diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index 8d02b9587ab2a930c23e7f52c5ee7f2e042be6d4..f2f8f37c576da49ca206c7b3eb333205cff26ca8 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -1927,6 +1927,7 @@ typedef struct {
bool no_alloc;
uint32_t slurmd_debug; /* remote slurmd debug level */
uint32_t pack_offset; /* Pack job offset or NO_VAL */
+ uint32_t task_offset; /* Pack job tsk offset or NO_VAL */
bool parallel_debug;
uint32_t profile; /* Level of acct_gather_profile {all | none} */
char *task_prolog;
diff --git a/src/api/step_io.c b/src/api/step_io.c
index 03810c2be0fa4b93a3307d1f32555f7e66a028e5..43407e128c6439d561b9c7a228bae3c446cd8a9f 100644
--- a/src/api/step_io.c
+++ b/src/api/step_io.c
@@ -598,6 +598,7 @@ static int _file_write(eio_obj_t *obj, List objs)
info->out_remaining,
info->out_msg->header.gtaskid,
info->cio->pack_offset,
+ info->cio->task_offset,
info->cio->label,
info->cio->taskid_width)) < 0) {
list_enqueue(info->cio->free_outgoing, info->out_msg);
@@ -1069,7 +1070,8 @@ _estimate_nports(int nclients, int cli_per_port)
client_io_t *client_io_handler_create(slurm_step_io_fds_t fds, int num_tasks,
int num_nodes, slurm_cred_t *cred,
- bool label, uint32_t pack_offset)
+ bool label, uint32_t pack_offset,
+ uint32_t task_offset)
{
client_io_t *cio;
int i;
@@ -1085,6 +1087,7 @@ client_io_t *client_io_handler_create(slurm_step_io_fds_t fds, int num_tasks,
cio->num_tasks = num_tasks;
cio->num_nodes = num_nodes;
cio->pack_offset = pack_offset;
+ cio->task_offset = task_offset;
cio->label = label;
if (cio->label)
diff --git a/src/api/step_io.h b/src/api/step_io.h
index 3c2bd8a138fdfe70cf20df2cb5d5267622bc8888..f1e7553fe0242c1853e23b0223566de908c46e1c 100644
--- a/src/api/step_io.h
+++ b/src/api/step_io.h
@@ -45,7 +45,9 @@ struct client_io {
int num_nodes;
bool label;
int taskid_width; /* characters needed for task_id label */
- uint32_t pack_offset;
+ uint32_t pack_offset; /* offset within a pack-job or NO_VAL */
+ uint32_t task_offset; /* task offset within a pack-job or NO_VAL */
+
char *io_key;
/* internal variables */
@@ -103,7 +105,8 @@ typedef struct client_io client_io_t;
*/
client_io_t *client_io_handler_create(slurm_step_io_fds_t fds, int num_tasks,
int num_nodes, slurm_cred_t *cred,
- bool label, uint32_t pack_offset);
+ bool label, uint32_t pack_offset,
+ uint32_t task_offset);
int client_io_handler_start(client_io_t *cio);
diff --git a/src/api/step_launch.c b/src/api/step_launch.c
index dfef26eda6854b0964c044f2d7b722749284842e..64fb7d8597c41a79fa4de31f210e22ef0b88d85f 100644
--- a/src/api/step_launch.c
+++ b/src/api/step_launch.c
@@ -165,6 +165,7 @@ void slurm_step_launch_params_t_init (slurm_step_launch_params_t *ptr)
ptr->cpu_freq_max = NO_VAL;
ptr->cpu_freq_gov = NO_VAL;
ptr->pack_offset = NO_VAL;
+ ptr->task_offset = NO_VAL;
}
/*
@@ -372,7 +373,8 @@ extern int slurm_step_launch(slurm_step_ctx_t *ctx,
launch.nnodes,
ctx->step_resp->cred,
params->labelio,
- params->pack_offset);
+ params->pack_offset,
+ params->task_offset);
if (ctx->launch_state->io.normal == NULL) {
rc = SLURM_ERROR;
goto fail1;
@@ -566,7 +568,8 @@ extern int slurm_step_launch_add(slurm_step_ctx_t *ctx,
launch.nnodes,
ctx->step_resp->cred,
params->labelio,
- params->pack_offset);
+ params->pack_offset,
+ params->task_offset);
if (ctx->launch_state->io.normal == NULL) {
rc = SLURM_ERROR;
goto fail1;
diff --git a/src/common/write_labelled_message.c b/src/common/write_labelled_message.c
index fd6868f737bc5d71e5a5101a9ada4b269060dfd1..a5ded00a9b832540a23a5e9306959ba989b671fa 100644
--- a/src/common/write_labelled_message.c
+++ b/src/common/write_labelled_message.c
@@ -48,18 +48,20 @@
#include "src/common/xmalloc.h"
#include "src/common/xstring.h"
-static char *_build_label(int taskid, int taskid_width, uint32_t pack_offset);
+static char *_build_label(int task_id, int task_id_width, uint32_t pack_offset,
+ uint32_t task_offset);
