From d1f9c73edbd05670e5f0a37cf41b445abc31e04c Mon Sep 17 00:00:00 2001
From: "Christopher J. Morrone" <morrone2@llnl.gov>
Date: Thu, 20 Jul 2006 22:45:15 +0000
Subject: [PATCH] Update slurm_launch_tasks to work with new location of task
layout sturcture. poe may have problems with the spawn_task changes.
---
src/api/spawn.c | 34 +++++++++++++++++-----------------
src/api/step_ctx.c | 25 ++++++++++---------------
src/api/step_ctx.h | 3 ---
src/api/step_launch.c | 12 ++++++------
src/slaunch/slaunch.c | 2 +-
5 files changed, 34 insertions(+), 42 deletions(-)
diff --git a/src/api/spawn.c b/src/api/spawn.c
index a8ea540766f..5a343fa8acd 100644
--- a/src/api/spawn.c
+++ b/src/api/spawn.c
@@ -118,8 +118,8 @@ extern int slurm_spawn (slurm_step_ctx ctx, int *fd_array)
}
/* validate fd_array and bind them to ports */
- sock_array = xmalloc(ctx->step_layout->num_hosts * sizeof(int));
- for (i=0; i<ctx->step_layout->num_hosts; i++) {
+ sock_array = xmalloc(ctx->step_resp->step_layout->num_hosts * sizeof(int));
+ for (i=0; i<ctx->step_resp->step_layout->num_hosts; i++) {
if (fd_array[i] < 0) {
slurm_seterrno(EINVAL);
free(sock_array);
@@ -135,14 +135,14 @@ extern int slurm_spawn (slurm_step_ctx ctx, int *fd_array)
}
msg_array_ptr = xmalloc(sizeof(spawn_task_request_msg_t) *
- ctx->step_layout->num_hosts);
+ ctx->step_resp->step_layout->num_hosts);
req_array_ptr = xmalloc(sizeof(slurm_msg_t) *
- ctx->step_layout->num_hosts);
+ ctx->step_resp->step_layout->num_hosts);
hostlist = hostlist_create(ctx->alloc_resp->node_list);
itr = hostlist_iterator_create(hostlist);
- for (i=0; i<ctx->step_layout->num_hosts; i++) {
+ for (i=0; i<ctx->step_resp->step_layout->num_hosts; i++) {
spawn_task_request_msg_t *r = &msg_array_ptr[i];
slurm_msg_t *m = &req_array_ptr[i];
@@ -156,13 +156,13 @@ extern int slurm_spawn (slurm_step_ctx ctx, int *fd_array)
r->envc = ctx->envc;
r->env = ctx->env;
r->cwd = ctx->cwd;
- r->nnodes = ctx->step_layout->num_hosts;
- r->nprocs = ctx->step_layout->num_tasks;
+ r->nnodes = ctx->step_resp->step_layout->num_hosts;
+ r->nprocs = ctx->step_resp->step_layout->num_tasks;
r->switch_job = ctx->step_resp->switch_job;
r->slurmd_debug = slurmd_debug;
/* Task specific message contents */
- r->global_task_id = ctx->step_layout->tids[i][0];
- r->cpus_allocated = ctx->step_layout->cpus[i];
+ r->global_task_id = ctx->step_resp->step_layout->tids[i][0];
+ r->cpus_allocated = ctx->step_resp->step_layout->cpus[i];
r->srun_node_id = (uint32_t) i;
r->io_port = ntohs(sock_array[i]);
m->msg_type = REQUEST_SPAWN_TASK;
@@ -170,7 +170,7 @@ extern int slurm_spawn (slurm_step_ctx ctx, int *fd_array)
j=0;
while((host = hostlist_next(itr))) {
- if(!strcmp(host,ctx->step_layout->host[i])) {
+ if(!strcmp(host,ctx->step_resp->step_layout->host[i])) {
free(host);
break;
}
@@ -178,14 +178,14 @@ extern int slurm_spawn (slurm_step_ctx ctx, int *fd_array)
free(host);
}
debug2("using %d %s with %d tasks\n", j,
- ctx->step_layout->host[i],
+ ctx->step_resp->step_layout->host[i],
r->nprocs);
hostlist_iterator_reset(itr);
memcpy(&m->address, &ctx->alloc_resp->node_addr[j],
sizeof(slurm_addr));
#if _DEBUG
printf("tid=%d, fd=%d, port=%u, node_id=%u\n",
