wreck: drop support for rank.N dirs in wreckrun

Now that all job allocation mechanisms (wreck internal, sched)
use R_lite for resources, drop support for rank.N dirs in
wreck job launch. Update tests accordingly.

src/modules/wreck/job.c

@@ -735,45 +735,6 @@ static int spawn_exec_handler (flux_t *h, struct wreck_job *job)
     return (-1);
 }
 
-/* Handle response to KVS look up of rank.N.
- * If it exists, spawn wrexecd.
- * This concludes the continuation chain started at runevent_cb().
- */
-static void runevent_fallback_continuation (flux_future_t *f, void *arg)
-{
-    struct wreck_job *job = arg;
-    flux_t *h = flux_future_get_flux (f);
-    const char *key = flux_kvs_lookup_get_key (f);
-
-    if (flux_future_get (f, NULL) < 0) {
-        flux_log (h, LOG_DEBUG, "No dir %s: %s", key, flux_strerror (errno));
-        goto done;
-    }
-    if (spawn_exec_handler (h, job) < 0)
-        goto done;
-done:
-    flux_future_destroy (f);
-}
-
-/* Send request to look up rank.N.
- * This function is continued in runevent_fallback_continuation().
- */
-static int runevent_fallback (flux_t *h, struct wreck_job *job)
-{
-    char key[MAX_JOB_PATH];
-    flux_future_t *f;
-
-    snprintf (key, sizeof (key), "%s.rank.%lu",
-              job->kvs_path, (unsigned long)broker_rank);
-    if (!(f = flux_kvs_lookup (h, FLUX_KVS_READDIR, key)))
-        return -1;
-    if (flux_future_then (f, -1., runevent_fallback_continuation, job) < 0) {
-        flux_future_destroy (f);
-        return -1;;
-    }
-    return 0;
-}
-
 /* Handle response to lookup of R_lite.  If this node is targetted,
  * spawn wrexecd.  If R_lite doesn't exist, fallback to old method
  * of looking up rank.N, with one more continuation.
@@ -789,10 +750,6 @@ static void runevent_continuation (flux_future_t *f, void *arg)
     if (flux_kvs_lookup_get (f, &R_lite) < 0) {
         if (broker_rank == 0)
             flux_log (h, LOG_INFO, "No %s: %s", key, flux_strerror (errno));
-        if (runevent_fallback (h, job) < 0) {
-            flux_log_error (h, "%s: fallback failed", __FUNCTION__);
-            goto done;
-        }
         goto done;
     }
     if (!(r = rcalc_create (R_lite))) {

t/t2000-wreck.t

@@ -223,25 +223,11 @@ test_expect_success 'wreckrun: -t2 -N${SIZE} sets correct ntasks in kvs' '
     n=$(flux kvs get --json ${LWJ}.ngpus) &&
     test "$n" = "4"
 '
-
-test_expect_success 'wreckrun: fallback to old rank.N.cores format works' '
-	flux wreckrun -N2 -n2 \
-             -P "lwj[\"rank.0.cores\"] = 1; lwj[\"rank.1.cores\"] = 1; lwj.R_lite = nil" \
-	     /bin/echo hello >oldrankN.out &&
-	LWJ=$(last_job_path) &&
-	test_must_fail flux kvs get ${LWJ}.R_lite &&
-	cat <<-EOF >oldrankN.expected &&
-	hello
-	hello
-	EOF
-	test_cmp oldrankN.expected oldrankN.out
-'
 test_expect_success 'wreckrun: job with more nodes than tasks fails' '
 	test_must_fail flux wreckrun -n2 \
-	    -P "for i=1,3 do lwj[\"rank.\"..i..\".cores\"] = 1 end; lwj.R_lite = nil" \
+	    -P "t={}; for i=1,3 do t[i]={rank=i-1,children={core=\"0\"}} end; lwj.R_lite = t" \
 	    hostname &&
 	LWJ=$(last_job_path) &&
-	test_must_fail flux kvs get ${LWJ}.R_lite &&
 	test "$(flux kvs get --json ${LWJ}.state)" = "failed"
 '
 cpus_allowed=${SHARNESS_TEST_SRCDIR}/scripts/cpus-allowed.lua