diff --git a/META b/META
index af3fb46bef458f42f16d09aacdaf00c06ac98889..7db18821c5f3dd0a14fe4a0496e7f73821008c5e 100644
--- a/META
+++ b/META
@@ -11,7 +11,7 @@
Minor: 6
Micro: 0
Version: 0.6.0
- Release: 0.pre6
+ Release: 0.pre8
API_CURRENT: 7
API_AGE: 4
API_REVISION: 0
diff --git a/NEWS b/NEWS
index a2c53a069f5bf9cf7449ee5c3f43b4067ec39051..c81cbf648c78e50d5789d59b6bd6ab2b656a9429 100644
--- a/NEWS
+++ b/NEWS
@@ -10,12 +10,32 @@ documents those changes that are of interest to users and admins.
-- made change to srun to start message thread before other threads
to make sure localtime doesn't interfere.
-* Changes in SLURM 0.6.0
-========================
+* Changes in SLURM 0.6.0-pre8
+=============================
+ -- Remove debugging xassert in switch/federation that were accidentally
+ committed
+ -- Make slurmd step manager retry slurm_container_destroy() indefinitely
+ instead of giving up after 30 seconds. If something prevents a job
+ step's processes from being killed, the job will be stuck in the
+ completing until the container destroy succeeds.
+
+* Changes in SLURM 0.6.0-pre7
+=============================
+ -- Disable localtime_r() calls from forked processes (semaphore set
+ in another pthread can deadlock calls to localtime_r made from
+ the forked process, this will be properly fixed in the next
+ major release of SLURM).
-- Added SLURM_LOCALID environment variable for spawned tasks
(Dan Palermo, HP).
-- Modify switch logic to restore state based exclusively upon
recovered job steps (not state save file).
+ -- Gracefully refuse job if there are too many job steps in slurmd.
+ -- Fix race condition in job completion that can leave nodes in
+ COMPLETING state after job is COMPLETED.
+ -- Added frees for BGL BrigeAPI strdups that were to this point unknown.
+ -- smap scrolls correctly for BGL systems.
+ -- slurm_pid2jobid() API call will now return the jobid for a step
+ manager slurmd process.
* Changes in SLURM 0.6.0-pre6
=============================
diff --git a/doc/man/man1/scontrol.1 b/doc/man/man1/scontrol.1
index 9f31857435705a81c61da5246c53371098a6be61..264acb9b0cdaecffe1921f0f79732373e70bcb0b 100644
--- a/doc/man/man1/scontrol.1
+++ b/doc/man/man1/scontrol.1
@@ -98,7 +98,8 @@ Print information one line per record.
.TP
\fIpidinfo\fP \fIPROC_ID\fP
Print the Slurm job id and scheduled termination time corresponding to the
-supplied process id, \fIPROC_ID\fP, on the current node.
+supplied process id, \fIPROC_ID\fP, on the current node. This will only
+work for processes which Slurm spawns and their descendants.
.TP
\fIping\fP
Ping the primary and secondary slurmctld daemon and report if
diff --git a/doc/man/man1/smap.1 b/doc/man/man1/smap.1
index 4f0d8ce8c65c24cfb88f38e67342f4250641bee3..e4a16c4c1f4f1dfd4b3c56bde8f9845c7b7b34d8 100644
--- a/doc/man/man1/smap.1
+++ b/doc/man/man1/smap.1
@@ -1,4 +1,4 @@
-.TH SMAP "1" "March 2005" "smap 0.6" "Slurm components"
+.TH SMAP "1" "September 2005" "smap 0.6" "Slurm components"
.SH "NAME"
smap \- graphically view information about SLURM jobs, partitions, and set
@@ -237,18 +237,6 @@ The default value is FALSE.
.I \fBRotate\fR
Equivalent to "Rotation=true".
.TP
-.I \fBMode = COPROC | VIRTUAL\fR
-Specify how the second processor on a c-node should be used.
-To use it for computation, specify VIRTUAL.
-To use it for communications, specify COPROC.
-The default value is COPROC.
-.TP
-.I \fBCoproc\fR
-Equivalent to "Mode=Coproc".
-.TP
-.I \fBVirtual\fR
-Equivalent to "Mode=Virtual".
-.TP
.I \fBElongation = TRUE | FALSE\fR
If TRUE, permit the geometry specified in the size parameter to be altered as
needed to fit available resources.
diff --git a/doc/man/man3/slurm_free_job_info_msg.3 b/doc/man/man3/slurm_free_job_info_msg.3
index 5418ac166aeb8d160eaeeaaf6df69f9c4e48de7e..1f8fd1363236bb98ea212c18c4651455dbac7e29 100644
--- a/doc/man/man3/slurm_free_job_info_msg.3
+++ b/doc/man/man3/slurm_free_job_info_msg.3
@@ -154,7 +154,8 @@ time limit.
record count, and array of job_table records for all jobs.
.LP
\fBslurm_pid2jobid\fR Returns a Slurm job id corresponding to the supplied
-local process id.
+local process id. This only works for processes which Slurm spawns and their
+descendants.
.LP
\fBslurm_print_job_info\fR Prints the contents of the data structure
describing a single job records from the data loaded by the
diff --git a/src/common/macros.h b/src/common/macros.h
index 988272604cde4935d0fbcb8a5d70cc87a94be4aa..c6d6aa1a99491bd9de403c5e2b8c1955f7858f8a 100644
--- a/src/common/macros.h
+++ b/src/common/macros.h
@@ -236,4 +236,15 @@ typedef enum {false, true} bool;
# endif
#endif
+/* localtime() can't be used after a fork due to possibly set semaphore.
