static int enter_request_cb (flux_t h, int typemask, zmsg_t **zmsg, void *arg)
{
ctx_t *ctx = arg;
barrier_t *b;
json_object *o = NULL;
char *sender = NULL;
const char *name;
int count, nprocs, hopcount;
if (flux_msg_decode (*zmsg, NULL, &o) < 0 || o == NULL
|| !(sender = flux_msg_sender (*zmsg))
|| util_json_object_get_string (o, "name", &name) < 0
|| util_json_object_get_int (o, "count", &count) < 0
|| util_json_object_get_int (o, "nprocs", &nprocs) < 0) {
flux_log (ctx->h, LOG_ERR, "%s: ignoring bad message", __FUNCTION__);
goto done;
}
if (!(b = zhash_lookup (ctx->barriers, name)))
b = barrier_create (ctx, name, nprocs);
/* Distinguish client (tracked) vs downstream barrier plugin (untracked).
* A client, distinguished by hopcount > 0, can only enter barrier once.
*/
if (util_json_object_get_int (o, "hopcount", &hopcount) < 0) {
if (barrier_add_client (b, sender, zmsg) < 0) {
flux_respond_errnum (ctx->h, zmsg, EEXIST);
flux_log (ctx->h, LOG_ERR,
"abort %s due to double entry by client %s",
name, sender);
if (exit_event_send (ctx->h, b->name, ECONNABORTED) < 0)
flux_log (ctx->h, LOG_ERR, "exit_event_send: %s", strerror (errno));
goto done;
}
}
/* If the count has been reached, terminate the barrier;
* o/w set timer to pass count upstream and zero it here.
*/
b->count += count;
if (b->count == b->nprocs) {
if (exit_event_send (ctx->h, b->name, 0) < 0)
flux_log (ctx->h, LOG_ERR, "exit_event_send: %s", strerror (errno));
} else if (!flux_treeroot (ctx->h) && !ctx->timer_armed) {
if (flux_tmouthandler_add (h, barrier_reduction_timeout_msec,
true, timeout_cb, ctx) < 0) {
flux_log (h, LOG_ERR, "flux_tmouthandler_add: %s",strerror (errno));
goto done;
}
ctx->timer_armed = true;
}
done:
if (o)
json_object_put (o);
if (*zmsg)
zmsg_destroy (zmsg);
if (sender)
free (sender);
return 0;
}
static int exit_event_cb (flux_t h, int typemask, zmsg_t **zmsg, void *arg)
{
ctx_t *ctx = arg;
barrier_t *b;
json_object *o = NULL;
const char *name;
int errnum;
if (flux_msg_decode (*zmsg, NULL, &o) < 0 || o == NULL
|| util_json_object_get_string (o, "name", &name) < 0
|| util_json_object_get_int (o, "errnum", &errnum) < 0) {
flux_log (h, LOG_ERR, "%s: bad message", __FUNCTION__);
goto done;
}
if ((b = zhash_lookup (ctx->barriers, name))) {
b->errnum = errnum;
zhash_foreach (b->clients, send_enter_response, b);
zhash_delete (ctx->barriers, name);
}
done:
if (o)
json_object_put (o);
if (*zmsg)
zmsg_destroy (zmsg);
return 0;
}
static int wait_for_lwj_watch_init (flux_t h, int64_t id)
{
int rc;
json_object *rpc_o;
json_object *rpc_resp;
rpc_o = util_json_object_new_object ();
util_json_object_add_string (rpc_o, "key", "lwj.next-id");
util_json_object_add_int64 (rpc_o, "val", id);
rc = flux_json_rpc (h, FLUX_NODEID_ANY, "sim_sched.lwj-watch", rpc_o, &rpc_resp);
if (rc >= 0) {
util_json_object_get_int (rpc_resp, "rc", &rc);
json_object_put (rpc_resp);
}
json_object_put (rpc_o);
return rc;
}
