Merge pull request #3486 from garlick/flux_resource_fixes

flux resource: allow drain, undrain, and status to work on any rank
flux-framework · Jan 26, 2021 · cbb77bd · cbb77bd
2 parents 4ca5702 + 7b8b17e
commit cbb77bd
Show file tree

Hide file tree

Showing 4 changed files with 58 additions and 12 deletions.
diff --git a/src/cmd/flux-resource.py b/src/cmd/flux-resource.py
@@ -48,14 +48,15 @@ def drain(args):
         flux.Flux(),
         "resource.drain",
         {"targets": args.targets, "reason": " ".join(args.reason)},
+        nodeid=0,
     ).get()
 
 
 def undrain(args):
     """
     Send an "undrain" request to resource module for args.targets
     """
-    RPC(flux.Flux(), "resource.undrain", {"targets": args.targets}).get()
+    RPC(flux.Flux(), "resource.undrain", {"targets": args.targets}, nodeid=0).get()
 
 
 class StatusLine:
@@ -273,7 +274,7 @@ def status(args):
     if args.from_stdin:
         resp = sys.stdin.read()
     else:
-        resp = RPC(flux.Flux(), "resource.status").get()
+        resp = RPC(flux.Flux(), "resource.status", nodeid=0).get()
 
     rstat = ResourceStatus.from_status_response(resp, fmt)
 

diff --git a/src/modules/resource/resource.c b/src/modules/resource/resource.c
@@ -156,15 +156,18 @@ static void status_cb (flux_t *h,
                        void *arg)
 {
     struct resource_ctx *ctx = arg;
-    const struct idset *online = monitor_get_up (ctx->monitor);
-    const struct idset *offline = monitor_get_down (ctx->monitor);
-    const struct idset *exclude = exclude_get (ctx->exclude);
     json_t *drain;
     const json_t *R;
     json_t *o = NULL;
+    const char *errstr = NULL;
 
     if (flux_request_decode (msg, NULL, NULL) < 0)
         goto error;
+    if (ctx->rank != 0) {
+        errno = EPROTO;
+        errstr = "this RPC only works on rank 0";
+        goto error;
+    }
     if (!(R = inventory_get (ctx->inventory)))
         goto error;
     if (!(drain = drain_get_info (ctx->drain)))
@@ -174,19 +177,25 @@ static void status_cb (flux_t *h,
         errno = ENOMEM;
         goto error;
     }
-    if (rutil_set_json_idset (o, "online", online) < 0)
+    if (rutil_set_json_idset (o,
+                              "online",
+                              monitor_get_up (ctx->monitor)) < 0)
         goto error;
-    if (rutil_set_json_idset (o, "offline", offline) < 0)
+    if (rutil_set_json_idset (o,
+                              "offline",
+                              monitor_get_down (ctx->monitor)) < 0)
         goto error;
-    if (rutil_set_json_idset (o, "exclude", exclude) < 0)
+    if (rutil_set_json_idset (o,
+                              "exclude",
+                              exclude_get (ctx->exclude)) < 0)
         goto error;
     if (flux_respond_pack (h, msg, "o", o) < 0) {
         flux_log_error (h, "error responding to resource.status request");
         json_decref (o);
     }
     return;
 error:
-    if (flux_respond_error (h, msg, errno, NULL) < 0)
+    if (flux_respond_error (h, msg, errno, errstr) < 0)
         flux_log_error (h, "error responding to resource.status request");
     json_decref (o);
 }

diff --git a/t/t2310-resource-module.t b/t/t2310-resource-module.t
@@ -113,8 +113,8 @@ test_expect_success 'flux resource reload fails on nonexistent XML directory' '
 '
 
 test_expect_success 'flux resource reload fails on empty XML directory' '
-    mkdir empty &&
-    test_must_fail flux resource reload -x $(pwd)/empty
+	mkdir empty &&
+	test_must_fail flux resource reload -x $(pwd)/empty
 '
 
 test_expect_success HAVE_JQ 'extract hwloc XML from JSON object' '
@@ -192,7 +192,20 @@ test_expect_success HAVE_JQ 'all ranks were drained' '
 '
 
 test_expect_success 'resource.get-xml blocks until all ranks are up' '
-    flux python ${SHARNESS_TEST_SRCDIR}/resource/get-xml-test.py
+	flux python ${SHARNESS_TEST_SRCDIR}/resource/get-xml-test.py
+'
+
+test_expect_success 'flux resource status works on rank > 0' '
+	flux exec -r 1 flux resource status
+'
+
+status_onrank() {
+	flux python -c "import flux; print(flux.Flux().rpc(\"resource.status\",nodeid=$1).get())"
+}
+
+test_expect_success 'resource.status RPC fails on rank > 0' '
+	test_must_fail status_onrank 1 2>status.err &&
+	grep "only works on rank 0" status.err
 '
 
 test_expect_success 'unload resource module' '

diff --git a/t/t2311-resource-drain.t b/t/t2311-resource-drain.t
@@ -128,4 +128,27 @@ test_expect_success 'drain with no args lists currently drained targets' '
 	grep "happy happy, joy joy" drain.out
 '
 
+test_expect_success 'drain/undrain works on rank > 0' '
+	flux exec -r 1 flux resource undrain 0 &&
+	flux exec -r 1 flux resource drain 0 whee drained again
+'
+
+drain_onrank() {
+	local op=$1
+	local nodeid=$2
+	local target=$3
+	flux python -c "import flux; print(flux.Flux().rpc(\"resource.$op\",{\"targets\":$target, \"reason\":\"\"}, nodeid=$nodeid).get())"
+}
+
+test_expect_success 'resource.drain RPC fails on rank > 0' '
+       test_must_fail drain_onrank drain 1 0 2>drain1.err &&
+       grep -i "unknown service method" drain1.err
+'
+
+test_expect_success 'resource.undrain RPC fails on rank > 0' '
+       test_must_fail drain_onrank undrain 1 0 2>undrain1.err &&
+       grep -i "unknown service method" undrain1.err
+'
+
+
 test_done