+ * until we modify slurmd and srun to exec immediately after the fork,
+ * we disable localtime(). This is a temporary patch. */
+#define DISABLE_LOCALTIME 1
+#include "src/common/xstring.h"
+#include "src/common/slurm_cred.h"
+#define disable_localtime() \
+ _STMT_START { \
+ _xstrftimecat(NULL, NULL); \
+ timestr(NULL,NULL,0); \
+ } _STMT_END
#endif /* !_MACROS_H */
diff --git a/src/common/parse_time.c b/src/common/parse_time.c
index 3634109e845d7b06fbd28d55c784a80b7f77a39e..eef7bd343845fc70471f268ca4f424775a98bd41 100644
--- a/src/common/parse_time.c
+++ b/src/common/parse_time.c
@@ -210,6 +210,7 @@ static int _get_date(char *time_str, int *pos, int *month, int *mday, int *year)
* now + count [minutes | hours | days | weeks]
*
* Invalid input results in message to stderr and return value of zero
+ * NOTE: not thread safe
*/
extern time_t parse_time(char *time_str)
{
diff --git a/src/common/slurm_cred.c b/src/common/slurm_cred.c
index 9cca8a8246fe04d1f5db627ce99c7917d90d3b6c..92071daced55c41976f4152c753b731be00fb9d8 100644
--- a/src/common/slurm_cred.c
+++ b/src/common/slurm_cred.c
@@ -185,8 +185,9 @@ static void _cred_state_pack(slurm_cred_ctx_t ctx, Buf buffer);
static void _job_state_pack_one(job_state_t *j, Buf buffer);
static void _cred_state_pack_one(cred_state_t *s, Buf buffer);
+#ifndef DISABLE_LOCALTIME
static char * timestr (const time_t *tp, char *buf, size_t n);
-
+#endif
slurm_cred_ctx_t
slurm_cred_creator_ctx_create(const char *path)
@@ -1103,10 +1104,19 @@ _credential_replayed(slurm_cred_ctx_t ctx, slurm_cred_t cred)
return false;
}
+#ifdef DISABLE_LOCALTIME
+extern char * timestr (const time_t *tp, char *buf, size_t n)
+#else
static char * timestr (const time_t *tp, char *buf, size_t n)
+#endif
{
char fmt[] = "%y%m%d%H%M%S";
struct tm tmval;
+#ifdef DISABLE_LOCALTIME
+ static int disabled = 0;
+ if (buf == NULL) disabled=1;
+ if (disabled) return NULL;
+#endif
if (!localtime_r (tp, &tmval))
error ("localtime: %m");
strftime (buf, n, fmt, &tmval);
diff --git a/src/common/slurm_cred.h b/src/common/slurm_cred.h
index fdfc913d0985c798dd0d6c28935d8295347a145f..65b96c21d19317a332f08188ba55262ddcf74b23 100644
--- a/src/common/slurm_cred.h
+++ b/src/common/slurm_cred.h
@@ -1,6 +1,6 @@
/*****************************************************************************\
- * src/common/slurm_cred.h - SLURM job credential operations
- * $Id$
+ * src/common/slurm_cred.h - SLURM job credential operations
+ * $Id$
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -226,5 +226,7 @@ int slurm_cred_get_signature(slurm_cred_t cred, char **datap, int *len);
*/
void slurm_cred_print(slurm_cred_t cred);
-
+#ifdef DISABLE_LOCALTIME
+extern char * timestr (const time_t *tp, char *buf, size_t n);
+#endif
#endif /* _HAVE_SLURM_CREDS_H */
diff --git a/src/common/xstring.c b/src/common/xstring.c
index e480459ff06cd68c189d4e673aeb52713b6df240..a7d1f20b4f216ccfaa31354307270e20cf07ba3b 100644
--- a/src/common/xstring.c
+++ b/src/common/xstring.c
@@ -149,10 +149,14 @@ void _xstrftimecat(char **buf, const char *fmt)
{
char p[256]; /* output truncated to 256 chars */
time_t t;
- struct tm *tm_ptr = NULL;
- static pthread_mutex_t localtime_lock = PTHREAD_MUTEX_INITIALIZER;
+ struct tm tm;
const char default_fmt[] = "%m/%d/%Y %H:%M:%S %Z";
+#ifdef DISABLE_LOCALTIME
+ static int disabled=0;
+ if (!buf) disabled=1;
+ if (disabled) return;
+#endif
if (fmt == NULL)
fmt = default_fmt;
@@ -160,13 +164,11 @@ void _xstrftimecat(char **buf, const char *fmt)
if (time(&t) == (time_t) -1)
fprintf(stderr, "time() failed\n");
- pthread_mutex_lock(&localtime_lock);
- if (!(tm_ptr = localtime(&t)))
- fprintf(stderr, "localtime() failed\n");
+ if (!localtime_r(&t, &tm))
+ fprintf(stderr, "localtime_r() failed\n");
- strftime(p, sizeof(p), fmt, tm_ptr);
+ strftime(p, sizeof(p), fmt, &tm);
- pthread_mutex_unlock(&localtime_lock);
_xstrcat(buf, p);
}
diff --git a/src/partition_allocator/partition_allocator.c b/src/partition_allocator/partition_allocator.c
index 38c5a524a7372722e48d9a56ae96a2504d0e3c69..766ca7818c32cf06c8dbd2eb20893666edf7e313 100644
--- a/src/partition_allocator/partition_allocator.c
+++ b/src/partition_allocator/partition_allocator.c
@@ -520,7 +520,6 @@ extern void print_pa_request(pa_request_t* pa_request)
debug(" rotate:\t%d", pa_request->rotate);
debug(" elongate:\t%d", pa_request->elongate);
debug("force contig:\t%d", pa_request->force_contig);
- debug(" node_use:\t%d", pa_request->node_use);
}
/**
@@ -1810,7 +1809,12 @@ extern int set_bp_map(void)
error("rm_get_data(RM_BPID): %d", rc);
continue;
}
-
+
+ if(!bp_id) {
+ error("No BP ID was returned from database");
+ continue;
+ }
+
if ((rc = rm_get_data(my_bp, RM_BPLoc, &bp_loc))
!= STATUS_OK) {
xfree(bp_map);
@@ -1833,11 +1837,11 @@ extern int set_bp_map(void)
list_push(bp_map_list, bp_map);
+ free(bp_id);
}
if ((rc = rm_free_BGL(bgl)) != STATUS_OK)
- error("rm_free_BGL(): %s", rc);
-
+ error("rm_free_BGL(): %s", rc);
#endif
_bp_map_initialized = true;
@@ -2146,6 +2150,12 @@ static int _set_external_wires(int dim, int count, pa_node_t* source,
error("rm_get_data(RM_FirstWire): %d", rc);
break;
}
+
+ if(!wire_id) {
+ error("No Wire ID was returned from database");
+ continue;
+ }
+
if(wire_id[7] != '_')
continue;
switch(wire_id[0]) {
@@ -2165,6 +2175,9 @@ static int _set_external_wires(int dim, int count, pa_node_t* source,
}
strncpy(from_node, wire_id+2, 4);
strncpy(to_node, wire_id+8, 4);
+
+ free(wire_id);
+
from_node[4] = '\0';
to_node[4] = '\0';
if ((rc = rm_get_data(my_wire, RM_WireFromPort, &my_port))
@@ -2177,7 +2190,6 @@ static int _set_external_wires(int dim, int count, pa_node_t* source,
error("rm_get_data(RM_PortID): %d", rc);
break;
}
-
if ((rc = rm_get_data(my_wire, RM_WireToPort, &my_port))
!= STATUS_OK) {
error("rm_get_data(RM_WireToPort): %d", rc);
