diff --git a/src/srun/io.c b/src/srun/io.c
index 926d607cca793dfe16d73be1c5d15c46c66899f6..444012aa5299bf5c4bf0f85125e4746e1ad40fde 100644
--- a/src/srun/io.c
+++ b/src/srun/io.c
@@ -38,9 +38,6 @@ void *io_thr(void *job_arg)
xassert(job != NULL);
- pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL);
- pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, NULL);
-
if (fcntl(job->iofd, F_SETFL, O_NONBLOCK) < 0)
error("Unable to set nonblocking I/O on fd\n");
@@ -68,7 +65,7 @@ void *io_thr(void *job_arg)
tv.tv_sec = 0;
tv.tv_usec = 500;
- while ((m = select(maxfd+1, &rset, NULL, NULL, &tv)) < 0) {
+ while ((m = select(maxfd+1, &rset, NULL, NULL, NULL)) < 0) {
if (errno != EINTR)
fatal("Unable to handle I/O: %m", errno);
}
diff --git a/src/srun/job.h b/src/srun/job.h
index 210513178ba051759eeb9cfe8c00c377ce4dae3d..ca242fb188381117d427e895c3bbbd62cd1bc9ab 100644
--- a/src/srun/job.h
+++ b/src/srun/job.h
@@ -22,6 +22,7 @@ typedef enum {
typedef enum {
SRUN_HOST_INIT = 0,
SRUN_HOST_CONTACTED,
+ SRUN_HOST_UNREACHABLE,
SRUN_HOST_REPLIED,
SRUN_HOST_DONE
} host_state_t;
diff --git a/src/srun/launch.c b/src/srun/launch.c
index b69633265f832df4e44ff6edad2d035b4f061d30..14ecede3abce5815a5efca7f2c6ecb5ce35ec21b 100644
--- a/src/srun/launch.c
+++ b/src/srun/launch.c
@@ -92,9 +92,10 @@ launch(void *arg)
debug2("launching on host %s", job->host[i]);
print_launch_msg(&msg);
- if (slurm_send_only_node_msg(&req) < 0)
- error("Unable to send launch request: %s",
- slurm_strerror(errno));
+ if (slurm_send_only_node_msg(&req) < 0) {
+ error("%s: %m", job->host[i]);
+ job->host_state[i] = SRUN_HOST_UNREACHABLE;
+ }
xfree(msg.global_task_ids);
}
diff --git a/src/srun/opt.c b/src/srun/opt.c
index 906610b093c8f1eb8375c36dd0dbd54a5e3a1b9a..b7ef2e6c773b7c337777bf6b3ac9872fcd6a4955 100644
--- a/src/srun/opt.c
+++ b/src/srun/opt.c
@@ -18,10 +18,15 @@
#include <pwd.h> /* getpwuid */
#include <ctype.h> /* isdigit */
#include <sys/param.h> /* MAXPATHLEN */
+#include <sys/stat.h>
+#include <unistd.h>
+#include <sys/types.h>
#include <src/common/log.h>
#include <src/common/xmalloc.h>
#include <src/common/xstring.h>
+#include <src/common/list.h>
+
#include "opt.h"
#include "env.h"
@@ -161,10 +166,10 @@ struct poptOption runTable[] = {
"location of stderr redirection",
"err"},
{"verbose", 'v', 0, 0, OPT_VERBOSE,
- "verbose operation", },
- {"debug", 'd', 0, 0, OPT_DEBUG,
+ "verbose operation (multiple -v's increase verbosity)", },
+ /*{"debug", 'd', 0, 0, OPT_DEBUG,
"enable debug",
- },
+ },*/
POPT_TABLEEND
};
@@ -248,6 +253,11 @@ static bool opt_verify(poptContext, bool, bool, bool);
*/
static void opt_list(void);
+/* search PATH for command
+ * returns full path
+ */
+static char * search_path(char *);
+
/*---[ end forward declarations of static functions ]---------------------*/
int initialize_and_process_args(int argc, char *argv[])
@@ -579,6 +589,7 @@ static void opt_args(int ac, char **av)
bool nprocs_set, nnodes_set, cpus_set;
const char **rest;
const char *arg;
+ char *fullpath;
poptContext optctx;
opt.progname = xbasename(av[0]);
@@ -733,6 +744,12 @@ static void opt_args(int ac, char **av)
for (i = 0; i < remote_argc; i++)
remote_argv[i] = strdup(rest[i]);
+ if ((fullpath = search_path(remote_argv[0])) != NULL) {
