From 213059efe7f8d1fc72d5f876ae949730d8e98b16 Mon Sep 17 00:00:00 2001
From: Danny Auble <da@llnl.gov>
Date: Tue, 30 May 2006 16:40:24 +0000
Subject: [PATCH] updated return codes from slurm_send_recv_rc_msg_only_one and
slurm_send_recv_contoller_rc_msg to either 0 or -1 and added error checking
to api/signal.c
---
src/api/signal.c | 15 ++++++++++-----
src/common/slurm_protocol_api.c | 23 ++++++++++++++---------
src/common/slurm_protocol_api.h | 4 +++-
src/slurmd/slurmstepd/mgr.c | 9 ++++-----
4 files changed, 31 insertions(+), 20 deletions(-)
diff --git a/src/api/signal.c b/src/api/signal.c
index 496c4e62df3..0cf04ad1b9e 100644
--- a/src/api/signal.c
+++ b/src/api/signal.c
@@ -274,8 +274,10 @@ static int _signal_batch_script_step(
msg.data = &rpc;
msg.address = allocation->node_addr[0];
- slurm_send_recv_rc_msg_only_one(&msg, &rc, 10);
-
+ if (slurm_send_recv_rc_msg_only_one(&msg, &rc, 0) < 0) {
+ error("_signal_batch_script_step: %m");
+ rc = -1;
+ }
return rc;
}
@@ -351,8 +353,11 @@ _thr_send_recv_rc_msg(void *args)
pthread_cond_t *cond = params->cond;
int *active = params->active;
- slurm_send_recv_rc_msg_only_one(params->msg,
- params->rc, params->timeout);
+ if (slurm_send_recv_rc_msg_only_one(params->msg, params->rc,
+ params->timeout) < 0) {
+ error("_thr_send_recv_rc_msg: %m");
+ *params->rc = -1;
+ }
xfree(args);
slurm_mutex_lock(lock);
@@ -546,7 +551,7 @@ static int _terminate_batch_script_step(
msg.address = allocation->node_addr[0];
i = slurm_send_recv_rc_msg_only_one(&msg, &rc, 10);
- if (i != SLURM_SUCCESS)
+ if (i != 0)
rc = i;
return rc;
diff --git a/src/common/slurm_protocol_api.c b/src/common/slurm_protocol_api.c
index 22f91290f4c..e8bce0ca3d3 100644
--- a/src/common/slurm_protocol_api.c
+++ b/src/common/slurm_protocol_api.c
@@ -1715,7 +1715,7 @@ int slurm_send_recv_rc_msg_only_one(slurm_msg_t *req, int *rc, int timeout)
slurm_fd fd = -1;
List ret_list = NULL;
ret_types_t *ret_type = NULL;
- int ret_c = SLURM_SUCCESS;
+ int ret_c = 0;
forward_init(&req->forward, NULL);
req->ret_list = NULL;
@@ -1723,9 +1723,8 @@ int slurm_send_recv_rc_msg_only_one(slurm_msg_t *req, int *rc, int timeout)
/* no need to init forward_struct_init here */
if ((fd = slurm_open_msg_conn(&req->address)) < 0) {
- return SLURM_SOCKET_ERROR;
+ return -1;
}
-
ret_list = _send_recv_rc_msg(fd, req, timeout);
if(ret_list) {
@@ -1737,12 +1736,15 @@ int slurm_send_recv_rc_msg_only_one(slurm_msg_t *req, int *rc, int timeout)
if(ret_type) {
*rc = ret_type->msg_rc;
- ret_c = ret_type->err;
+ // make sure we only send 0 or -1 for an error
+ if(ret_type->err != 0)
+ //ret_c = ret_type->err;
+ ret_c = -1;
destroy_ret_types(ret_type);
}
list_destroy(ret_list);
} else
- ret_c = SLURM_ERROR;
+ ret_c = -1;
return ret_c;
}
@@ -1754,7 +1756,7 @@ int slurm_send_recv_controller_rc_msg(slurm_msg_t *req, int *rc)
slurm_fd fd = -1;
List ret_list = NULL;
ret_types_t *ret_type = NULL;
- int ret_val = SLURM_SUCCESS;
+ int ret_val = 0;
forward_init(&req->forward, NULL);
req->ret_list = NULL;