@@ -2188,6 +2200,7 @@ static int _set_external_wires(int dim, int count, pa_node_t* source,
error("rm_get_data(RM_PortID): %d", rc);
break;
}
+
coord = find_bp_loc(from_node);
if(coord[X]>=DIM_SIZE[X]
|| coord[Y]>=DIM_SIZE[Y]
diff --git a/src/partition_allocator/partition_allocator.h b/src/partition_allocator/partition_allocator.h
index cfa7d75111a4eac3da2e6586498233bdf47f1b97..ac94e2ec151881db7334d84f9deb6430a394236d 100644
--- a/src/partition_allocator/partition_allocator.h
+++ b/src/partition_allocator/partition_allocator.h
@@ -121,7 +121,6 @@ typedef struct {
int conn_type;
int rotate_count;
int elongate_count;
- int node_use;
bool rotate;
bool elongate;
bool force_contig;
diff --git a/src/plugins/select/bluegene/bgl_job_run.c b/src/plugins/select/bluegene/bgl_job_run.c
index 7f7572ddb2bf42948710a5091714e9ef8d37e4e3..22fc9d25828788fe500844e61f4d34cc75bf610d 100644
--- a/src/plugins/select/bluegene/bgl_job_run.c
+++ b/src/plugins/select/bluegene/bgl_job_run.c
@@ -396,10 +396,22 @@ static void _term_agent(bgl_update_t *bgl_update_ptr)
part_id, bgl_err_str(rc));
continue;
}
+
+ if(!part_id) {
+ error("No partitionID returned from Database");
+ continue;
+ }
+
debug2("looking at partition %s looking for %s\n",
part_id, bgl_update_ptr->bgl_part_id);
- if (strcmp(part_id, bgl_update_ptr->bgl_part_id) != 0)
+
+ if (strcmp(part_id, bgl_update_ptr->bgl_part_id) != 0) {
+ free(part_id);
continue;
+ }
+
+ free(part_id);
+
if ((rc = rm_get_data(job_elem, RM_JobDBJobID, &job_id))
!= STATUS_OK) {
error("rm_get_data(RM_JobDBJobID): %s",
diff --git a/src/plugins/select/bluegene/bgl_part_info.c b/src/plugins/select/bluegene/bgl_part_info.c
index 33781d606160c5c38eb0f7f87639cab6aa60a983..ec481ffe47fa1765841beae1671224488cf9b2b9 100644
--- a/src/plugins/select/bluegene/bgl_part_info.c
+++ b/src/plugins/select/bluegene/bgl_part_info.c
@@ -223,9 +223,14 @@ extern int update_partition_list()
updated = -1;
break;
}
- if(strncmp("RMP", name,3))
+ if(!name) {
+ error("No Partition ID was returned from database");
continue;
-
+ }
+ if(strncmp("RMP", name, 3)) {
+ free(name);
+ continue;
+ }
bgl_record = find_bgl_record(name);
if(bgl_record == NULL) {
@@ -235,7 +240,9 @@ extern int update_partition_list()
if ((rc = pm_destroy_partition(name))
!= STATUS_OK) {
if(rc == PARTITION_NOT_FOUND) {
- debug("partition %s is not found");
+ debug("partition %s is not found",
+ name);
+ free(name);
break;
}
error("pm_destroy_partition(%s): %s",
@@ -249,9 +256,11 @@ extern int update_partition_list()
bgl_err_str(rc));
} else
debug("done\n");
+ free(name);
continue;
}
-
+ free(name);
+
slurm_mutex_lock(&part_state_mutex);
if ((rc = rm_get_data(part_ptr, RM_PartitionMode, &node_use))
diff --git a/src/plugins/select/bluegene/bgl_switch_connections.c b/src/plugins/select/bluegene/bgl_switch_connections.c
index 74379b5531cb5abf725c8b85bbf9cb01a8f69d2d..c0c49c98d038d3c8e72c39133ebcd5567277bccc 100644
--- a/src/plugins/select/bluegene/bgl_switch_connections.c
+++ b/src/plugins/select/bluegene/bgl_switch_connections.c
@@ -1,6 +1,8 @@
/*****************************************************************************\
* bgl_switch_connections.c - Blue Gene switch management functions,
* establish switch connections
+ *
+ * $Id$
*****************************************************************************
* Copyright (C) 2004 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -454,12 +456,18 @@ extern int configure_partition_switches(bgl_record_t * bgl_record)
rc = SLURM_ERROR;
goto cleanup;
}
-
+
+ if(!bpid) {
+ error("No BP ID was returned from database");
+ continue;
+ }
+
found_bpid = 0;
for (i=0; i<switch_count; i++) {
if(i) {
if ((rc = rm_get_data(bgl, RM_NextSwitch,
- &curr_switch)) != STATUS_OK) {
+ &curr_switch))
+ != STATUS_OK) {
fatal("rm_get_data: RM_NextSwitch: %s",
bgl_err_str(rc));
list_iterator_destroy(bgl_itr);
@@ -468,7 +476,8 @@ extern int configure_partition_switches(bgl_record_t * bgl_record)
}
} else {
if ((rc = rm_get_data(bgl, RM_FirstSwitch,
- &curr_switch)) != STATUS_OK) {
+ &curr_switch))
+ != STATUS_OK) {
fatal("rm_get_data: "
"RM_FirstSwitch: %s",
bgl_err_str(rc));
@@ -485,15 +494,25 @@ extern int configure_partition_switches(bgl_record_t * bgl_record)
rc = SLURM_ERROR;
goto cleanup;
}
-
+
+ if(!curr_bpid) {
+ error("No BP ID was returned from database");
+ continue;
+ }
+
if (!strcasecmp((char *)bpid, (char *)curr_bpid)) {
coord_switch[found_bpid] = curr_switch;
found_bpid++;
- if(found_bpid==PA_SYSTEM_DIMENSIONS)
+ if(found_bpid==PA_SYSTEM_DIMENSIONS) {
+ free(curr_bpid);
break;
- }
+ }
+ }
+ free(curr_bpid);
}
-
+
+ free(bpid);
+
if(found_bpid==PA_SYSTEM_DIMENSIONS) {
debug2("adding midplane %d%d%d\n",
@@ -503,17 +522,17 @@ extern int configure_partition_switches(bgl_record_t * bgl_record)
switch_itr = list_iterator_create(bgl_bp->switch_list);
while((bgl_switch = list_next(switch_itr)) != NULL) {
- if ((rc = rm_get_data(coord_switch
- [bgl_switch->dim],
- RM_SwitchID,&name2))
- != STATUS_OK) {
- fatal("rm_get_data: RM_SwitchID: %s",
- bgl_err_str(rc));
- list_iterator_destroy(bgl_itr);
- list_iterator_destroy(switch_itr);
- rc = SLURM_ERROR;
- goto cleanup;
- }
+ /* if ((rc = rm_get_data(coord_switch */
+/* [bgl_switch->dim], */
+/* RM_SwitchID,&name2)) */
+/* != STATUS_OK) { */
+/* fatal("rm_get_data: RM_SwitchID: %s",*/
+/* bgl_err_str(rc)); */
+/* list_iterator_destroy(bgl_itr); */
+/* list_iterator_destroy(switch_itr); */
+/* rc = SLURM_ERROR; */
+/* goto cleanup; */
+/* } */
debug2("adding switch dim %d\n",
bgl_switch->dim);
diff --git a/src/plugins/select/bluegene/bluegene.c b/src/plugins/select/bluegene/bluegene.c
index cfc245cc6317f39603f19cb5ddf093ef3af440dd..705e7e73c401c07857e9aa0a306a484de40378a7 100644
--- a/src/plugins/select/bluegene/bluegene.c
+++ b/src/plugins/select/bluegene/bluegene.c
