get SWIM to use SSG dping fn

src/ssg-internal.h

@@ -65,7 +65,6 @@ typedef struct ssg_group_descriptor
 typedef struct ssg_group_view
 {
     unsigned int size;
-    ssg_member_state_t *member_list;
     ssg_member_state_t *member_map;
 } ssg_group_view_t;
 
@@ -76,6 +75,7 @@ typedef struct ssg_group
     ssg_group_view_t view;
     ssg_group_descriptor_t *descriptor;
     swim_context_t *swim_ctx;
+    ssg_member_state_t *member_list;
     ssg_membership_update_cb update_cb;
     void *update_cb_dat;
     UT_hash_handle hh;

src/ssg-rpc.c

@@ -248,7 +248,7 @@ DEFINE_MARGO_RPC_HANDLER(ssg_group_attach_recv_ult)
 static int ssg_group_view_serialize(
     ssg_group_view_t *view, void **buf, hg_size_t *buf_size)
 {
-    ssg_member_state_t *member_state;
+    ssg_member_state_t *member_state, *tmp;
     hg_size_t view_buf_size = 0;
     void *view_buf;
     void *buf_p, *str_p;
@@ -257,7 +257,7 @@ static int ssg_group_view_serialize(
     *buf_size = 0;
 
     /* first determine view size */
-    LL_FOREACH(view->member_list, member_state)
+    HASH_ITER(hh, view->member_map, member_state, tmp)
     {
         view_buf_size += strlen(member_state->addr_str) + 1;
     }
@@ -267,7 +267,7 @@ static int ssg_group_view_serialize(
         return SSG_FAILURE;
 
     buf_p = view_buf;
-    LL_FOREACH(view->member_list, member_state)
+    HASH_ITER(hh, view->member_map, member_state, tmp)
     {
         str_p = member_state->addr_str;
         strcpy(buf_p, str_p);

src/ssg.c

@@ -123,21 +123,51 @@ int ssg_finalize()
  *** SSG group management routines ***
  *************************************/
 
-/* XXX */
 static int ssg_get_swim_dping_target(
     void *group_data,
     hg_addr_t *target_addr,
-    swim_member_state_t *target_ms)
+    swim_member_state_t *target_swim_ms);
+static void ssg_gen_rand_member_list(
+    ssg_group_t *g);
+
+static int ssg_get_swim_dping_target(
+    void *group_data,
+    hg_addr_t *target_addr,
+    swim_member_state_t *target_swim_ms)
 {
     ssg_group_t *g = (ssg_group_t *)group_data;
+    ssg_member_state_t *target_ms;
 
     assert(g != NULL);
 
-    /* get a random group member, return addr and state */
+    /* if current member list is empty, generate a new random one */
+    if (g->member_list == NULL)
+        ssg_gen_rand_member_list(g);
+
+    /* pull random member off head of list and return addr */
+    target_ms = g->member_list;
+    LL_DELETE(g->member_list, target_ms);
+    *target_addr = target_ms->addr;
+    *target_swim_ms = target_ms->swim_state;
+
+    printf("%lu: pinging %lu\n", g->self_id, target_ms->id);
 
     return 0;
 }
 
+static void ssg_gen_rand_member_list(ssg_group_t *g)
+{
+    ssg_member_state_t *ms, *tmp;
+
+    /* XXX generate random LL permutation from membership map */
+    HASH_ITER(hh, g->view.member_map, ms, tmp)
+    {
+        LL_APPEND(g->member_list, ms);
+    }
+
+    return;
+}
+
 ssg_group_id_t ssg_group_create(
     const char * group_name,
     const char * const group_addr_strs[],
@@ -197,6 +227,9 @@ ssg_group_id_t ssg_group_create(
         goto fini;
     }
 
+    /* generate random list permutation of the member map */
+    ssg_gen_rand_member_list(g);
+
     /* initialize swim failure detector */
     // TODO: we should probably barrier or sync somehow to avoid rpc failures
     // due to timing skew of different ranks initializing swim
@@ -942,7 +975,6 @@ void ssg_apply_membership_update(
             member_state);
         if (member_state)
         {
-            LL_DELETE(g->view.member_list, member_state);
             HASH_DELETE(hh, g->view.member_map, member_state);
             margo_addr_free(ssg_inst->mid, member_state->addr);
             free(member_state->addr_str);
@@ -1105,9 +1137,8 @@ static int ssg_group_view_create(
                 if (self_id)               
                     *self_id = tmp_ms->id;
 