static int _write_line(int fd, char *prefix, char *suffix, void *buf, int len);
/*
- * fd is the file descriptor to write to
- * buf is the char buffer to write
- * len is the buffer length in bytes
- * taskid is will be used in the label
- * pack_offset is the offset within a pack-job or NO_VAL
- * label if true, prepend each line of the buffer with a
+ * fd is the file descriptor to write to
+ * buf is the char buffer to write
+ * len is the buffer length in bytes
+ * task_id is will be used in the label
+ * pack_offset is the offset within a pack-job or NO_VAL
+ * task_offset is the task offset within a pack-job or NO_VAL
+ * label if true, prepend each line of the buffer with a
* label for the task id
- * taskid_width is the number of digits to use for the task id
+ * task_id_width is the number of digits to use for the task id
*
* Write as many lines from the message as possible. Return
* the number of bytes from the message that have been written,
@@ -69,9 +71,9 @@ static int _write_line(int fd, char *prefix, char *suffix, void *buf, int len);
* in a '\n'), then add a newline to the output file, but only
* in label mode.
*/
-extern int write_labelled_message(int fd, void *buf, int len, int taskid,
- uint32_t pack_offset, bool label,
- int taskid_width)
+extern int write_labelled_message(int fd, void *buf, int len, int task_id,
+ uint32_t pack_offset, uint32_t task_offset,
+ bool label, int task_id_width)
{
void *start, *end;
char *prefix = NULL, *suffix = NULL;
@@ -80,8 +82,11 @@ extern int write_labelled_message(int fd, void *buf, int len, int taskid,
int line_len;
int rc = -1;
- if (label)
- prefix = _build_label(taskid, taskid_width, pack_offset);
+ if (label) {
+ prefix = _build_label(task_id, task_id_width, pack_offset,
+ task_offset);
+ }
+
while (remaining > 0) {
start = buf + written;
end = memchr(start, '\n', remaining);
@@ -118,14 +123,22 @@ done:
/*
* Build line label. Call xfree() to release returned memory
*/
-static char *_build_label(int taskid, int taskid_width, uint32_t pack_offset)
+static char *_build_label(int task_id, int task_id_width,
+ uint32_t pack_offset, uint32_t task_offset)
{
char *buf = NULL;
- if (pack_offset == NO_VAL)
- xstrfmtcat(buf, "%*d: ", taskid_width, taskid);
- else
- xstrfmtcat(buf, "P%u %*d: ", pack_offset, taskid_width, taskid);
+ if (pack_offset != NO_VAL) {
+ if (task_offset == NO_VAL) {
+ xstrfmtcat(buf, "P%u %*d: ", pack_offset, task_id_width,
+ task_id);
+ } else {
+ xstrfmtcat(buf, "%*d: ", task_id_width,
+ (task_id + task_offset));
+ }
+ } else {
+ xstrfmtcat(buf, "%*d: ", task_id_width, task_id);
+ }
return buf;
}
diff --git a/src/common/write_labelled_message.h b/src/common/write_labelled_message.h
index 12ec33155de98232618f85d60f334ede050f12aa..4c9d76bce0d33813e5121ca503ba359558e18ac9 100644
--- a/src/common/write_labelled_message.h
+++ b/src/common/write_labelled_message.h
@@ -42,14 +42,15 @@
#include "slurm/slurm.h"
/*
- * fd is the file descriptor to write to
- * buf is the char buffer to write
- * len is the buffer length in bytes
- * taskid is will be used in the label
- * pack_offset is the offset within a pack-job or NO_VAL
- * label if true, prepend each line of the buffer with a
+ * fd is the file descriptor to write to
+ * buf is the char buffer to write
+ * len is the buffer length in bytes
+ * task_id is will be used in the label
+ * pack_offset is the offset within a pack-job or NO_VAL
+ * task_offset is the task offset within a pack-job or NO_VAL
+ * label if true, prepend each line of the buffer with a
* label for the task id
- * taskid_width is the number of digits to use for the task id
+ * task_id_width is the number of digits to use for the task id
*
* Write as many lines from the message as possible. Return
* the number of bytes from the message that have been written,
@@ -59,8 +60,8 @@
* in a '\n'), then add a newline to the output file, but only
* in label mode.