@@ -110,6 +110,8 @@ extern int init_bgl(void)
/* Purge all plugin variables */
extern void fini_bgl(void)
{
+ int rc;
+
_set_bgl_lists();
if (bgl_list) {
@@ -135,7 +137,8 @@ extern void fini_bgl(void)
#ifdef HAVE_BGL_FILES
if(bgl)
- rm_free_BGL(bgl);
+ if ((rc = rm_free_BGL(bgl)) != STATUS_OK)
+ error("rm_free_BGL(): %s", bgl_err_str(rc));
#endif
pa_fini();
}
@@ -318,20 +321,26 @@ extern int remove_all_users(char *bgl_part_id, char *user_name)
RM_PartitionFirstUser,
&user))
!= STATUS_OK) {
- error("rm_get_partition(%s): %s",
+ error("rm_get_data(%s): %s",
bgl_part_id,
bgl_err_str(rc));
returnc = REMOVE_USER_ERR;
break;
}
}
-
- if(!strcmp(user, slurmctld_conf.slurm_user_name))
+ if(!user) {
+ error("No user was returned from database");
continue;
-
+ }
+ if(!strcmp(user, slurmctld_conf.slurm_user_name)) {
+ free(user);
+ continue;
+ }
+
if(user_name) {
if(!strcmp(user, user_name)) {
returnc = REMOVE_USER_FOUND;
+ free(user);
continue;
}
}
@@ -344,7 +353,8 @@ extern int remove_all_users(char *bgl_part_id, char *user_name)
debug("user %s isn't on partition %s",
user,
bgl_part_id);
- }
+ }
+ free(user);
}
if ((rc = rm_free_partition(part_ptr)) != STATUS_OK) {
error("rm_free_partition(): %s", bgl_err_str(rc));
@@ -759,9 +769,10 @@ extern int bgl_free_partition(bgl_record_t *bgl_record)
debug("partition %s is not found");
break;
}
- error("pm_destroy_partition(%s): %s",
+ error("pm_destroy_partition(%s): %s "
+ "State = %d",
bgl_record->bgl_part_id,
- bgl_err_str(rc));
+ bgl_err_str(rc), bgl_record->state);
}
}
@@ -1615,7 +1626,6 @@ static int _update_bgl_record_state(List bgl_destroy_list)
rm_partition_t *part_ptr = NULL;
ListIterator itr;
bgl_record_t* bgl_record = NULL;
- int found = 0;
if(!bgl_destroy_list) {
return SLURM_SUCCESS;
@@ -1665,18 +1675,19 @@ static int _update_bgl_record_state(List bgl_destroy_list)
func_rc = SLURM_ERROR;
break;
}
- found = 0;
+ if (!name) {
+ error("RM_Partition is NULL");
+ continue;
+ }
+
itr = list_iterator_create(bgl_destroy_list);
while ((bgl_record = (bgl_record_t*) list_next(itr))) {
- if(bgl_record->bgl_part_id)
- if(!strcmp(bgl_record->bgl_part_id, name)) {
- found = 1;
- break;
- }
- }
- list_iterator_destroy(itr);
+ if(!bgl_record->bgl_part_id)
+ continue;
+ if(strcmp(bgl_record->bgl_part_id, name)) {
+ continue;
+ }
- if(found) {
slurm_mutex_lock(&part_state_mutex);
if ((rc = rm_get_data(part_ptr,
RM_PartitionState,
@@ -1690,11 +1701,13 @@ static int _update_bgl_record_state(List bgl_destroy_list)
name, bgl_record->state, state);
bgl_record->state = state;
}
- slurm_mutex_unlock(&part_state_mutex);
+ slurm_mutex_unlock(&part_state_mutex);
+ break;
}
+ list_iterator_destroy(itr);
+ free(name);
}
-clean_up:
if ((rc = rm_free_partition_list(part_list)) != STATUS_OK) {
error("rm_free_partition_list(): %s", bgl_err_str(rc));
}
diff --git a/src/plugins/select/bluegene/partition_sys.c b/src/plugins/select/bluegene/partition_sys.c
index 6aa153538ca62cfff55a8d94b6c624177e43f307..d92826568d2b1827f030205d94520370b7d6c334 100755
--- a/src/plugins/select/bluegene/partition_sys.c
+++ b/src/plugins/select/bluegene/partition_sys.c
@@ -167,8 +167,14 @@ static int _post_allocate(bgl_record_t *bgl_record)
error("rm_get_data(RM_PartitionID): %s", bgl_err_str(rc));
bgl_record->bgl_part_id = xstrdup("UNKNOWN");
} else {
+ if(!part_id) {
+ error("No Partition ID was returned from database");
+ return SLURM_ERROR;
+ }
bgl_record->bgl_part_id = xstrdup(part_id);
+ free(part_id);
+
xfree(bgl_record->target_name);
bgl_record->target_name =
xstrdup(slurmctld_conf.slurm_user_name);
@@ -268,42 +274,53 @@ int read_bgl_partitions()
bgl_err_str(rc));
continue;
}
- if(strncmp("RMP",part_name,3))
+
+ if(!part_name) {
+ error("No Partition ID was returned from database");
continue;
-
+ }
+
+ if(strncmp("RMP", part_name, 3)) {
+ free(part_name);
+ continue;
+ }
if(bgl_recover)
if ((rc = rm_get_partition(part_name, &part_ptr))
!= STATUS_OK) {
error("Partition %s doesn't exist.",
part_name);
rc = SLURM_ERROR;
+ free(part_name);
break;
}
-
/* New BGL partition record */
bgl_record = xmalloc(sizeof(bgl_record_t));
list_push(bgl_curr_part_list, bgl_record);
bgl_record->bgl_part_id = xstrdup(part_name);
+
+ free(part_name);
+
bgl_record->state = -1;
if ((rc = rm_get_data(part_ptr, RM_PartitionBPNum, &bp_cnt))
- != STATUS_OK) {
+ != STATUS_OK) {
error("rm_get_data(RM_BPNum): %s", bgl_err_str(rc));
bp_cnt = 0;
}
if(bp_cnt==0)
- continue;
-
+ goto clean_up;
+
bgl_record->bgl_part_list = list_create(NULL);
bgl_record->hostlist = hostlist_create(NULL);
for (i=0; i<bp_cnt; i++) {
if(i) {
if ((rc = rm_get_data(part_ptr,
- RM_PartitionNextBP, &bp_ptr))
- != STATUS_OK) {
+ RM_PartitionNextBP,
+ &bp_ptr))
+ != STATUS_OK) {
error("rm_get_data(RM_NextBP): %s",
bgl_err_str(rc));
rc = SLURM_ERROR;
@@ -317,6 +334,8 @@ int read_bgl_partitions()
error("rm_get_data(RM_FirstBP): %s",
bgl_err_str(rc));
rc = SLURM_ERROR;
+ if (bgl_recover)
+ rm_free_partition(part_ptr);
return rc;
}
}
@@ -327,9 +346,16 @@ int read_bgl_partitions()
rc = SLURM_ERROR;
break;
}
-
+
+ if(!bpid) {
+ error("No BP ID was returned from database");
+ continue;
+ }
+
coord = find_bp_loc(bpid);
-
+
+ free(bpid);
+
if(!coord)
fatal("No contact with db2. Shutting down.");
@@ -382,9 +408,19 @@ int read_bgl_partitions()
xstrdup(slurmctld_conf.
slurm_user_name);
} else {
- rm_get_data(part_ptr, RM_PartitionFirstUser,
- &user_name);
+ user_name = NULL;
+ if ((rc = rm_get_data(part_ptr, RM_PartitionFirstUser,
+ &user_name)) != STATUS_OK) {
+ error("rm_get_data(RM_PartitionFirstUser): %s",
+ bgl_err_str(rc));
+ }
+ if(!user_name) {
+ error("No user name was "
+ "returned from database");
+ goto clean_up;
+ }
bgl_record->user_name = xstrdup(user_name);
+
if(!bgl_record->boot_state)
bgl_record->target_name =
xstrdup(slurmctld_conf.
