Merge pull request flux-framework#1457 from garlick/hwloc_lite

Colllection of small fixes for splash project (part 2)

src/broker/broker.c

@@ -1477,8 +1477,11 @@ static int handle_event (broker_ctx_t *ctx, const flux_msg_t *msg)
     }
     ctx->event_recv_seq = seq;
 
-    (void)overlay_mcast_child (ctx->overlay, msg);
-    (void)overlay_sendmsg_relay (ctx->overlay, msg);
+    if (overlay_mcast_child (ctx->overlay, msg) < 0)
+        flux_log_error (ctx->h, "%s: overlay_mcast_child", __FUNCTION__);
+    if (overlay_sendmsg_relay (ctx->overlay, msg) < 0)
+        flux_log_error (ctx->h, "%s: overlay_sendmsg_relay", __FUNCTION__);
+
 
     /* Internal services may install message handlers for events.
      */

src/broker/overlay.c

@@ -453,15 +453,9 @@ const char *overlay_get_relay (overlay_t *ov)
 
 int overlay_sendmsg_relay (overlay_t *ov, const flux_msg_t *msg)
 {
-    int rc = -1;
-
-    if (!ov->relay || !ov->relay->zs) {
-        errno = EINVAL;
-        goto done;
-    }
-    rc = flux_msg_sendzsock (ov->relay->zs, msg);
-done:
-    return rc;
+    if (!ov->relay)
+        return 0;
+    return flux_msg_sendzsock (ov->relay->zs, msg);
 }
 
 static void child_cb (flux_reactor_t *r, flux_watcher_t *w,

src/common/libjsc/jstatctl.c

@@ -622,6 +622,29 @@ static int query_rdesc (flux_t *h, int64_t j, json_object **jcb)
     return 0;
 }
 
+int jsc_query_rdesc_efficiently (flux_t *h, int64_t jobid,
+                                 int64_t *nnodes, int64_t *ntasks,
+                                 int64_t *ncores, int64_t *walltime)
+{
+    if (nnodes) {
+        if (extract_raw_nnodes (h, jobid, nnodes) < 0)
+            return -1;
+    }
+    if (ntasks) {
+        if (extract_raw_ntasks (h, jobid, ntasks) < 0)
+            return -1;
+    }
+    if (ncores) {
+        if (extract_raw_ncores (h, jobid, ncores) < 0)
+            return -1;
+    }
+    if (walltime) {
+        if (extract_raw_walltime (h, jobid, walltime) < 0)
+            return -1;
+    }
+    return 0;
+}
+
 static int query_rdl (flux_t *h, int64_t j, json_object **jcb)
 {
     char *rdlstr = NULL;

src/common/libjsc/jstatctl.h

@@ -122,6 +122,15 @@ int jsc_update_jcb (flux_t *h, int64_t jobid, const char *key, const char *jcb);
 const char *jsc_job_num2state (job_state_t s);
 int jsc_job_state2num (const char *s);
 
+/**
+ * Accessor for nnodes, ntasks, ncores, walltime that circumvents
+ * JSON encode/decode.  N.B. this is a workaround for performance issu
+ * encountered when scheduling high throughput workloads.
+ */
+int jsc_query_rdesc_efficiently (flux_t *h, int64_t jobid,
+                                 int64_t *nnodes, int64_t *ntasks,
+                                 int64_t *ncores, int64_t *walltime);
+
 #ifdef __cplusplus
 }
 #endif

src/modules/resource-hwloc/resource.c

@@ -70,12 +70,10 @@ static int ctx_hwloc_init (flux_t *h, resource_ctx_t *ctx)
         flux_log_error (h, "flux_topology_init");
         goto done;
     }
-    if (hwloc_topology_set_flags (ctx->topology,
-                                HWLOC_TOPOLOGY_FLAG_WHOLE_IO) < 0) {
-        flux_log_error (h, "hwloc_topology_set_flags");
-        goto done;
-    }
-
+    if (hwloc_topology_ignore_type (ctx->topology, HWLOC_OBJ_CACHE) < 0)
+        flux_log (h, LOG_ERR, "hwloc_topology_ignore_type OBJ_CACHE failed");
+    if (hwloc_topology_ignore_type (ctx->topology, HWLOC_OBJ_GROUP) < 0)
+        flux_log (h, LOG_ERR, "hwloc_topology_ignore_type OBJ_GROUP failed");
     key = xasprintf ("config.resource.hwloc.xml.%" PRIu32, ctx->rank);
     if (!(f = flux_kvs_lookup (h, 0, key))) {
         flux_log_error (h, "flux_kvs_lookup");