*/
-extern int write_labelled_message(int fd, void *buf, int len, int taskid,
- uint32_t pack_offset, bool label,
- int taskid_width);
+extern int write_labelled_message(int fd, void *buf, int len, int task_id,
+ uint32_t pack_offset, uint32_t task_offset,
+ bool label, int task_id_width);
#endif
diff --git a/src/plugins/launch/aprun/launch_aprun.c b/src/plugins/launch/aprun/launch_aprun.c
index 6f35b0e3ee2f2c75c051655ea214c409e3964576..3978e3ab75781f4b027ee6d06ced529787a6a1cd 100644
--- a/src/plugins/launch/aprun/launch_aprun.c
+++ b/src/plugins/launch/aprun/launch_aprun.c
@@ -703,8 +703,7 @@ extern int launch_p_handle_multi_prog_verify(int command_pos, opt_t *opt_local)
extern int launch_p_create_job_step(srun_job_t *job, bool use_all_cpus,
void (*signal_function)(int),
- sig_atomic_t *destroy_job, opt_t *opt_local,
- int pack_offset)
+ sig_atomic_t *destroy_job, opt_t *opt_local)
{
if (opt_local->launch_cmd) {
int i = 0;
@@ -773,8 +772,7 @@ extern int launch_p_step_launch(srun_job_t *job, slurm_step_io_fds_t *cio_fds,
return rc;
}
-extern int launch_p_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local,
- int pack_offset)
+extern int launch_p_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local)
{
return SLURM_SUCCESS;
}
diff --git a/src/plugins/launch/poe/launch_poe.c b/src/plugins/launch/poe/launch_poe.c
index 0419cf53227943cd5e8217eff3d5b4ceaaa04789..82ea5b0d53adf9fbb9cbdef6b00c21cc77a8464a 100644
--- a/src/plugins/launch/poe/launch_poe.c
+++ b/src/plugins/launch/poe/launch_poe.c
@@ -173,7 +173,7 @@ static void _setenv(const char *name, const char *value, int overwrite,
{
char *key = NULL;
- if (pack_offset == -1) {
+ if (pack_offset == NO_VAL) {
setenv(name, value, overwrite);
} else {
xstrfmtcat(key, "%s_PACK_GROUP_%d", name, pack_offset);
@@ -357,8 +357,7 @@ extern int launch_p_handle_multi_prog_verify(int command_pos, opt_t *opt_local)
extern int launch_p_create_job_step(srun_job_t *job, bool use_all_cpus,
void (*signal_function)(int),
- sig_atomic_t *destroy_job, opt_t *opt_local,
- int pack_offset)
+ sig_atomic_t *destroy_job, opt_t *opt_local)
{
char dname[512], value[32];
char *protocol = "mpi";
@@ -773,7 +772,7 @@ extern int launch_p_create_job_step(srun_job_t *job, bool use_all_cpus,
" -stdoutmode unordered");
}
- _propagate_srun_opts(nnodes, ntasks, opt_local, pack_offset);
+ _propagate_srun_opts(nnodes, ntasks, opt_local, job->pack_offset);
setenv("SLURM_STARTED_STEP", "YES", 1);
//disable_status = opt_local->disable_status;
//quit_on_intr = opt_local->quit_on_intr;
@@ -849,8 +848,7 @@ extern int launch_p_step_launch(srun_job_t *job, slurm_step_io_fds_t *cio_fds,
return rc;
}
-extern int launch_p_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local,
- int pack_offset)
+extern int launch_p_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local)
{
return SLURM_SUCCESS;
}
diff --git a/src/plugins/launch/runjob/launch_runjob.c b/src/plugins/launch/runjob/launch_runjob.c
index 519099a1f9b35dfa1b77b609d15eaec816a4bcdf..ea460ce61210fbaff2dc46e14a58d919fc65cf87 100644
--- a/src/plugins/launch/runjob/launch_runjob.c
+++ b/src/plugins/launch/runjob/launch_runjob.c