- ctx->step_layout->tids[i][0],
+ ctx->step_resp->step_layout->tids[i][0],
fd_array[i], r->io_port, i);
#endif
}
@@ -317,7 +317,7 @@ static void _dump_ctx(slurm_step_ctx ctx)
}
}
- for (i=0; i<ctx->step_layout->num_hosts; i++) {
+ for (i=0; i<ctx->step_resp->step_layout->num_hosts; i++) {
printf("host=%s cpus=%u tasks=%u",
ctx->host[i], ctx->cpus[i], ctx->tasks[i]);
for (j=0; j<ctx->tasks[i]; j++)
@@ -337,19 +337,19 @@ static int _p_launch(slurm_msg_t *req, slurm_step_ctx ctx)
int rc = SLURM_SUCCESS, i;
thd_t *thd;
- thd = xmalloc(sizeof(thd_t) * ctx->step_layout->num_hosts);
+ thd = xmalloc(sizeof(thd_t) * ctx->step_resp->step_layout->num_hosts);
if (thd == NULL) {
slurm_seterrno(ENOMEM);
return SLURM_ERROR;
}
- for (i=0; i<ctx->step_layout->num_hosts; i++) {
+ for (i=0; i<ctx->step_resp->step_layout->num_hosts; i++) {
thd[i].state = DSH_NEW;
thd[i].req = &req[i];
}
/* start all the other threads (up to _MAX_THREAD_COUNT active) */
- for (i=0; i<ctx->step_layout->num_hosts; i++) {
+ for (i=0; i<ctx->step_resp->step_layout->num_hosts; i++) {
/* wait until "room" for another thread */
slurm_mutex_lock(&thread_mutex);
while (threads_active >= _MAX_THREAD_COUNT) {
@@ -378,7 +378,7 @@ static int _p_launch(slurm_msg_t *req, slurm_step_ctx ctx)
/* wait for all tasks to terminate */
slurm_mutex_lock(&thread_mutex);
- for (i=0; i<ctx->step_layout->num_hosts; i++) {
+ for (i=0; i<ctx->step_resp->step_layout->num_hosts; i++) {
while (thd[i].state < DSH_DONE) {
/* wait until another thread completes*/
pthread_cond_wait(&thread_cond, &thread_mutex);
diff --git a/src/api/step_ctx.c b/src/api/step_ctx.c
index a177dcaf435..f1fbe740a19 100644
--- a/src/api/step_ctx.c
+++ b/src/api/step_ctx.c
@@ -83,19 +83,12 @@ slurm_step_ctx_create (job_step_create_request_msg_t *step_req)
ctx = xmalloc(sizeof(struct slurm_step_ctx_struct));
ctx->launch_state = NULL;
- ctx->step_layout = step_layout_create(alloc_resp, step_resp, step_req);
-
ctx->magic = STEP_CTX_MAGIC;
ctx->job_id = step_req->job_id;
ctx->user_id = step_req->user_id;
ctx->step_req = _copy_step_req(step_req);
ctx->step_resp = step_resp;
ctx->alloc_resp = alloc_resp;
- if (task_layout(ctx->step_layout) != SLURM_SUCCESS) {
- slurm_step_ctx_destroy((slurm_step_ctx)ctx);
- errno = ESLURM_BAD_DIST;
- return NULL;
- }
return (slurm_step_ctx)ctx;
}
@@ -132,18 +125,20 @@ slurm_step_ctx_get (slurm_step_ctx ctx, int ctx_key, ...)
break;
case SLURM_STEP_CTX_TASKS:
uint32_array_pptr = (uint32_t **) va_arg(ap, void *);
- *uint32_array_pptr = ctx->step_layout->tasks;
+ *uint32_array_pptr = ctx->step_resp->step_layout->tasks;
break;
case SLURM_STEP_CTX_TID:
node_inx = va_arg(ap, uint32_t);
- if ((node_inx < 0) || (node_inx > ctx->step_layout->num_hosts)) {
+ if ((node_inx < 0)
+ || (node_inx > ctx->step_resp->step_layout->num_hosts)) {
slurm_seterrno(EINVAL);
rc = SLURM_ERROR;
break;
}
uint32_array_pptr = (uint32_t **) va_arg(ap, void *);
- *uint32_array_pptr = ctx->step_layout->tids[node_inx];
+ *uint32_array_pptr =
+ ctx->step_resp->step_layout->tids[node_inx];
break;
case SLURM_STEP_CTX_RESP:
@@ -161,21 +156,22 @@ slurm_step_ctx_get (slurm_step_ctx ctx, int ctx_key, ...)