@@ -392,6 +428,8 @@ int read_bgl_partitions()
else
bgl_record->target_name =
xstrdup(user_name);
+
+ free(user_name);
}
if((pw_ent = getpwnam(bgl_record->user_name))
@@ -419,8 +457,8 @@ int read_bgl_partitions()
bgl_record->part_lifecycle = STATIC;
- if ((rc = rm_free_partition(part_ptr))
- != STATUS_OK) {
+clean_up: if (bgl_recover
+ && ((rc = rm_free_partition(part_ptr)) != STATUS_OK)) {
error("rm_free_partition(): %s", bgl_err_str(rc));
}
}
diff --git a/src/plugins/select/bluegene/sfree.c b/src/plugins/select/bluegene/sfree.c
index 9d3373983cde04494f68edc370f0a9b5969b8220..e1726086ab942047a65e7b25f49805879d463da0 100644
--- a/src/plugins/select/bluegene/sfree.c
+++ b/src/plugins/select/bluegene/sfree.c
@@ -205,11 +205,22 @@ int main(int argc, char *argv[])
break;
}
- if(strncmp("RMP", bgl_part_id, 3))
+
+ if(!bgl_part_id) {
+ error("No Part ID was returned from database");
continue;
+ }
+ if(strncmp("RMP", bgl_part_id, 3)) {
+ free(bgl_part_id);
+ continue;
+ }
+
delete_record = xmalloc(sizeof(delete_record_t));
delete_record->bgl_part_id = xstrdup(bgl_part_id);
+
+ free(bgl_part_id);
+
delete_record->state = -1;
list_push(delete_record_list, delete_record);
@@ -297,7 +308,6 @@ static int _update_bgl_record_state()
int j, rc, i, num_parts = 0;
rm_partition_state_t state = -2;
rm_partition_t *part_ptr = NULL;
- int found=0;
delete_record_t *delete_record = NULL;
ListIterator itr;
@@ -345,20 +355,21 @@ static int _update_bgl_record_state()
state = -1;
break;
}
- found = 0;
+ if(!name) {
+ error("No Partition ID was returned from database");
+ continue;
+ }
+
itr = list_iterator_create(delete_record_list);
while ((delete_record =
(delete_record_t*) list_next(itr))) {
- if(delete_record->bgl_part_id)
- if(!strcmp(delete_record->bgl_part_id, name)) {
- found = 1;
- break;
- }
- }
- list_iterator_destroy(itr);
+ if(!delete_record->bgl_part_id)
+ continue;
+ if(strcmp(delete_record->bgl_part_id, name)) {
+ continue;
+ }
- if(found) {
if(state == -1)
goto clean_up;
else if(j>=num_parts) {
@@ -377,7 +388,10 @@ static int _update_bgl_record_state()
"(RM_PartitionState): %s",
_bgl_err_str(rc));
}
+ break;
}
+ list_iterator_destroy(itr);
+ free(name);
}
clean_up:
if ((rc = rm_free_partition_list(part_list)) != STATUS_OK) {
@@ -439,9 +453,17 @@ static void _term_jobs_on_part(char *bgl_part_id)
part_id, _bgl_err_str(rc));
continue;
}
-
- if (strcmp(part_id, bgl_part_id) != 0)
+
+ if(!part_id) {
+ error("No Partition ID was returned from database");
continue;
+ }
+
+ if (strcmp(part_id, bgl_part_id) != 0) {
+ free(part_id);
+ continue;
+ }
+ free(part_id);
job_found = 1;
if ((rc = rm_get_data(job_elem, RM_JobDBJobID, &job_id))
!= STATUS_OK) {
@@ -450,9 +472,9 @@ static void _term_jobs_on_part(char *bgl_part_id)
continue;
}
info("got job_id %d",job_id);
- if((rc = _remove_job(job_id)) == INTERNAL_ERROR)
+ if((rc = _remove_job(job_id)) == INTERNAL_ERROR) {
goto not_removed;
-
+ }
}
if(job_found == 0)
info("No jobs on bglblock %s", bgl_part_id);
diff --git a/src/plugins/select/bluegene/state_test.c b/src/plugins/select/bluegene/state_test.c
index 807cc8f0000dfa210a553fc6f1bb8194517e3ae4..c49284c2202a52c5cd40836a837749dc396e7587 100644
--- a/src/plugins/select/bluegene/state_test.c
+++ b/src/plugins/select/bluegene/state_test.c
@@ -101,8 +101,18 @@ static void _configure_node_down(rm_bp_id_t bp_id, rm_BGL_t *bgl)
error("rm_get_data(RM_BPID): %s", bgl_err_str(rc));
continue;
}
- if (strcmp(bp_id, bpid) != 0) /* different base partition */
+
+ if(!bpid) {
+ error("No BPID was returned from database");
continue;
+ }
+
+ if (strcmp(bp_id, bpid) != 0) { /* different base partition */
+ free(bpid);
+ continue;
+ }
+ free(bpid);
+
if ((rc = rm_get_data(my_bp, RM_BPState, &bp_state))
!= STATUS_OK) {
error("rm_get_data(RM_BPState): %s", bgl_err_str(rc));
@@ -277,7 +287,14 @@ static void _test_down_switches(rm_BGL_t *bgl)
bgl_err_str(rc));
continue;
}
+
+ if(!bp_id) {
+ error("No BPID was returned from database");
+ continue;
+ }
+
_configure_node_down(bp_id, bgl);
+ free(bp_id);
}
}
#endif
diff --git a/src/plugins/switch/federation/federation.c b/src/plugins/switch/federation/federation.c
index 206831578cf6b864b7d1fb97e1ec282027f9c3f2..3e4e831516f210b49b0a019741e51f0071d17175 100644
--- a/src/plugins/switch/federation/federation.c
+++ b/src/plugins/switch/federation/federation.c
@@ -1522,10 +1522,6 @@ _job_step_window_state(fed_jobinfo_t *jp, hostlist_t hl, enum NTBL_RC state)
|| (hostlist_is_empty(hl)))
return SLURM_ERROR;
- xassert(jp->tables_per_task);
- xassert(jp->tableinfo);
- xassert(jp->tableinfo[0].table_length);
-
if ((jp->tables_per_task == 0)
|| !jp->tableinfo
|| (jp->tableinfo[0].table_length == 0))