@@ -443,8 +443,7 @@ extern int launch_p_handle_multi_prog_verify(int command_posm, opt_t *opt_local)
extern int launch_p_create_job_step(srun_job_t *job, bool use_all_cpus,
void (*signal_function)(int),
- sig_atomic_t *destroy_job, opt_t *opt_local,
- int pack_offset)
+ sig_atomic_t *destroy_job, opt_t *opt_local)
{
if (opt_local->launch_cmd) {
int i = 0;
@@ -486,8 +485,7 @@ extern int launch_p_step_launch(srun_job_t *job, slurm_step_io_fds_t *cio_fds,
return SLURM_SUCCESS;
}
-extern int launch_p_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local,
- int pack_offset, uint32_t task_offset)
+extern int launch_p_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local)
{
return SLURM_SUCCESS;
}
diff --git a/src/plugins/launch/slurm/launch_slurm.c b/src/plugins/launch/slurm/launch_slurm.c
index c88ed4d155acccad7c080c45e00635b565b63bce..7536275aa5acd79f39f49ea4a9d702df1e131ee4 100644
--- a/src/plugins/launch/slurm/launch_slurm.c
+++ b/src/plugins/launch/slurm/launch_slurm.c
@@ -555,8 +555,7 @@ extern int launch_p_handle_multi_prog_verify(int command_pos, opt_t *opt_local)
extern int launch_p_create_job_step(srun_job_t *job, bool use_all_cpus,
void (*signal_function)(int),
- sig_atomic_t *destroy_job, opt_t *opt_local,
- int pack_offset)
+ sig_atomic_t *destroy_job, opt_t *opt_local)
{
if (launch_common_create_job_step(job, use_all_cpus,
signal_function, destroy_job,
@@ -666,6 +665,7 @@ extern int launch_p_step_launch(srun_job_t *job, slurm_step_io_fds_t *cio_fds,
launch_params.remote_input_filename = fname_remote_string(job->ifname);
launch_params.remote_error_filename = fname_remote_string(job->efname);
launch_params.pack_offset = job->pack_offset;
+ launch_params.task_offset = job->task_offset;
launch_params.partition = job->partition;
launch_params.profile = opt_local->profile;
launch_params.task_prolog = opt_local->task_prolog;
@@ -788,8 +788,7 @@ cleanup:
return rc;
}
-extern int launch_p_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local,
- int pack_offset)
+extern int launch_p_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local)
{
int rc = 0;
@@ -800,9 +799,9 @@ extern int launch_p_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local,
retry_step_begin = false;
slurm_step_ctx_destroy(job->step_ctx);
if (got_alloc)
- rc = create_job_step(job, true, opt_local, pack_offset);
+ rc = create_job_step(job, true, opt_local);
else
- rc = create_job_step(job, false, opt_local,pack_offset);
+ rc = create_job_step(job, false, opt_local);
if (rc < 0)
exit(error_exit);
rc = -1;
diff --git a/src/sattach/sattach.c b/src/sattach/sattach.c
index 1703c93b7900a8cd1c5134b352c05069fc4849f8..d0e12e4b59ac76022f89c0d8e4154844d1e093cf 100644
--- a/src/sattach/sattach.c
+++ b/src/sattach/sattach.c
@@ -174,7 +174,7 @@ int sattach(int argc, char **argv)
io = client_io_handler_create(opt.fds, layout->task_cnt,
layout->node_cnt, fake_cred,
- opt.labelio, NO_VAL);
+ opt.labelio, NO_VAL, NO_VAL);
client_io_handler_start(io);
if (opt.pty) {
diff --git a/src/slurmd/slurmstepd/io.c b/src/slurmd/slurmstepd/io.c
index b38d40792acc63848f26101ffdb39967c33d9ffe..f9e0b8a8098d9ab6292848066f2b960615e171a5 100644
--- a/src/slurmd/slurmstepd/io.c
