From 042fe16d4280430f6f649664161f2cbb706dff4e Mon Sep 17 00:00:00 2001
From: Mark Grondona <mgrondona@llnl.gov>
Date: Mon, 5 Aug 2002 22:01:01 +0000
Subject: [PATCH] o more changes in srun related to signals
---
src/srun/launch.c | 1 +
src/srun/msg.c | 8 ++++++--
src/srun/srun.c | 36 +++++++++++++++++++++++-------------
3 files changed, 30 insertions(+), 15 deletions(-)
diff --git a/src/srun/launch.c b/src/srun/launch.c
index 33d112f8590..b69633265f8 100644
--- a/src/srun/launch.c
+++ b/src/srun/launch.c
@@ -82,6 +82,7 @@ launch(void *arg)
msg.tasks_to_launch = job->ntask[i];
msg.global_task_ids =
(uint32_t *) xmalloc(job->ntask[i]*sizeof(uint32_t));
+ msg.srun_node_id = (uint32_t)i;
for (j = 0; j < job->ntask[i]; j++)
msg.global_task_ids[j] = taskid++;
diff --git a/src/srun/msg.c b/src/srun/msg.c
index 9d489460114..2a50f88a829 100644
--- a/src/srun/msg.c
+++ b/src/srun/msg.c
@@ -18,14 +18,18 @@ _launch_handler(job_t *job, slurm_msg_t *resp)
launch_tasks_response_msg_t *msg =
(launch_tasks_response_msg_t *) resp->data;
- debug2("recieved launch resp from %s", msg->node_name);
+ debug2("recieved launch resp from %s nodeid=%d", msg->node_name,
+ msg->srun_node_id);
if (msg->return_code != 0) {
error("recvd return code %d from %s", msg->return_code,
msg->node_name);
return;
} else {
- /* job->host_state[msg->host_id] = SRUN_HOST_REPLIED; */
+
+ if (msg->srun_node_id > 0 && msg->srun_node_id < job->nhosts)
+ job->host_state[msg->srun_node_id] =
+ SRUN_HOST_REPLIED;
}
}
diff --git a/src/srun/srun.c b/src/srun/srun.c
index 348cccdf2e9..d12424b1214 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -180,8 +180,10 @@ main(int ac, char **av)
pthread_kill(job->lid, SIGTERM);
pthread_kill(job->jtid, SIGTERM);
- pthread_kill(job->ioid, SIGTERM);
pthread_kill(job->sigid, SIGTERM);
+ fflush(stderr);
+ fflush(stdout);
+ pthread_kill(job->ioid, SIGTERM);
exit(0);
}
@@ -227,7 +229,7 @@ allocate_nodes(void)
if (rc == SLURM_FAILURE) {
error("Unable to allocate resources: %s",
slurm_strerror(errno));
- exit(1);
+ return NULL;
}
return resp;
@@ -320,26 +322,34 @@ sig_thr(void *arg)
{
job_t *job = (job_t *)arg;
sigset_t set;
+ static time_t last_intr = 0;
int signo;
struct sigaction action;
while (1) {
sigfillset(&set);
- pthread_sigmask(SIG_UNBLOCK, &set, NULL);
+ pthread_sigmask(SIG_BLOCK, &set, NULL);
sigwait(&set, &signo);
debug2("recvd signal %d", signo);
switch (signo) {
- case SIGINT:
- fwd_signal(job, SIGINT);
- pthread_mutex_lock(&job->state_mutex);
- job->state = SRUN_JOB_OVERDONE;
- pthread_cond_signal(&job->state_cond);
- pthread_mutex_unlock(&job->state_mutex);
- break;
case SIGTERM:
pthread_exit(0);
break;
+ case SIGINT:
+ if (time(NULL) - last_intr > 1) {
+ info("sending Ctrl-C to remote tasks");
+ last_intr = time(NULL);
+ fwd_signal(job, signo);
+ } else { /* second Ctrl-C in half as many seconds */
+ /* terminate job */
+ info("forcing termination");
+ pthread_mutex_lock(&job->state_mutex);
+ job->state = SRUN_JOB_OVERDONE;
+ pthread_cond_signal(&job->state_cond);
+ pthread_mutex_unlock(&job->state_mutex);
+ }
+ break;
default:
fwd_signal(job, signo);
break;
@@ -349,14 +359,14 @@ sig_thr(void *arg)
pthread_exit(0);
}
-void
+ void
fwd_signal(job_t *job, int signo)
{
int i;
slurm_msg_t req;
slurm_msg_t resp;
kill_tasks_msg_t msg;
-
+
debug("forward signal %d to job", signo);
req.msg_type = REQUEST_KILL_TASKS;
@@ -368,7 +378,7 @@ fwd_signal(job_t *job, int signo)
for (i = 0; i < job->nhosts; i++) {
slurm_set_addr_uint(&req.address, slurm_get_slurmd_port(),
- ntohl(job->iaddr[i]));
+ ntohl(job->iaddr[i]));
debug("sending kill req to %s", job->host[i]);
if (slurm_send_recv_node_msg(&req, &resp) < 0)
error("Unable to send signal to host %s",
--
GitLab