break;
case SLURM_STEP_CTX_NUM_HOSTS:
uint32_ptr = (uint32_t *) va_arg(ap, void *);
- *uint32_ptr = ctx->step_layout->num_hosts;
+ *uint32_ptr = ctx->step_resp->step_layout->num_hosts;
break;
case SLURM_STEP_CTX_CPUS:
uint32_array_pptr = (uint32_t **) va_arg(ap, void *);
- *uint32_array_pptr = ctx->step_layout->cpus;
+ *uint32_array_pptr = ctx->step_resp->step_layout->cpus;
break;
case SLURM_STEP_CTX_HOST:
node_inx = va_arg(ap, uint32_t);
- if ((node_inx < 0) || (node_inx > ctx->step_layout->num_hosts)) {
+ if ((node_inx < 0)
+ || (node_inx > ctx->step_resp->step_layout->num_hosts)) {
slurm_seterrno(EINVAL);
rc = SLURM_ERROR;
break;
}
char_array_pptr = (char **) va_arg(ap, void *);
- *char_array_pptr = ctx->step_layout->host[node_inx];
+ *char_array_pptr = ctx->step_resp->step_layout->host[node_inx];
break;
default:
slurm_seterrno(EINVAL);
@@ -276,7 +272,6 @@ slurm_step_ctx_destroy (slurm_step_ctx ctx)
slurm_seterrno(EINVAL);
return SLURM_ERROR;
}
- step_layout_destroy(ctx->step_layout);
_free_step_req(ctx->step_req);
slurm_free_job_step_create_response_msg(ctx->step_resp);
slurm_free_resource_allocation_response_msg(ctx->alloc_resp);
diff --git a/src/api/step_ctx.h b/src/api/step_ctx.h
index d4700a77565..ee733b46edc 100644
--- a/src/api/step_ctx.h
+++ b/src/api/step_ctx.h
@@ -83,9 +83,6 @@ struct slurm_step_ctx_struct {
uint32_t envc; /* count of env vars */
char **env; /* environment variables */
- slurm_step_layout_t *step_layout; /* holds info about how the task is
- laid out */
-
/* Used by slurm_step_launch(), but not slurm_spawn() */
struct step_launch_state *launch_state;
};
diff --git a/src/api/step_launch.c b/src/api/step_launch.c
index 19ba1fe0e5d..5b6895a3a85 100644
--- a/src/api/step_launch.c
+++ b/src/api/step_launch.c
@@ -198,9 +198,9 @@ int slurm_step_launch (slurm_step_ctx ctx,
/* job->step_layout->tasks[i] = 1; */
/* } */
- launch.tasks_to_launch = ctx->step_layout->tasks;
- launch.cpus_allocated = ctx->step_layout->cpus;
- launch.global_task_ids = ctx->step_layout->tids;
+ launch.tasks_to_launch = ctx->step_resp->step_layout->tasks;
+ launch.cpus_allocated = ctx->step_resp->step_layout->cpus;
+ launch.global_task_ids = ctx->step_resp->step_layout->tids;
ctx->launch_state->client_io = _setup_step_client_io(
ctx, params->local_fds, params->labelio);
@@ -559,13 +559,13 @@ static int _launch_tasks(slurm_step_ctx ctx,
msg.ret_list = NULL;
msg.orig_addr.sin_addr.s_addr = 0;
msg.buffer = buffer;
- memcpy(&msg.address, &ctx->alloc_resp->node_addr[0],
+ memcpy(&msg.address, &ctx->step_resp->step_layout->node_addr[0],
sizeof(slurm_addr));
timeout = slurm_get_msg_timeout();
forward_set_launch(&msg.forward,
- ctx->step_req->node_count,
+ ctx->step_resp->step_layout->num_hosts,
&zero,
- ctx->step_layout,
+ ctx->step_resp->step_layout,
itr,
timeout);
hostlist_iterator_destroy(itr);
diff --git a/src/slaunch/slaunch.c b/src/slaunch/slaunch.c
index bf2556f8d05..d5575f0c8ab 100644
--- a/src/slaunch/slaunch.c
+++ b/src/slaunch/slaunch.c
@@ -191,7 +191,7 @@ int slaunch(int argc, char **argv)
/* FIXME - don't peek into the step context, that's cheating! */
_setup_local_fds(¶ms.local_fds, (int)step_ctx->job_id,
(int)step_ctx->step_resp->job_step_id,
- step_ctx->step_layout);
+ step_ctx->step_resp->step_layout);
params.parallel_debug = opt.parallel_debug ? true : false;
params.task_start_callback = _task_start;
params.task_finish_callback = _task_finish;
--
GitLab