+++ b/src/slurmd/slurmstepd/io.c
@@ -533,8 +533,10 @@ _local_file_write(eio_obj_t *obj, List objs)
io_hdr_packed_size();
}
- /* This code to make a buffer, fill it, unpack its contents, and free
- it is just used to read the header to get the global task id. */
+ /*
+ * This code to make a buffer, fill it, unpack its contents, and free
+ * it is just used to read the header to get the global task id.
+ */
header_tmp_buf = create_buf(client->out_msg->data,
client->out_msg->length);
if (!header_tmp_buf) {
@@ -545,8 +547,10 @@ _local_file_write(eio_obj_t *obj, List objs)
header_tmp_buf->head = NULL; /* CLANG false positive bug here */
free_buf(header_tmp_buf);
- /* A zero-length message indicates the end of a stream from one
- of the tasks. Just free the message and return. */
+ /*
+ * A zero-length message indicates the end of a stream from one
+ * of the tasks. Just free the message and return.
+ */
if (header.length == 0) {
_free_outgoing_msg(client->out_msg, client->job);
client->out_msg = NULL;
@@ -558,7 +562,8 @@ _local_file_write(eio_obj_t *obj, List objs)
(client->out_msg->length - client->out_remaining);
n = write_labelled_message(obj->fd, buf, client->out_remaining,
header.gtaskid, client->job->pack_offset,
- client->labelio, client->taskid_width);
+ client->job->task_offset, client->labelio,
+ client->taskid_width);
if (n < 0) {
client->out_eof = true;
_free_all_outgoing_msgs(client->msg_queue, client->job);
diff --git a/src/slurmd/slurmstepd/slurmstepd_job.h b/src/slurmd/slurmstepd/slurmstepd_job.h
index 62b057c99edd9438f8b35e95a0a14396c390d3d3..6438e745408689d224bbc3e175165572864cfb81 100644
--- a/src/slurmd/slurmstepd/slurmstepd_job.h
+++ b/src/slurmd/slurmstepd/slurmstepd_job.h
@@ -135,6 +135,8 @@ typedef struct {
uint32_t nodeid; /* relative position of this node in job */
uint32_t node_tasks; /* number of tasks on *this* node */
uint32_t pack_offset; /* pack job offset or NO_VAL */
+//FIXME-PACK - To populate
+ uint32_t task_offset; /* pack job task offset or NO_VAL */
uint16_t *task_cnts; /* Number of tasks on each node in job */
uint32_t cpus_per_task; /* number of cpus desired per task */
uint32_t debug; /* debug level for job slurmd */
diff --git a/src/srun/libsrun/allocate.c b/src/srun/libsrun/allocate.c
index d159e554bee72366311f19929a61eb0e0612bae6..2f92e8818fc27ea70132e1cef49da934f70b0fac 100644
--- a/src/srun/libsrun/allocate.c
+++ b/src/srun/libsrun/allocate.c
@@ -1139,10 +1139,9 @@ job_desc_msg_destroy(job_desc_msg_t *j)
}
}
-extern int create_job_step(srun_job_t *job, bool use_all_cpus, opt_t *opt_local,
- int pack_offset)
+extern int create_job_step(srun_job_t *job, bool use_all_cpus, opt_t *opt_local)
{
return launch_g_create_job_step(job, use_all_cpus,
_signal_while_allocating,
- &destroy_job, opt_local, pack_offset);
+ &destroy_job, opt_local);
}
diff --git a/src/srun/libsrun/allocate.h b/src/srun/libsrun/allocate.h
index e250ec80333beebe65c8e672f1e893b9207a610d..07e55cb067b258077ace4347caa944f1a3924b36 100644
--- a/src/srun/libsrun/allocate.h
+++ b/src/srun/libsrun/allocate.h
@@ -105,12 +105,10 @@ extern List existing_allocation(void);
* After returning, 'j' is filled in with information for job step.
