Merge pull request #3104 from garlick/module_set_running

libflux: add flux_module_set_running()

etc/gen-cmdhelp.py

@@ -31,7 +31,7 @@
 visited = dict()
 
 for (path, cmd, descr, author, section) in man_pages:
-    if section is not 1 or path in visited:
+    if section != 1 or path in visited:
         continue
     visited[path] = True
     with open(f"{docsdir}/{path}.rst", "r", encoding='utf-8') as f:

src/common/libflux/module.c

@@ -19,6 +19,7 @@
 
 #include "module.h"
 #include "message.h"
+#include "keepalive.h"
 #include "rpc.h"
 
 #include "src/common/libutil/log.h"
@@ -125,6 +126,25 @@ bool flux_module_debug_test (flux_t *h, int flag, bool clear)
     return true;
 }
 
+int flux_module_set_running (flux_t *h)
+{
+    flux_msg_t *msg;
+    int rc = -1;
+
+    if (!h) {
+        errno = EINVAL;
+        return -1;
+    }
+    if (!(msg = flux_keepalive_encode (0, FLUX_MODSTATE_RUNNING)))
+        return -1;
+    if (flux_send (h, msg, 0) < 0)
+        goto done;
+    rc = 0;
+done:
+    flux_msg_decref (msg);
+    return rc;
+}
+
 /*
  * vi:tabstop=4 shiftwidth=4 expandtab
  */

src/common/libflux/module.h

@@ -62,6 +62,13 @@ char *flux_modfind (const char *searchpath, const char *modname,
  */
 bool flux_module_debug_test (flux_t *h, int flag, bool clear);
 
+/* Set module state to RUNNING.  This transition occurs automatically when the
+ * reactor is entered, but this function can set the state to RUNNING early,
+ * e.g. if flux module load must complete before the module enters the reactor.
+ * Returns 0 on success, -1 on error with errno set.
+ */
+int flux_module_set_running (flux_t *h);
+
 #ifdef __cplusplus
 }
 #endif

src/common/libflux/test/module.c

@@ -20,6 +20,7 @@
 #include <flux/core.h>
 
 #include "src/common/libtap/tap.h"
+#include "src/common/libtestutil/util.h"
 
 /* N.B. FAKE1 and FAKE2 are defined with -D on the CC command line.
  * They are set to the full path of two test modules, module_fake1.so
@@ -182,6 +183,22 @@ void test_debug (void)
     flux_handle_destroy (h);
 }
 
+void test_set_running (void)
+{
+    flux_t *h;
+
+    if (!(h = loopback_create (0)))
+        BAIL_OUT ("loopback_create failed");
+
+    ok (flux_module_set_running (h) == 0,
+        "flux_module_set_running returns success");
+    errno = 0;
+    ok (flux_module_set_running (NULL) < 0 && errno == EINVAL,
+        "flux_module_set_running h=NULL fails with EINVAL");
+
+    flux_close (h);
+}
+
 int main (int argc, char *argv[])
 {
 
@@ -190,6 +207,7 @@ int main (int argc, char *argv[])
     test_modname ();
     test_modfind ();
     test_debug ();
+    test_set_running ();
 
     done_testing();
     return (0);

src/modules/sched-simple/sched.c

@@ -643,6 +643,12 @@ int mod_main (flux_t *h, int argc, char **argv)
     zlistx_set_comparator (ss->queue, jobreq_cmp);
     zlistx_set_destructor (ss->queue, jobreq_destructor);
 
+    /* Let `flux module load simple-sched` return before synchronous
+     * initialization with resource and job-manager modules.
+     */
+    if (flux_module_set_running (h) < 0)
+        goto done;
+
     if (simple_sched_init (h, ss) < 0)
         goto done;
     if (flux_msg_handler_addvec (h, htab, ss, &handlers) < 0) {

t/Makefile.am

@@ -287,6 +287,7 @@ endif
 check_LTLIBRARIES = \
 	module/parent.la \
 	module/child.la \
+	module/running.la \
 	request/req.la \
 	ingest/job-manager-dummy.la \
 	job-manager/sched-dummy.la \
@@ -456,6 +457,12 @@ module_child_la_LDFLAGS = $(fluxmod_ldflags) -module -rpath /nowher
 module_child_la_LIBADD = \
 	$(test_ldadd) $(LIBDL) $(LIBUTIL)
 
+module_running_la_SOURCES = module/running.c
+module_running_la_CPPFLAGS = $(test_cppflags)
+module_running_la_LDFLAGS = $(fluxmod_ldflags) -module -rpath /nowher
+module_running_la_LIBADD = \
+	$(test_ldadd) $(LIBDL) $(LIBUTIL)
+
 barrier_tbarrier_SOURCES = barrier/tbarrier.c
 barrier_tbarrier_CPPFLAGS = $(test_cppflags)
 barrier_tbarrier_LDADD = \

t/module/running.c

@@ -0,0 +1,47 @@
+/************************************************************\
+ * Copyright 2020 Lawrence Livermore National Security, LLC
+ * (c.f. AUTHORS, NOTICE.LLNS, COPYING)
+ *
+ * This file is part of the Flux resource manager framework.
+ * For details, see https://github.com/flux-framework.
+ *
+ * SPDX-License-Identifier: LGPL-3.0
+\************************************************************/
+
+#if HAVE_CONFIG_H
+#include "config.h"
+#endif
+#include <flux/core.h>
+
+int mod_main (flux_t *h, int argc, char *argv[])
+{
+    flux_msg_t *msg;
+    struct flux_match match;
+
+    if (flux_event_subscribe (h, "running.go") < 0)
+        return -1;
+    if (flux_module_set_running (h) < 0)
+        return -1;
+    match = FLUX_MATCH_EVENT;
+    match.topic_glob = "running.go";
+    if (!(msg = flux_recv (h, match, 0))) {
+        flux_log_error (h, "flux_recv");
+        return -1;
+    }
+    flux_log (h, LOG_DEBUG, "received event");
+    flux_msg_destroy (msg);
+
+    if (flux_reactor_run (flux_get_reactor (h), 0) < 0) {
+        flux_log_error (h, "flux_reactor_run");
+        return -1;
+    }
+    return 0;
+}
+
+/* This dso extends a comms module named "parent".
+ */
+MOD_NAME ("running");
+
+/*
+ * vi:tabstop=4 shiftwidth=4 expandtab
+ */

t/t0003-module.t

@@ -267,4 +267,15 @@ test_expect_success 'flux module load "noexist" fails' '
 	grep -q "not found" noexist.out
 '
 
+test_expect_success 'flux_module_set_running - load test module' '
+	run_timeout 10 \
+		flux module load ${FLUX_BUILD_DIR}/t/module/.libs/running.so
+'
+test_expect_success 'flux_module_set_running - signal module to enter reactor' '
+	flux event pub running.go
+'
+test_expect_success 'flux_module_set_running - remove test module' '
+	flux module remove running
+'
+
 test_done