@@ -1762,7 +1764,7 @@ int slurm_send_recv_controller_rc_msg(slurm_msg_t *req, int *rc)
/* no need to init forward_struct_init here */
if ((fd = slurm_open_controller_conn()) < 0)
- return SLURM_SOCKET_ERROR;
+ return -1;
ret_list = _send_recv_rc_msg(fd, req, 0);
if(ret_list) {
@@ -1773,12 +1775,15 @@ int slurm_send_recv_controller_rc_msg(slurm_msg_t *req, int *rc)
if(ret_type) {
*rc = ret_type->msg_rc;
- ret_val = ret_type->err;
+ // make sure we only send 0 or -1 for an error
+ if(ret_type->err != 0)
+ //ret_c = ret_type->err;
+ ret_val = -1;
destroy_ret_types(ret_type);
}
list_destroy(ret_list);
} else
- ret_val = SLURM_ERROR;
+ ret_val = -1;
return ret_val;
}
diff --git a/src/common/slurm_protocol_api.h b/src/common/slurm_protocol_api.h
index 537a90a7138..c19225a59c8 100644
--- a/src/common/slurm_protocol_api.h
+++ b/src/common/slurm_protocol_api.h
@@ -557,13 +557,15 @@ List slurm_send_recv_rc_packed_msg(slurm_msg_t *req, int timeout);
List slurm_send_recv_rc_msg(slurm_msg_t *req, int timeout);
/*
- * Same as above, but only to one node
+ * Same as above, but only to one node
+ * returns 0 on success, -1 on failure and sets errno
*/
int slurm_send_recv_rc_msg_only_one(slurm_msg_t *req, int *rc, int timeout);
/*
* Same as above, but send to controller
+ * returns 0 on success, -1 on failure and sets errno
*/
int slurm_send_recv_controller_rc_msg(slurm_msg_t *req, int *rc);
diff --git a/src/slurmd/slurmstepd/mgr.c b/src/slurmd/slurmstepd/mgr.c
index e77486a474a..7c23af6ae54 100644
--- a/src/slurmd/slurmstepd/mgr.c
+++ b/src/slurmd/slurmstepd/mgr.c
@@ -537,8 +537,7 @@ _one_step_complete_msg(slurmd_job_t *job, int first, int last)
/* this is the base of the tree, its parent is slurmctld */
debug3("Rank %d sending complete to slurmctld, range %d to %d",
step_complete.rank, first, last);
- if (slurm_send_recv_controller_rc_msg(&req, &rc)
- != SLURM_SUCCESS)
+ if (slurm_send_recv_controller_rc_msg(&req, &rc) < 0)
error("Rank %d failed sending step completion message"
" to slurmctld (parent)", step_complete.rank);
goto finished;
@@ -554,13 +553,13 @@ _one_step_complete_msg(slurmd_job_t *job, int first, int last)
if (i)
sleep(1);
retcode = slurm_send_recv_rc_msg_only_one(&req, &rc, 10);
- if (retcode == SLURM_SUCCESS && rc == 0)
+ if (retcode == 0 && rc == 0)
goto finished;
}
/* on error AGAIN, send to the slurmctld instead */
debug3("Rank %d sending complete to slurmctld instead, range %d to %d",
step_complete.rank, first, last);
- if (slurm_send_recv_controller_rc_msg(&req, &rc) != SLURM_SUCCESS)
+ if (slurm_send_recv_controller_rc_msg(&req, &rc) < 0)
error("Rank %d failed sending step completion message"
" directly to slurmctld", step_complete.rank);
finished:
@@ -1391,7 +1390,7 @@ _complete_batch_script(slurmd_job_t *job, int err, int status)
/* Note: these log messages don't go to slurmd.log from here */
for (i=0; i<=MAX_RETRY; i++) {
- if (slurm_send_recv_controller_rc_msg(&req_msg, &rc) >= 0)
+ if (slurm_send_recv_controller_rc_msg(&req_msg, &rc) == 0)
break;
info("Retrying job complete RPC for %u.%u",
job->jobid, job->stepid);
--
GitLab