* IN use_all_cpus - true to use every CPU allocated to the job
* IN opt_local - options used to create job step
- * IN pack_offset - offset within a pack job, -1 if not part of pack job
*
* Returns -1 if job step creation failure, 0 otherwise
*/
-int create_job_step(srun_job_t *j, bool use_all_cpus, opt_t *opt_local,
- int pack_offset);
+int create_job_step(srun_job_t *j, bool use_all_cpus, opt_t *opt_local);
/* set the job for debugging purpose */
void set_allocate_job(srun_job_t *job);
diff --git a/src/srun/libsrun/launch.c b/src/srun/libsrun/launch.c
index c589ed39b12075bdfcb548b872854b1dd9cce440..a1e754ba5552bb1fa8839ee721ac636c1f868e18 100644
--- a/src/srun/libsrun/launch.c
+++ b/src/srun/libsrun/launch.c
@@ -52,15 +52,15 @@ typedef struct {
int (*handle_multi_prog) (int command_pos, opt_t *opt_local);
int (*create_job_step) (srun_job_t *job, bool use_all_cpus,
void (*signal_function)(int),
- sig_atomic_t *destroy_job, opt_t *opt_local,
- int pack_offset);
+ sig_atomic_t *destroy_job,
+ opt_t *opt_local);
int (*step_launch) (srun_job_t *job,
slurm_step_io_fds_t *cio_fds,
uint32_t *global_rc,
slurm_step_launch_callbacks_t *
step_callbacks, opt_t *opt_local);
int (*step_wait) (srun_job_t *job, bool got_alloc,
- opt_t *opt_local, int pack_offset);
+ opt_t *opt_local);
int (*step_terminate) (void);
void (*print_status) (void);
void (*fwd_signal) (int signal);
@@ -508,14 +508,13 @@ extern int launch_g_handle_multi_prog_verify(int command_pos, opt_t *opt_local)
extern int launch_g_create_job_step(srun_job_t *job, bool use_all_cpus,
void (*signal_function)(int),
- sig_atomic_t *destroy_job, opt_t *opt_local,
- int pack_offset)
+ sig_atomic_t *destroy_job, opt_t *opt_local)
{
if (launch_init() < 0)
return SLURM_ERROR;
return (*(ops.create_job_step))(job, use_all_cpus, signal_function,
- destroy_job, opt_local, pack_offset);
+ destroy_job, opt_local);
}
extern int launch_g_step_launch(srun_job_t *job, slurm_step_io_fds_t *cio_fds,
@@ -530,13 +529,12 @@ extern int launch_g_step_launch(srun_job_t *job, slurm_step_io_fds_t *cio_fds,
opt_local);
}
-extern int launch_g_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local,
- int pack_offset)
+extern int launch_g_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local)
{
if (launch_init() < 0)
return SLURM_ERROR;
- return (*(ops.step_wait))(job, got_alloc, opt_local, pack_offset);
+ return (*(ops.step_wait))(job, got_alloc, opt_local);
}
extern int launch_g_step_terminate(void)
diff --git a/src/srun/libsrun/launch.h b/src/srun/libsrun/launch.h
index d7c9a26c087c4165502e1f89bf5d2d57bf680a09..e064200daddf14e4a8f5b3992cd6e746482215d9 100644
--- a/src/srun/libsrun/launch.h
+++ b/src/srun/libsrun/launch.h
@@ -124,14 +124,13 @@ extern int launch_g_handle_multi_prog_verify(int command_pos, opt_t *opt_local);
* IN destroy_job - pointer to a global flag signifying if the job was
* canceled while allocating.