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 3d74fc4f3edccfe7453e40d9c99e4e641fee98b1..8c5f7ff11a1d8bd014d553912e25b94276f9f30e 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -1576,6 +1576,9 @@ extern int job_complete(uint32_t job_id, uid_t uid, bool requeue,
return ESLURM_USER_ID_MISSING;
}
+ if (job_ptr->job_state & JOB_COMPLETING)
+ return SLURM_SUCCESS; /* avoid replay */
+
if (job_ptr->job_state == JOB_RUNNING)
job_comp_flag = JOB_COMPLETING;
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 8184e9cc41fd0e1c96c44097ed332663c22d6cd8..7e9a1e7d733931cf9e5f624f8e7a70a0e00547db 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -1356,7 +1356,7 @@ extern int validate_nodes_via_front_end(uint32_t job_count,
hostlist_uniq(reg_hostlist);
hostlist_ranged_string(reg_hostlist, sizeof(host_str),
host_str);
- debug("Nodes %s have registerd", host_str);
+ debug("Nodes %s have registered", host_str);
hostlist_destroy(reg_hostlist);
}
if (return_hostlist) {
diff --git a/src/slurmd/mgr.c b/src/slurmd/mgr.c
index 241f4260bb38eaf661b3d3ab337e7547dad18cd1..b36a561061982a88452758a1d89cde9134d1c8e2 100644
--- a/src/slurmd/mgr.c
+++ b/src/slurmd/mgr.c
@@ -866,7 +866,7 @@ _kill_running_tasks(slurmd_job_t *job)
List steps;
ListIterator i;
job_step_t *s = NULL;
- int limit = 0;
+ int delay = 1;
if (job->batch)
return;
@@ -884,12 +884,16 @@ _kill_running_tasks(slurmd_job_t *job)
if (s->cont_id) {
slurm_container_signal(s->cont_id, SIGKILL);
- /* Try destroying the container up to 30 times */
- while (slurm_container_destroy(s->cont_id) != SLURM_SUCCESS
- && limit < 30) {
+ /* Spin until the container is successfully destroyed */
+ while (slurm_container_destroy(s->cont_id) != SLURM_SUCCESS) {
slurm_container_signal(s->cont_id, SIGKILL);
- sleep(1);
- limit++;
+ sleep(delay);
+ if (delay < 120) {
+ delay *= 2;
+ } else {
+ error("Unable to destroy container, job %u.%u",
+ job->jobid, job->stepid);
+ }
}
}
diff --git a/src/slurmd/proctrack.c b/src/slurmd/proctrack.c
index d07aaafcd075b69ce78fce9ace60b38cad78521b..bc2efea7cd89373ada1f8676a4238b614830130d 100644
--- a/src/slurmd/proctrack.c
+++ b/src/slurmd/proctrack.c
@@ -188,7 +188,8 @@ slurm_proctrack_init( void )
}
if ( _proctrack_get_ops( g_proctrack_context ) == NULL ) {
- error( "cannot resolve proctrack plugin operations" );
+ error( "cannot resolve proctrack plugin operations for %s",
+ proctrack_type );
_proctrack_context_destroy( g_proctrack_context );
g_proctrack_context = NULL;
retval = SLURM_ERROR;
diff --git a/src/slurmd/req.c b/src/slurmd/req.c
index 4e5b0f4207228fd4af88c7acb6b6d36d99a2d4e8..160126112b51e584d9bf9274b437ac31130011cf 100644
--- a/src/slurmd/req.c
+++ b/src/slurmd/req.c
@@ -244,6 +244,9 @@ _fork_new_slurmd(void)
return ((int) pid);
}
+#ifdef DISABLE_LOCALTIME
+ disable_localtime();
+#endif
if (close(fds[0]) < 0)
error("Unable to close read-pipe in child: %m");
@@ -456,6 +459,14 @@ _rpc_launch_tasks(slurm_msg_t *msg, slurm_addr *cli)
goto done;
}
+ /* Make an effort to not overflow shm records */
+ if (shm_free_steps() < 2) {
+ errnum = ESLURMD_TOOMANYSTEPS;
+ error("reject task %u.%u, too many steps", req->job_id,
+ req->job_step_id);
+ goto done;
+ }
+
/* xassert(slurm_cred_jobid_cached(conf->vctx, req->job_id));*/
/* Run job prolog if necessary */
@@ -517,6 +528,14 @@ _rpc_spawn_task(slurm_msg_t *msg, slurm_addr *cli)
goto done;
}
+ /* Make an effort to not overflow shm records */
+ if (shm_free_steps() < 2) {
+ errnum = ESLURMD_TOOMANYSTEPS;
+ error("reject task %u.%u, too many steps", req->job_id,
+ req->job_step_id);
+ goto done;
+ }
+
slurmd_get_addr(cli, &port, host, sizeof(host));
info("spawn task %u.%u request from %u@%s", req->job_id,
req->job_step_id, req->uid, host);
@@ -614,6 +633,14 @@ _rpc_batch_job(slurm_msg_t *msg, slurm_addr *cli)
goto done;
}
+ /* Make an effort to not overflow shm records */
+ if (shm_free_steps() < 2) {
+ rc = ESLURMD_TOOMANYSTEPS;
+ error("reject job %u, too many steps", req->job_id);
+ _prolog_error(req, rc);
+ goto done;
+ }
+
if (req->step_id != NO_VAL && req->step_id != 0)
first_job_run = false;
@@ -875,15 +902,27 @@ static void _rpc_pid2jid(slurm_msg_t *msg, slurm_addr *cli)
job_id_response_msg_t resp;
bool found = false;
uint32_t my_cont = slurm_container_find(req->job_pid);
+ List steps = shm_get_steps();
+ ListIterator i = list_iterator_create(steps);
+ job_step_t *s = NULL;
if (my_cont == 0) {
- verbose("slurm_container_find(%u): process not found",
- (uint32_t) req->job_pid);
+ debug("slurm_container_find(%u): process not found",
+ (uint32_t) req->job_pid);
+ /*
+ * Check if the job_pid matches the pid of a job step slurmd.
+ * LCRM gets confused if a session leader process
+ * (the job step slurmd) is not labelled as a process in the
+ * job step.