-                /* add self state to membership view */
-                LL_PREPEND(view->member_list, tmp_ms);
-                HASH_ADD(hh, view->member_map, id, sizeof(ssg_member_id_t), tmp_ms);
+                /* XXX: add self state to membership view? or somewhere static? */
+                //HASH_ADD(hh, view->member_map, id, sizeof(ssg_member_id_t), tmp_ms);
 
                 /* don't look up our own address, we already know it */
                 continue;
@@ -1196,7 +1227,6 @@ static void ssg_group_lookup_ult(
         &l->member_state->addr);
     if (l->out == HG_SUCCESS)
     {
-        LL_APPEND(l->view->member_list, l->member_state);
         HASH_ADD(hh, l->view->member_map, id, sizeof(ssg_member_id_t),
             l->member_state);
     }
@@ -1214,15 +1244,13 @@ static void ssg_group_view_destroy(
     ssg_member_state_t *state, *tmp;
 
     /* destroy state for all group members */
-    LL_FOREACH_SAFE(view->member_list, state, tmp)
+    HASH_ITER(hh, view->member_map, state, tmp)
     {
-        LL_DELETE(view->member_list, state);
         HASH_DEL(view->member_map, state);
         free(state->addr_str);
         margo_addr_free(ssg_inst->mid,  state->addr);
         free(state);
     }
-    view->member_list = NULL;
     view->member_map = NULL;
 
     return;

src/swim-fd/swim-fd-ping.c

@@ -18,6 +18,8 @@
 #include "swim-fd.h"
 #include "swim-fd-internal.h"
 
+#if 0
+
 /* NOTE these defines must be kept in sync with defs in
  * ssg.h & swim-internal.h
  */
@@ -443,3 +445,4 @@ static hg_return_t hg_proc_swim_message_t(hg_proc_t proc, void *data)
 
     return(hret);
 }
+#endif

src/swim-fd/swim-fd.c

@@ -4,8 +4,6 @@
  * See COPYRIGHT in top-level directory.
  */
 
-#include <ssg-config.h>
-
 #include <stdio.h>
 #include <stdlib.h>
 #include <unistd.h>
@@ -50,11 +48,6 @@ static void swim_update_suspected_members(
     ssg_group_t *g, double susp_timeout);
 static void swim_add_recent_member_update(
     ssg_group_t *g, swim_member_update_t update);
-static int swim_get_rand_group_member(
-    ssg_group_t *g, ssg_member_id_t *member_id);
-static int swim_get_rand_group_member_set(
-    ssg_group_t *g, ssg_member_id_t *member_ids, int num_members,
-    ssg_member_id_t excluded_id);
 #endif
 
 /******************************************************
@@ -604,52 +597,4 @@ static void swim_add_recent_member_update(
 
     return;
 }
-
-static int swim_get_rand_group_member(
-    ssg_group_t *g, ssg_member_id_t *member_id)
-{
-    int ret = swim_get_rand_group_member_set(g, member_id, 1,
-        SSG_MEMBER_ID_INVALID);
-
-    return(ret);
-}
-
-static int swim_get_rand_group_member_set(
-    ssg_group_t *g, ssg_member_id_t *member_ids, int num_members,
-    ssg_member_id_t excluded_id)
-{
-    unsigned int i, j, rand_ndx = 0;
-    ssg_member_id_t r, rand_member;
-
-    if(num_members == 0)
-        return(0);
-
-    /* TODO: what data structure could we use to avoid looping to look
-     * for a set of random ranks
-     */
-    do
-    {
-        r = rand() % g->view.size;
-        for(i = 0; i < g->view.size; i++)
-        {
-            rand_member = (r + i) % g->view.size;
-            if(rand_member == g->self_id || rand_member == excluded_id ||
-                !(g->view.member_states[rand_member].is_member))
-                continue;
-            for(j = 0; j < rand_ndx; j++)
-            {
-                if(rand_member == member_ids[j])
-                    break;
-            }
-            if(j == rand_ndx)
-                break;
-        }
-        if(i == g->view.size)
-            break;
-
-        member_ids[rand_ndx++] = rand_member;
-    } while((int)rand_ndx < num_members);
-
-    return(rand_ndx);
-}
 #endif