From 6b371f524d4d5a2e9d9d02240bbfab82c5e4ed59 Mon Sep 17 00:00:00 2001
From: Danny Auble <da@llnl.gov>
Date: Tue, 17 Oct 2006 18:45:30 +0000
Subject: [PATCH] fixes after the merge
---
src/common/forward.c | 18 +++++++++-------
src/common/slurm_protocol_api.c | 38 ++++++++++++++++-----------------
src/common/slurm_protocol_api.h | 5 ++---
src/slurmctld/agent.c | 1 -
src/slurmd/slurmd/slurmd.c | 3 +--
src/srun/allocate.c | 4 ++--
src/srun/opt.c | 2 +-
7 files changed, 35 insertions(+), 36 deletions(-)
diff --git a/src/common/forward.c b/src/common/forward.c
index ec62d72a708..428ff914945 100644
--- a/src/common/forward.c
+++ b/src/common/forward.c
@@ -107,13 +107,7 @@ void *_forward_thread(void *arg)
xfree(fwd_msg->header.forward.nodelist);
fwd_msg->header.forward.nodelist = xstrdup(buf);
fwd_msg->header.forward.cnt = hostlist_count(hl);
- if(fwd_msg->header.forward.cnt>0) {
- steps = (fwd_msg->header.forward.cnt+1) /
- slurm_get_tree_width();
- fwd_msg->timeout = (1000*steps);
- steps++;
- fwd_msg->timeout += (start_timeout*steps);
- }
+
debug3("forward: along with %s",
fwd_msg->header.forward.nodelist);
@@ -167,7 +161,15 @@ void *_forward_thread(void *arg)
}
goto cleanup;
}
-
+
+ if(fwd_msg->header.forward.cnt>0) {
+ steps = (fwd_msg->header.forward.cnt+1) /
+ slurm_get_tree_width();
+ fwd_msg->timeout = (1000*steps);
+ steps++;
+ fwd_msg->timeout += (start_timeout*steps);
+ }
+
ret_list = slurm_receive_msgs(fd, steps, fwd_msg->timeout);
if(!ret_list || (fwd_msg->header.forward.cnt != 0
diff --git a/src/common/slurm_protocol_api.c b/src/common/slurm_protocol_api.c
index 97a24bd2ab5..b67819f9de8 100644
--- a/src/common/slurm_protocol_api.c
+++ b/src/common/slurm_protocol_api.c
@@ -839,7 +839,7 @@ int slurm_receive_msg(slurm_fd fd, slurm_msg_t *msg, int timeout)
/* Forward message to other nodes */
if(header.forward.cnt > 0) {
error("We need to forward this to other nodes use "
- "slurm_receive_and_forward_msgs instead");
+ "slurm_receive_msg_and_forward instead");
}
if ((auth_cred = g_slurm_auth_unpack(buffer)) == NULL) {
@@ -925,8 +925,9 @@ List slurm_receive_msgs(slurm_fd fd, int steps, int timeout)
/* convert secs to msec */
timeout = slurm_get_msg_timeout() * 1000;
if(steps) {
- steps++;
orig_timeout = timeout/steps;
+ steps--;
+ orig_timeout -= (1000*steps);
}
debug4("orig_timeout was %d we have %d steps and a timeout of %d",
orig_timeout, steps, timeout);
@@ -1055,12 +1056,11 @@ total_return:
* IN open_fd - file descriptor to receive msg on
* IN/OUT msg - a slurm_msg struct to be filled in by the function
* we use the orig_addr from this var for forwarding.
- * IN steps - how many steps down the tree we have to wait for
* IN timeout - how long to wait in milliseconds
* RET int - returns 0 on success, -1 on failure and sets errno
*/
-int slurm_receive_and_forward_msgs(slurm_fd fd, slurm_addr *orig_addr,
- slurm_msg_t *msg, int steps, int timeout)
+int slurm_receive_msg_and_forward(slurm_fd fd, slurm_addr *orig_addr,
+ slurm_msg_t *msg, int timeout)
{
char *buf = NULL;
size_t buflen = 0;
@@ -1068,8 +1068,6 @@ int slurm_receive_and_forward_msgs(slurm_fd fd, slurm_addr *orig_addr,
int rc;
void *auth_cred = NULL;
Buf buffer;
- ret_data_info_t *ret_data_info = NULL;
- int orig_timeout = timeout;
xassert(fd >= 0);
@@ -1091,21 +1089,16 @@ int slurm_receive_and_forward_msgs(slurm_fd fd, slurm_addr *orig_addr,
if (timeout <= 0)
/* convert secs to msec */
timeout = slurm_get_msg_timeout() * 1000;
- if(steps) {
- steps++;
- orig_timeout = timeout/steps;
- }
-
- if(orig_timeout >= (slurm_get_msg_timeout() * 10000)) {
+
+ if(timeout >= (slurm_get_msg_timeout() * 10000)) {
error("You are sending a message with timeout's greater "
"than %d seconds, your's is %d seconds",
(slurm_get_msg_timeout() * 10),
(timeout/1000));
- } else if(orig_timeout < 1000) {
+ } else if(timeout < 1000) {
debug("You are sending a message with a very short timeout of "