+ */
+ while ((s = list_next(i))) {
+ if (s->mpid == req->job_pid) {
+ resp.job_id = s->jobid;
+ found = true;
+ break;
+ }
+ }
} else {
- List steps = shm_get_steps();
- ListIterator i = list_iterator_create(steps);
- job_step_t *s = NULL;
-
while ((s = list_next(i))) {
if (s->cont_id == my_cont) {
resp.job_id = s->jobid;
@@ -891,9 +930,9 @@ static void _rpc_pid2jid(slurm_msg_t *msg, slurm_addr *cli)
break;
}
}
- list_iterator_destroy(i);
- list_destroy(steps);
}
+ list_iterator_destroy(i);
+ list_destroy(steps);
if (found) {
resp_msg.address = msg->address;
diff --git a/src/slurmd/shm.c b/src/slurmd/shm.c
index 7d7c7c47c0d9c9f24ce3b5b8a86965dfe93d60d5..ec807bba79bed2015a75ef4b61623e50e21ecf4e 100644
--- a/src/slurmd/shm.c
+++ b/src/slurmd/shm.c
@@ -1,6 +1,6 @@
/*****************************************************************************\
- * src/slurmd/shm.c - slurmd shared memory routines
- * $Id$
+ * src/slurmd/shm.c - slurmd shared memory routines
+ * $Id$
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -269,6 +269,20 @@ shm_get_steps(void)
return l;
}
+extern int
+shm_free_steps(void)
+{
+ int i, rc = 0;
+
+ xassert(slurmd_shm != NULL);
+ _shm_lock();
+ for (i = 0; i < MAX_JOB_STEPS; i++) {
+ if (slurmd_shm->step[i].state == SLURMD_JOB_UNUSED)
+ rc++;
+ }
+ _shm_unlock();
+ return rc;
+}
static bool
_job_step_mgr_still_running(job_step_t *step)
diff --git a/src/slurmd/shm.h b/src/slurmd/shm.h
index ab7c4f581111184456485f647059ddb200325def..369f2568ffe72a5f6a45a23a10e11a9a279c2468 100644
--- a/src/slurmd/shm.h
+++ b/src/slurmd/shm.h
@@ -1,6 +1,6 @@
/*****************************************************************************\
- * src/slurmd/shm.h - shared memory routines for slurmd
- * $Id$
+ * src/slurmd/shm.h - shared memory routines for slurmd
+ * $Id$
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -263,6 +263,10 @@ int shm_update_job_timelimit(uint32_t jobid, time_t newlim);
*/
int shm_update_step_timelimit(uint32_t jobid, uint32_t stepid, time_t newlim);
+/*
+ * Return count of free shm records
+ */
+extern int shm_free_steps(void);
/*
* Return job step timelimit
diff --git a/src/slurmd/slurmd.c b/src/slurmd/slurmd.c
index b73bf5b31130f603f2fbfd99d652a94495c9907f..4633f16c43590a8e24e2603ebc49181c220c6ef7 100644
--- a/src/slurmd/slurmd.c
+++ b/src/slurmd/slurmd.c
@@ -657,7 +657,8 @@ _slurmd_init()
*/
_update_logging();
_print_conf();
- slurm_proctrack_init();
+ if (slurm_proctrack_init() != SLURM_SUCCESS)
+ return SLURM_FAILURE;
if (getrlimit(RLIMIT_NOFILE,&rlim) == 0) {
rlim.rlim_cur = rlim.rlim_max;
diff --git a/src/smap/configure_functions.c b/src/smap/configure_functions.c
index 533d6a7904121531eaad30c6c7431ad127e331cb..ee95c3fdd7c16b71d46bac013bb187db1725fba7 100644
--- a/src/smap/configure_functions.c
+++ b/src/smap/configure_functions.c
@@ -108,7 +108,6 @@ static int _create_allocation(char *com, List allocated_partitions)
request->rotate = false;
request->elongate = false;
request->force_contig = false;
- request->node_use = -1;
while(i<len) {
@@ -521,7 +520,6 @@ static int _copy_allocation(char *com, List allocated_partitions)
request->geometry[Z] = allocated_part->request->geometry[Z];
request->size = allocated_part->request->size;
request->conn_type=allocated_part->request->conn_type;
- request->node_use=allocated_part->request->node_use;
request->rotate =allocated_part->request->rotate;
request->elongate = allocated_part->request->elongate;
request->force_contig = allocated_part->request->force_contig;
@@ -615,24 +613,6 @@ static int _save_allocation(char *com, List allocated_partitions)
else
conn_type = "MESH";
- if(allocated_part->request->node_use != -1) {
- if(allocated_part->request->node_use
- == COPROCESSOR)
- mode_type = "COPROCESSOR";
- else
- mode_type = "VIRTUAL";
-
- sprintf(save_string, "Nodes=%s Type=%s "
- "Use=%s\n",
- allocated_part->request->save_name,
- conn_type, mode_type);
- } else {
- sprintf(save_string, "Nodes=%s "
- "Type=%s\n",
- allocated_part->request->save_name,
- conn_type);
- }
-
fputs (save_string,file_ptr);
}
fclose (file_ptr);
@@ -646,15 +626,9 @@ static void _print_header_command(void)
mvwprintw(pa_system_ptr->text_win, pa_system_ptr->ycord,
pa_system_ptr->xcord, "ID");
pa_system_ptr->xcord += 4;
- /* mvwprintw(pa_system_ptr->text_win, pa_system_ptr->ycord, */
-/* pa_system_ptr->xcord, "PARTITION"); */
-/* pa_system_ptr->xcord += 10; */
mvwprintw(pa_system_ptr->text_win, pa_system_ptr->ycord,
pa_system_ptr->xcord, "TYPE");
pa_system_ptr->xcord += 7;
- mvwprintw(pa_system_ptr->text_win, pa_system_ptr->ycord,
- pa_system_ptr->xcord, "CONF");
- pa_system_ptr->xcord += 9;
mvwprintw(pa_system_ptr->text_win, pa_system_ptr->ycord,
pa_system_ptr->xcord, "CONTIG");
pa_system_ptr->xcord += 7;
@@ -681,9 +655,6 @@ static void _print_text_command(allocated_part_t *allocated_part)
mvwprintw(pa_system_ptr->text_win, pa_system_ptr->ycord,
pa_system_ptr->xcord, "%c",allocated_part->letter);
pa_system_ptr->xcord += 4;
- /* mvwprintw(pa_system_ptr->text_win, pa_system_ptr->ycord, */
- /* pa_system_ptr->xcord, "PARTITION"); */
- /* pa_system_ptr->xcord += 10; */
if(allocated_part->request->conn_type==TORUS)
mvwprintw(pa_system_ptr->text_win, pa_system_ptr->ycord,
pa_system_ptr->xcord, "TORUS");
@@ -692,22 +663,6 @@ static void _print_text_command(allocated_part_t *allocated_part)
pa_system_ptr->xcord, "MESH");
pa_system_ptr->xcord += 7;
- if(allocated_part->request->node_use != -1) {
- if(allocated_part->request->node_use == COPROCESSOR)
- mvwprintw(pa_system_ptr->text_win,
- pa_system_ptr->ycord,
- pa_system_ptr->xcord, "coproc");
- else
- mvwprintw(pa_system_ptr->text_win,
- pa_system_ptr->ycord,
- pa_system_ptr->xcord, "virtual");
- } else {
- mvwprintw(pa_system_ptr->text_win, pa_system_ptr->ycord,
- pa_system_ptr->xcord, "both");
- }
-
- pa_system_ptr->xcord += 9;
-
if(allocated_part->request->force_contig)