* IN opt_local - options used for step creation
- * IN pack_offset - offset within a pack job, -1 if not part of pack job
*
* RETURN SLURM_SUCCESS on success || SLURM_ERROR else wise
*/
extern int launch_g_create_job_step(srun_job_t *job, bool use_all_cpus,
void (*signal_function)(int),
- sig_atomic_t *destroy_job, opt_t *opt_local,
- int pack_offset);
+ sig_atomic_t *destroy_job,
+ opt_t *opt_local);
/*
* launch_g_step_launch() is called to launch the job step that
@@ -155,12 +154,11 @@ extern int launch_g_step_launch(srun_job_t *job, slurm_step_io_fds_t *cio_fds,
* IN/OUT job - the job waiting to finish.
* IN got_alloc - if the resource allocation was created inside srun
* IN opt_local - options used for step creation
- * IN pack_offset - offset within a pack job, -1 if not part of pack job
*
* RETURN SLURM_SUCCESS on success || SLURM_ERROR else wise
*/
-extern int launch_g_step_wait(srun_job_t *job, bool got_alloc, opt_t *opt_local,
- int pack_offset);
+extern int launch_g_step_wait(srun_job_t *job, bool got_alloc,
+ opt_t *opt_local);
/*
* launch_g_step_terminate() is called to end the job step.
diff --git a/src/srun/libsrun/opt.c b/src/srun/libsrun/opt.c
index 996fd8380ee9e94a9672ba021007be300a28bee2..2b970f00f877edf6ebfed9bc36030358666e7bb4 100644
--- a/src/srun/libsrun/opt.c
+++ b/src/srun/libsrun/opt.c
@@ -584,7 +584,7 @@ extern int initialize_and_process_args(int argc, char **argv, int *argc_off)
/* Massage ntasks value earlier than normal */
if (!opt.ntasks_set)
opt.ntasks = _get_task_count();
- launch_g_create_job_step(NULL, 0, NULL, NULL, &opt, -1);
+ launch_g_create_job_step(NULL, 0, NULL, NULL, &opt);
exit(0);
}
pending_append = true;
diff --git a/src/srun/libsrun/srun_job.c b/src/srun/libsrun/srun_job.c
index b8a56108103306ae7f685978833333da4cbdde18..371ebc1e12302aeda415d9297f12b989ff7a7c3f 100644
--- a/src/srun/libsrun/srun_job.c
+++ b/src/srun/libsrun/srun_job.c
@@ -647,7 +647,8 @@ static int _create_job_step(srun_job_t *job, bool use_all_cpus,
{
ListIterator opt_iter = NULL, job_iter;
opt_t *opt_local = &opt;
- int pack_offset = 0, rc = 0;
+ uint32_t pack_offset = 0, task_offset = 0;
+ int rc = 0;
if (srun_job_list) {
if (opt_list)
@@ -658,18 +659,20 @@ static int _create_job_step(srun_job_t *job, bool use_all_cpus,
opt_local = (opt_t *) list_next(opt_iter);
if (!opt_local)
fatal("%s: opt_list too short", __func__);
- rc = create_job_step(job, use_all_cpus, opt_local,
- pack_offset);
+ job->pack_offset = pack_offset;
+ job->task_offset = task_offset;
+ rc = create_job_step(job, use_all_cpus, opt_local);
if (rc < 0)
break;
pack_offset++;
+ task_offset += job->ntasks;
}
list_iterator_destroy(job_iter);
if (opt_iter)
list_iterator_destroy(opt_iter);
return rc;
} else if (job) {
- return create_job_step(job, use_all_cpus, &opt, -1);
+ return create_job_step(job, use_all_cpus, &opt);
} else {
return -1;
}
@@ -682,7 +685,7 @@ extern void create_srun_job(void **p_job, bool *got_alloc,
List job_resp_list = NULL, srun_job_list = NULL;
