_wait_for_finalize(mid); /* XXX: probably need a conditional here, so we can cleanup after blocking */
     mobject_server_shutdown(mid);
     margo_finalize(mid);
-    //margo_wait_for_finalize(mid);
     MPI_Finalize();
 
     return 0;

src/mobject-server.c

@@ -14,16 +14,19 @@
 
 #include "mobject-server.h"
 
+
 typedef struct mobject_server_context
 {
     /* XXX bake, sds-keyval stuff */
     ssg_group_id_t gid;
 } mobject_server_context_t;
 
+static int mobject_server_store_cluster_info(const char *file_name);
+
 /* XXX one global mobject server state struct */
 mobject_server_context_t *g_srv_ctx = NULL;
 
-int mobject_server_init(margo_instance_id mid)
+int mobject_server_init(margo_instance_id mid, const char *cluster_file)
 {
     int ret;
 
@@ -58,12 +61,25 @@ int mobject_server_init(margo_instance_id mid)
         return -1;
     }
 
+    /* write cluster connect info to file for clients to find later */
+    ret = mobject_server_store_cluster_info(cluster_file);
+    if (ret != 0)
+    {
+        fprintf(stderr, "Error: unable to store mobject cluster info to file %s\n",
+            cluster_file);
+        /* XXX: this call is performed by one process under the covers, and we
+         * don't currently have a way to propagate this error to the entire cluster
+         */
+        ssg_group_destroy(g_srv_ctx->gid);
+        ssg_finalize();
+        return -1;
+    }
+
     /* XXX cleanup? */
 
     return 0;
 }
 
-#if 0
 int mobject_server_register(margo_instance_id mid, const char *poolname)
 {
     int ret=0;
@@ -90,3 +106,19 @@ void mobject_server_shutdown(margo_instance_id mid)
 
     return;
 }
+
+static int mobject_server_store_cluster_info(const char *file_name)
+{
+    int my_id;
+    int ret;
+
+    assert(g_srv_ctx);
+
+    /* only have one group member do this */
+    my_id = ssg_get_group_self_id(g_srv_ctx->gid);
+    if (my_id != 0)
+        return 0;
+
+    ret = ssg_group_id_store(file_name, g_srv_ctx->gid);
+    return ret;
+}