+ free(remote_argv[0]);
+ remote_argv[0] = fullpath;
+ }
+
+
if (!opt_verify(optctx, nnodes_set, cpus_set, nprocs_set)) {
poptPrintUsage(optctx, stderr, 0);
exit(1);
@@ -752,6 +769,10 @@ opt_verify(poptContext optctx,
{
bool verified = true;
+ if (opt.no_alloc && !opt.nodelist) {
+ error("must specify a node list with -Z, --no-allocate.");
+ verified = false;
+ }
if (mode == MODE_ATTACH) { /* attach to a running job */
if (nodes_set || cpus_set || procs_set) {
@@ -771,7 +792,7 @@ opt_verify(poptContext optctx,
/* XXX what other args are incompatible with attach mode? */
- } else {
+ } else { /* mode != MODE_ATTACH */
if (mode == MODE_ALLOCATE) {
@@ -831,6 +852,65 @@ opt_verify(poptContext optctx,
return verified;
}
+static List
+create_path_list(void)
+{
+ List l = list_create(&free);
+ char *path = strdup(getenv("PATH"));
+ char *c, *lc;
+
+ if (!path) {
+ error("Error in PATH environment variable");
+ list_destroy(l);
+ return NULL;
+ }
+
+ c = lc = path;
+
+ while (*c != '\0') {
+ if (*c == ':') {
+ /* nullify and push token onto list */
+ *c = '\0';
+ if (lc != NULL && strlen(lc) > 0)
+ list_append(l, strdup(lc));
+ lc = ++c;
+ } else
+ c++;
+ }
+
+ if (strlen(lc) > 0)
+ list_append(l, strdup(lc));
+
+ free(path);
+
+ return l;
+}
+
+static char *
+search_path(char *cmd)
+{
+ List l = create_path_list();
+ ListIterator i = list_iterator_create(l);
+ char *path, *fullpath;
+ struct stat stat_buf;
+
+ fullpath = xmalloc(1);
+
+ while ((path = list_next(i))) {
+ xstrcat(fullpath, path);
+ xstrcatchar(fullpath, '/');
+ xstrcat(fullpath, cmd);
+
+ if ( (stat(fullpath, &stat_buf) == 0)
+ && (stat_buf.st_mode & (S_IXUSR | S_IXGRP | S_IXOTH))) {
+ list_destroy(l);
+ return fullpath;
+ } else
+ xfree(fullpath);
+ }
+ return NULL;
+}
+
#ifdef __DEBUG
/* generate meaningful output message based on io type and "filename" */
diff --git a/src/srun/opt.h b/src/srun/opt.h
index 9edeb7772602c324c86594c47ab2e07084cb78db..fa3e1cb3baf1d90e016658c5bdf1a5603e2b7dce 100644
--- a/src/srun/opt.h
+++ b/src/srun/opt.h
@@ -18,11 +18,7 @@
#include <getopt.h>
*/
-#ifndef HAVE_POPT_H
-# include <src/popt/popt.h>
-#else
-# include <popt.h>
-#endif
+#include <popt.h>
#include <src/common/macros.h> /* true and false */
diff --git a/src/srun/srun.c b/src/srun/srun.c
index d12424b1214bfca1aaf0cd616138799da6d67379..76037529e85d9768d7c9b5a6efc0e0057269c7ac 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -178,9 +178,16 @@ main(int ac, char **av)
if (!opt.no_alloc)
slurm_complete_job(job->jobid);
+ /* kill launch thread */
pthread_kill(job->lid, SIGTERM);
+
+ /* kill msg server thread */
pthread_kill(job->jtid, SIGTERM);
+
+ /* kill signal thread */
pthread_kill(job->sigid, SIGTERM);
+
+ /* flush stdio and kill io thread */
fflush(stderr);
fflush(stdout);
pthread_kill(job->ioid, SIGTERM);
@@ -359,7 +366,7 @@ sig_thr(void *arg)
pthread_exit(0);
}
- void
+void
fwd_signal(job_t *job, int signo)
{
int i;
@@ -377,6 +384,9 @@ fwd_signal(job_t *job, int signo)
msg.signal = (uint32_t) signo;
for (i = 0; i < job->nhosts; i++) {
+ if (job->host_state[i] != SRUN_HOST_REPLIED)
+ continue;
+
slurm_set_addr_uint(&req.address, slurm_get_slurmd_port(),
ntohl(job->iaddr[i]));
debug("sending kill req to %s", job->host[i]);