Fix a race condition caused by a bad attribute flag that created an OR instead of an AND condition check

This commit was SVN r32587.

orte/mca/errmgr/default_orted/errmgr_default_orted.c

@@ -396,7 +396,8 @@ static void proc_errors(int fd, short args, void *cbdata)
             orte_set_attribute(&jdata->attributes, ORTE_JOB_FAIL_NOTIFIED, ORTE_ATTR_LOCAL, NULL, OPAL_BOOL);
         }
         /* if the proc has terminated, notify the state machine */
-        if (ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_COMPLETE) &&
+        if (ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_IOF_COMPLETE) &&
+            ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_WAITPID) &&
             !ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_RECORDED)) {
             ORTE_ACTIVATE_PROC_STATE(proc, ORTE_PROC_STATE_TERMINATED);
         }
@@ -493,7 +494,8 @@ static void proc_errors(int fd, short args, void *cbdata)
             orte_set_attribute(&jdata->attributes, ORTE_JOB_FAIL_NOTIFIED, ORTE_ATTR_LOCAL, NULL, OPAL_BOOL);
         }
         /* if the proc has terminated, notify the state machine */
-        if (ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_COMPLETE) &&
+        if (ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_IOF_COMPLETE) &&
+            ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_WAITPID) &&
             !ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_RECORDED)) {
             ORTE_ACTIVATE_PROC_STATE(proc, ORTE_PROC_STATE_TERMINATED);
         }

orte/mca/odls/base/odls_base_default_fns.c

@@ -1873,7 +1873,8 @@ int orte_odls_base_default_kill_local_procs(opal_pointer_array_t *procs,
             /* check for everything complete - this will remove
              * the child object from our local list
              */
-            if (ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_COMPLETE)) {
+            if (ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_IOF_COMPLETE) &&
+                ORTE_FLAG_TEST(child, ORTE_PROC_FLAG_WAITPID)) {
                 ORTE_ACTIVATE_PROC_STATE(&child->name, child->state);
             }
         }

orte/mca/state/orted/state_orted.c

@@ -282,7 +282,7 @@ static void track_procs(int fd, short argc, void *cbdata)
         if (NULL != orte_iof.close) {
             orte_iof.close(proc, ORTE_IOF_STDIN);
         }
-        if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_COMPLETE) &&
+        if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_WAITPID) &&
             !ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_RECORDED)) {
             ORTE_ACTIVATE_PROC_STATE(proc, ORTE_PROC_STATE_TERMINATED);
         }
@@ -292,7 +292,7 @@ static void track_procs(int fd, short argc, void *cbdata)
          * successful launch for short-lived procs
          */
         ORTE_FLAG_SET(pdata, ORTE_PROC_FLAG_WAITPID);
-        if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_COMPLETE) &&
+        if (ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_IOF_COMPLETE) &&
             !ORTE_FLAG_TEST(pdata, ORTE_PROC_FLAG_RECORDED)) {
             ORTE_ACTIVATE_PROC_STATE(proc, ORTE_PROC_STATE_TERMINATED);
         }

orte/util/attr.h

@@ -134,7 +134,6 @@ typedef uint16_t orte_proc_flags_t;
 #define ORTE_PROC_FLAG_AS_MPI        0x0080  // proc is MPI process
 #define ORTE_PROC_FLAG_IOF_COMPLETE  0x0100  // IOF has completed
 #define ORTE_PROC_FLAG_WAITPID       0x0200  // waitpid fired
-#define ORTE_PROC_FLAG_COMPLETE      0x0300  // both IOF and waitpid have been reported
 #define ORTE_PROC_FLAG_RECORDED      0x0400  // termination has been recorded