mvwprintw(pa_system_ptr->text_win, pa_system_ptr->ycord,
pa_system_ptr->xcord, "Y");
@@ -748,12 +703,14 @@ static void _print_text_command(allocated_part_t *allocated_part)
void get_command(void)
{
char com[255];
- //static node_info_msg_t *node_info_ptr;
+
int text_width, text_startx;
allocated_part_t *allocated_part = NULL;
int i=0;
+ int count=0;
+
WINDOW *command_win;
- List allocated_partitions;
+ List allocated_partitions;
ListIterator results_i;
if(params.commandline) {
@@ -796,22 +753,29 @@ void get_command(void)
pa_system_ptr->xcord++;
}
pa_system_ptr->ycord++;
- pa_system_ptr->xcord=1;
- memset(error_string,0,255);
-
+ pa_system_ptr->xcord=1;
+ memset(error_string,0,255);
}
results_i = list_iterator_create(allocated_partitions);
+
+ count = list_count(allocated_partitions)
+ - (LINES-(pa_system_ptr->ycord+5));
+
+ if(count<0)
+ count=0;
+ i=0;
while((allocated_part = list_next(results_i)) != NULL) {
- _print_text_command(allocated_part);
+ if(i>=count)
+ _print_text_command(allocated_part);
+ i++;
}
- list_iterator_destroy(results_i);
-
+ list_iterator_destroy(results_i);
wnoutrefresh(pa_system_ptr->text_win);
wnoutrefresh(pa_system_ptr->grid_win);
doupdate();
clear_window(command_win);
- //wclear(command_win);
+
box(command_win, 0, 0);
mvwprintw(command_win, 0, 3,
"Input Command: (type quit to change view, "
@@ -824,6 +788,8 @@ void get_command(void)
_delete_allocated_parts(allocated_partitions);
pa_fini();
exit(0);
+ } if (!strcmp(com, "quit")) {
+ break;
} else if (!strncasecmp(com, "resolve", 7) ||
!strncasecmp(com, "r ", 2)) {
_resolve(com);
diff --git a/src/smap/partition_functions.c b/src/smap/partition_functions.c
index 3dbd3f06303fd7e8f38decef844b3baf7707687b..35f4a2a70f471b10a8517d8a12876ade0574331b 100644
--- a/src/smap/partition_functions.c
+++ b/src/smap/partition_functions.c
@@ -69,7 +69,7 @@ static int _list_match_all(void *object, void *key);
static int _in_slurm_partition(db2_block_info_t *db2_info_ptr,
int *first,
int *last);
-static int _print_rest(db2_block_info_t *block_ptr, int *count);
+static int _print_rest(db2_block_info_t *block_ptr);
#endif
extern void get_slurm_part()
@@ -316,12 +316,12 @@ extern void get_bgl_part()
list_next(itr)) != NULL) {
if (params.commandline)
block_ptr->printed = 1;
- else
+ else {
if(count>=text_line_cnt)
block_ptr->printed = 1;
-
- _print_rest(block_ptr, &count);
-
+ }
+ _print_rest(block_ptr);
+ count++;
}
list_iterator_destroy(itr);
@@ -798,63 +798,13 @@ static int _in_slurm_partition(db2_block_info_t *db2_info_ptr,
}
-static int _print_rest(db2_block_info_t *block_ptr, int *count)
+static int _print_rest(db2_block_info_t *block_ptr)
{
partition_info_t part;
db2_block_info_t *db2_info_ptr = NULL;
ListIterator itr;
int set = 0;
-/* part.total_nodes = 0; */
-
-/* if (block_list) { */
-/* itr = list_iterator_create(block_list); */
-/* while ((db2_info_ptr = (db2_block_info_t*) list_next(itr)) */
-/* != NULL) { */
-/* if(!strcmp(block_ptr->bgl_block_name, */
-/* db2_info_ptr->bgl_block_name)) { */
-/* if(set == 2) */
-/* break; */
-/* set = 0; */
-/* break; */
-/* } */
-/* if((block_ptr->start[X]==db2_info_ptr->start[X] && */
-/* block_ptr->start[Y]==db2_info_ptr->start[Y] && */
-/* block_ptr->start[Z]==db2_info_ptr->start[Z]) && */
-/* (block_ptr->end[X]==db2_info_ptr->end[X] && */
-/* block_ptr->end[Y]==db2_info_ptr->end[Y] && */
-/* block_ptr->end[Z]==db2_info_ptr->end[Z])) { */
-/* set = 1; */
-/* break; */
-/* } */
-
-/* if((block_ptr->start[X]<=db2_info_ptr->start[X] && */
-/* block_ptr->start[Y]<=db2_info_ptr->start[Y] && */
-/* block_ptr->start[Z]<=db2_info_ptr->start[Z]) && */
-/* (block_ptr->end[X]>=db2_info_ptr->end[X] && */
-/* block_ptr->end[Y]>=db2_info_ptr->end[Y] && */
-/* block_ptr->end[Z]>=db2_info_ptr->end[Z])) { */
-/* set = 2; */
-/* continue; */
-/* } */
-/* } */
-/* list_iterator_destroy(itr); */
-/* } */
-
-/* if (set == 1) { */
-/* block_ptr->letter_num=db2_info_ptr->letter_num; */
-/* part.total_nodes += set_grid_bgl(block_ptr->start, */
-/* block_ptr->end, */
-/* block_ptr->letter_num, */
-/* set); */
-/* } else { */
-/* block_ptr->letter_num=*count; */
-/* part.total_nodes += set_grid_bgl(block_ptr->start, */
-/* block_ptr->end, */
-/* block_ptr->letter_num, */
-/* set); */
-/* (*count)++; */
-/* } */
part.total_nodes = block_ptr->size;
if(block_ptr->slurm_part_name)
part.name = block_ptr->slurm_part_name;
diff --git a/src/smap/smap.c b/src/smap/smap.c
index 062ae99a2f641c7cc9614ff3e0d11279f42000ba..adf929d6aefc17800d85667906395263194ef460 100644
--- a/src/smap/smap.c
+++ b/src/smap/smap.c
@@ -101,18 +101,24 @@ int main(int argc, char *argv[])
if(params.partition[0] == 'r')
params.partition[0] = 'R';
if(params.partition[0] != 'R') {
- char *rack_mid = find_bp_rack_mid(params.partition);
+ i = strlen(params.partition);
+ i -= 3;
+ if(i<0) {
+ printf("No real block was entered\n");
+ goto part_fini;
+ }
+ char *rack_mid = find_bp_rack_mid(params.partition+i);
if(rack_mid)
printf("X=%c Y=%c Z=%c resolves to %s\n",
- params.partition[X],
- params.partition[Y],
- params.partition[Z],
+ params.partition[X+i],
+ params.partition[Y+i],
+ params.partition[Z+i],
rack_mid);
else
printf("X=%c Y=%c Z=%c has no resolve\n",
- params.partition[X],
- params.partition[Y],
- params.partition[Z]);
+ params.partition[X+i],
+ params.partition[Y+i],
+ params.partition[Z+i]);
} else {
int *coord = find_bp_loc(params.partition);
diff --git a/src/srun/msg.c b/src/srun/msg.c
index 6fa5bdaa6305c73bacddd264dd18e21f5a561b8f..c0f6cab7ef0a45d9e5f0ca5ed7a1935132b9e2be 100644
--- a/src/srun/msg.c
+++ b/src/srun/msg.c
@@ -998,7 +998,10 @@ msg_thr_create(srun_job_t *job)
if((job->forked_msg->par_msg->pid = fork()) == -1)
return SLURM_ERROR; // there was an error
else if (job->forked_msg->par_msg->pid == 0)
- { // child:
+ { // child:
+#ifdef DISABLE_LOCALTIME
+ disable_localtime();
+#endif
setsid();
message_thread = 1;
close(job->forked_msg->