ListIterator opt_iter, resp_iter;
srun_job_t *job = NULL;
- int max_pack_offset, pack_offset = -1;
+ int i, max_pack_offset, pack_offset = -1;
opt_t *opt_local;
uint32_t my_job_id = 0;
bool begin_error_logged = false;
@@ -714,7 +717,7 @@ extern void create_srun_job(void **p_job, bool *got_alloc,
error("Job creation failure.");
exit(error_exit);
}
- if (create_job_step(job, false, &opt, -1) < 0)
+ if (create_job_step(job, false, &opt) < 0)
exit(error_exit);
} else if ((job_resp_list = existing_allocation())) {
srun_job_list = list_create(NULL);
@@ -812,11 +815,14 @@ extern void create_srun_job(void **p_job, bool *got_alloc,
} /* More pack job components */
list_iterator_destroy(resp_iter);
- if (list_count(srun_job_list) == 0) {
+ i = list_count(srun_job_list);
+ if (i == 0) {
error("No directives to start application on any available "
"pack job components");
exit(error_exit);
}
+ if (i == 1)
+ FREE_NULL_LIST(srun_job_list); /* Just use "job" */
max_pack_offset = get_max_pack_group();
pack_offset = list_count(job_resp_list) - 1;
if (max_pack_offset > pack_offset) {
@@ -1094,6 +1100,8 @@ static srun_job_t *_job_create_structure(allocation_info_t *ainfo,
job->nodelist = xstrdup(ainfo->nodelist);
job->partition = xstrdup(ainfo->partition);
job->stepid = ainfo->stepid;
+ job->pack_offset = NO_VAL;
+ job->task_offset = NO_VAL;
#if defined HAVE_BG
//#if defined HAVE_BGQ && defined HAVE_BG_FILES
diff --git a/src/srun/libsrun/srun_job.h b/src/srun/libsrun/srun_job.h
index 93c648dd057174317bd5e020d32e55ef0f38f3dc..7c0c54148a085ed84e6e8e4ea8b992a72255e850 100644
--- a/src/srun/libsrun/srun_job.h
+++ b/src/srun/libsrun/srun_job.h
@@ -81,7 +81,9 @@ typedef struct srun_job {
int fir_nodeid;
uint32_t jobid; /* assigned job id */
uint32_t stepid; /* assigned step id */
- uint32_t pack_offset; /* pack job offset */
+ uint32_t pack_offset; /* pack job offset or NO_VAL */
+//FIXME-PACK - Where is this set??
+ uint32_t task_offset; /* pack job tsk offset or NO_VAL */
uint32_t cpu_count; /* allocated CPUs */
uint32_t nhosts; /* node count */
diff --git a/src/srun/srun.c b/src/srun/srun.c
index f26f173d5ca2fc9f8243ff3335af857b63f9d58d..9e0bdee969e99ef73460cb2e3a6913ee5eec9d81 100644
--- a/src/srun/srun.c
+++ b/src/srun/srun.c
@@ -202,8 +202,10 @@ static void *_launch_one_app(void *data)
slurm_step_io_fds_t cio_fds = SLURM_STEP_IO_FDS_INITIALIZER;
slurm_step_launch_callbacks_t step_callbacks;
+//FIXME-PACK - Perhaps we set pack_offset at some earlier point?
+//FIXME-PACK - Add support for overlapping rank IDs
if (opt_local->pack_grp_bits)
- job->pack_offset = bit_ffs(opt_local->pack_grp_bits);
+ job->pack_offset = bit_ffs(opt_local->pack_grp_bits); // SET HERE
else
job->pack_offset = NO_VAL;
@@ -234,7 +236,7 @@ relaunch:
if (!launch_g_step_launch(job, &cio_fds, &global_rc, &step_callbacks,
opt_local)) {
- if (launch_g_step_wait(job, got_alloc, opt_local, -1) == -1)
+ if (launch_g_step_wait(job, got_alloc, opt_local) == -1)
goto relaunch;
}