Quiet errors that aren't really errors

src/slurmd/common/stepd_api.c

@@ -181,6 +181,7 @@ stepd_signal_container(step_loc_t step, void *auth_cred, int signal)
 	Buf buf;
 	int buf_len;
 	int rc;
+	int errnum = 0;
 
 	fd = step_connect(step);
 	if (fd == -1)
@@ -197,11 +198,13 @@ stepd_signal_container(step_loc_t step, void *auth_cred, int signal)
 	safe_write(fd, &buf_len, sizeof(int));
 	safe_write(fd, get_buf_data(buf), buf_len);
 
-	/* Receive the return code */
+	/* Receive the return code and errno */
 	safe_read(fd, &rc, sizeof(int));
+	safe_read(fd, &errnum, sizeof(int));
 
 	free_buf(buf);
 	close(fd);
+	errno = errnum;
 	return rc;
 rwfail:
 	close(fd);

src/slurmd/common/stepd_api.h

@@ -121,7 +121,7 @@ pid_t stepd_daemon_pid(step_loc_t step);
 
 #define safe_read(fd, ptr, size) do {					\
 		if (read(fd, ptr, size) != size) {			\
-			error("%s:%d: %s: read (%d bytes) failed: %m",	\
+			debug("%s:%d: %s: read (%d bytes) failed: %m",	\
 			      __FILE__, __LINE__, __CURRENT_FUNC__,	\
 			      (int)size);				\
 			goto rwfail;					\
@@ -130,7 +130,7 @@ pid_t stepd_daemon_pid(step_loc_t step);
 
 #define safe_write(fd, ptr, size) do {					\
 		if (write(fd, ptr, size) != size) {			\
-			error("%s:%d: %s: write (%d bytes) failed: %m",	\
+			debug("%s:%d: %s: write (%d bytes) failed: %m",	\
 			      __FILE__, __LINE__, __CURRENT_FUNC__,	\
 			      (int)size);				\
 			goto rwfail;					\

src/slurmd/slurmd/req.c

@@ -1068,7 +1068,7 @@ _kill_all_active_steps(void *auth_cred, uint32_t jobid, int sig, bool batch)
 		debug2("container signal %d to job %u.%u",
 		       sig, jobid, stepd->stepid);
 		if (stepd_signal_container(*stepd, auth_cred, sig) < 0)
-			error("kill jid %d: %m", jobid);
+			debug("kill jid %d: %m", jobid);
 	}
 	list_iterator_destroy(i);
 	list_destroy(steps);

src/slurmd/slurmstepd/req.c

@@ -509,7 +509,8 @@ _handle_signal_container(int fd, slurmd_job_t *job)
 		debug("kill container req from uid %ld for job %u.%u "
 		      "owned by uid %ld",
 		      (long)uid, job->jobid, job->stepid, (long)job->uid);
-		rc = EPERM;
+		rc = -1;
+		errno = EPERM;
 		goto done;
 	}
 
@@ -519,7 +520,8 @@ _handle_signal_container(int fd, slurmd_job_t *job)
 	if (job->cont_id == 0) {
 		debug ("step %u.%u invalid container [cont_id:%u]", 
 			job->jobid, job->stepid, job->cont_id);
-		rc = ESLURMD_JOB_NOTRUNNING;
+		rc = -1;
+		errno = ESLURMD_JOB_NOTRUNNING;
 		goto done;
 	}
 
@@ -534,8 +536,9 @@ _handle_signal_container(int fd, slurmd_job_t *job)
 	}
 
 done:
-	/* Send the return code */
+	/* Send the return code and errno */
 	safe_write(fd, &rc, sizeof(int));
+	safe_write(fd, &errno, sizeof(int));
 rwfail:
 	return;
 }