"%d milliseconds", timeout);
- }
-
+ }
/*
* Receive a msg. slurm_msg_recvfrom() will read the message
@@ -1134,14 +1127,21 @@ int slurm_receive_and_forward_msgs(slurm_fd fd, slurm_addr *orig_addr,
rc = SLURM_PROTOCOL_VERSION_ERROR;
goto total_return;
}
- //info("ret_cnt = %d",header.ret_cnt);
if(header.ret_cnt > 0) {
- while((ret_data_info = list_pop(header.ret_list)))
- list_push(msg->ret_list, ret_data_info);
+ error("we recieved more than one message back use "
+ "slurm_receive_msgs instead");
header.ret_cnt = 0;
list_destroy(header.ret_list);
header.ret_list = NULL;
}
+ //info("ret_cnt = %d",header.ret_cnt);
+ /* if(header.ret_cnt > 0) { */
+/* while((ret_data_info = list_pop(header.ret_list))) */
+/* list_push(msg->ret_list, ret_data_info); */
+/* header.ret_cnt = 0; */
+/* list_destroy(header.ret_list); */
+/* header.ret_list = NULL; */
+/* } */
/*
* header.orig_addr will be set to where the first message
* came from if this is a forward else we set the
diff --git a/src/common/slurm_protocol_api.h b/src/common/slurm_protocol_api.h
index 0e343c525cc..2259188da8e 100644
--- a/src/common/slurm_protocol_api.h
+++ b/src/common/slurm_protocol_api.h
@@ -351,12 +351,11 @@ List slurm_receive_msgs(slurm_fd fd, int steps, int timeout);
*
* IN open_fd - file descriptor to receive msg on
* OUT resp - a slurm_msg struct to be filled in by the function
- * IN steps - how many steps down the tree we have to wait for
* IN timeout - how long to wait in milliseconds
* RET int - returns 0 on success, -1 on failure and sets errno
*/
-int slurm_receive_and_forward_msgs(slurm_fd fd, slurm_addr *orig_addr,
- slurm_msg_t *resp, int steps, int timeout);
+int slurm_receive_msg_and_forward(slurm_fd fd, slurm_addr *orig_addr,
+ slurm_msg_t *resp, int timeout);
/**********************************************************************\
* send message functions
diff --git a/src/slurmctld/agent.c b/src/slurmctld/agent.c
index f3ef27f7e2c..554cafa55cb 100644
--- a/src/slurmctld/agent.c
+++ b/src/slurmctld/agent.c
@@ -1029,7 +1029,6 @@ static void _queue_agent_retry(agent_info_t * agent_info_ptr, int count)
j = 0;
for (i = 0; i < agent_info_ptr->thread_count; i++) {
if(!thread_ptr[i].ret_list) {
- char ip_buf[32];
if (thread_ptr[i].state != DSH_NO_RESP)
continue;
diff --git a/src/slurmd/slurmd/slurmd.c b/src/slurmd/slurmd/slurmd.c
index fea862f71cb..395ca18322b 100644
--- a/src/slurmd/slurmd/slurmd.c
+++ b/src/slurmd/slurmd/slurmd.c
@@ -364,8 +364,7 @@ _service_connection(void *arg)
debug3("in the service_connection");
slurm_msg_t_init(msg);
- if((rc = slurm_receive_and_forward_msgs(
- con->fd, con->cli_addr, msg, 0, 0))
+ if((rc = slurm_receive_msg_and_forward(con->fd, con->cli_addr, msg, 0))
!= SLURM_SUCCESS) {
error("service_connection: slurm_receive_msg: %m");
/* if this fails we need to make sure the nodes we forward
diff --git a/src/srun/allocate.c b/src/srun/allocate.c
index 714195fb29e..788296e7a31 100644
--- a/src/srun/allocate.c
+++ b/src/srun/allocate.c
@@ -667,10 +667,10 @@ create_job_step(srun_job_t *job)
/* Number of hosts in job may not have been initialized yet if
- * --jobid was used or only SLURM_JOBID was set in user env.
+ * --jobid was used or only SLURM_JOBID was set in user env.
* Reset the value here just in case.
*/
- job->nhosts = job->step_layout->num_hosts;
+ job->nhosts = job->step_layout->node_cnt;
if(!job->step_layout) {
error("step_layout not returned");
diff --git a/src/srun/opt.c b/src/srun/opt.c
index abab6989bc3..ddc346f0407 100644
--- a/src/srun/opt.c
+++ b/src/srun/opt.c
@@ -923,7 +923,7 @@ static void _opt_default()
opt.max_launch_time = 120;/* 120 seconds to launch job */
opt.max_exit_timeout= 60; /* Warn user 60 seconds after task exit */
/* Default launch msg timeout */
- opt.msg_timeout = SLURM_MESSAGE_TIMEOUT_SEC_STATIC;
+ opt.msg_timeout = slurm_get_msg_timeout();
for (i=0; i<SYSTEM_DIMENSIONS; i++)
opt.geometry[i] = (uint16_t) NO_VAL;
--
GitLab