static int _unpack_step_complete_msg(dbd_step_comp_msg_t **msg,
uint16_t rpc_version, Buf buffer)
{
uint32_t uint32_tmp;
dbd_step_comp_msg_t *msg_ptr = xmalloc(sizeof(dbd_step_comp_msg_t));
*msg = msg_ptr;
if (rpc_version >= SLURM_MIN_PROTOCOL_VERSION) {
safe_unpack32(&msg_ptr->assoc_id, buffer);
safe_unpack64(&msg_ptr->db_index, buffer);
safe_unpack_time(&msg_ptr->end_time, buffer);
safe_unpack32(&msg_ptr->exit_code, buffer);
jobacctinfo_unpack((struct jobacctinfo **)&msg_ptr->jobacct,
rpc_version, PROTOCOL_TYPE_DBD, buffer, 1);
safe_unpack32(&msg_ptr->job_id, buffer);
safe_unpack_time(&msg_ptr->job_submit_time, buffer);
safe_unpackstr_xmalloc(&msg_ptr->job_tres_alloc_str,
&uint32_tmp, buffer);
safe_unpack32(&msg_ptr->req_uid, buffer);
safe_unpack_time(&msg_ptr->start_time, buffer);
safe_unpack16(&msg_ptr->state, buffer);
safe_unpack32(&msg_ptr->step_id, buffer);
safe_unpack32(&msg_ptr->total_tasks, buffer);
} else
goto unpack_error;
return SLURM_SUCCESS;
unpack_error:
debug2("slurmdbd_unpack_step_complete_msg:"
"unpack_error: size_buf(buffer) %u",
size_buf(buffer));
slurmdbd_free_step_complete_msg(msg_ptr);
*msg = NULL;
return SLURM_ERROR;
}
extern int jobacctinfo_getinfo(
jobacctinfo_t *jobacct, enum jobacct_data_type type, void *data,
uint16_t protocol_version)
{
int rc = SLURM_SUCCESS;
int *fd = (int *)data;
uint32_t *uint32 = (uint32_t *) data;
uint64_t *uint64 = (uint64_t *) data;
double *dub = (double *) data;
jobacct_id_t *jobacct_id = (jobacct_id_t *) data;
struct rusage *rusage = (struct rusage *)data;
struct jobacctinfo *send = (struct jobacctinfo *) data;
char *buf = NULL;
if (!plugin_polling)
return SLURM_SUCCESS;
/* jobacct needs to be allocated before this is called. */
xassert(jobacct);
switch (type) {
case JOBACCT_DATA_TOTAL:
memcpy(send, jobacct, sizeof(struct jobacctinfo));
break;
case JOBACCT_DATA_PIPE:
if (protocol_version >= SLURM_MIN_PROTOCOL_VERSION) {
int len;
Buf buffer;
safe_read(*fd, &len, sizeof(int));
buf = xmalloc(len);
safe_read(*fd, buf, len);
buffer = create_buf(buf, len);
jobacctinfo_unpack(&jobacct, protocol_version,
PROTOCOL_TYPE_SLURM, buffer, 0);
free_buf(buffer);
}
break;
case JOBACCT_DATA_RUSAGE:
memset(rusage, 0, sizeof(struct rusage));
rusage->ru_utime.tv_sec = jobacct->user_cpu_sec;
rusage->ru_utime.tv_usec = jobacct->user_cpu_usec;
rusage->ru_stime.tv_sec = jobacct->sys_cpu_sec;
rusage->ru_stime.tv_usec = jobacct->sys_cpu_usec;
break;
case JOBACCT_DATA_MAX_RSS:
*uint64 = jobacct->max_rss;
break;
case JOBACCT_DATA_MAX_RSS_ID:
*jobacct_id = jobacct->max_rss_id;
break;
case JOBACCT_DATA_TOT_RSS:
*uint64 = jobacct->tot_rss;
break;
case JOBACCT_DATA_MAX_VSIZE:
*uint64 = jobacct->max_vsize;
break;
case JOBACCT_DATA_MAX_VSIZE_ID:
*jobacct_id = jobacct->max_vsize_id;
break;
case JOBACCT_DATA_TOT_VSIZE:
*uint64 = jobacct->tot_vsize;
break;
case JOBACCT_DATA_MAX_PAGES:
*uint64 = jobacct->max_pages;
break;
case JOBACCT_DATA_MAX_PAGES_ID:
*jobacct_id = jobacct->max_pages_id;
break;
case JOBACCT_DATA_TOT_PAGES:
*uint64 = jobacct->tot_pages;
break;
case JOBACCT_DATA_MIN_CPU:
*uint32 = jobacct->min_cpu;
break;
case JOBACCT_DATA_MIN_CPU_ID:
*jobacct_id = jobacct->min_cpu_id;
break;
case JOBACCT_DATA_TOT_CPU:
*dub = jobacct->tot_cpu;
break;
case JOBACCT_DATA_ACT_CPUFREQ:
*uint32 = jobacct->act_cpufreq;
break;
case JOBACCT_DATA_CONSUMED_ENERGY:
*uint64 = jobacct->energy.consumed_energy;
break;
case JOBACCT_DATA_MAX_DISK_READ:
*dub = jobacct->max_disk_read;
break;
case JOBACCT_DATA_MAX_DISK_READ_ID:
*jobacct_id = jobacct->max_disk_read_id;
break;
case JOBACCT_DATA_TOT_DISK_READ:
*dub = jobacct->tot_disk_read;
break;
case JOBACCT_DATA_MAX_DISK_WRITE:
*dub = jobacct->max_disk_write;
break;
case JOBACCT_DATA_MAX_DISK_WRITE_ID:
*jobacct_id = jobacct->max_disk_write_id;
break;
case JOBACCT_DATA_TOT_DISK_WRITE:
*dub = jobacct->tot_disk_write;
break;
default:
debug("jobacct_g_set_getinfo data_type %d invalid", type);
}
return rc;
rwfail:
xfree(buf);
return SLURM_ERROR;
}
static int
_handle_completion(int fd, stepd_step_rec_t *job, uid_t uid)
{
int rc = SLURM_SUCCESS;
int errnum = 0;
int first;
int last;
jobacctinfo_t *jobacct = NULL;
int step_rc;
char* buf;
int len;
Buf buffer;
int version; /* For future use */
bool lock_set = false;
debug("_handle_completion for job %u.%u",
job->jobid, job->stepid);
debug3(" uid = %d", uid);
if (!_slurm_authorized_user(uid)) {
debug("step completion message from uid %ld for job %u.%u ",
(long)uid, job->jobid, job->stepid);
rc = -1;
errnum = EPERM;
/* Send the return code and errno */
safe_write(fd, &rc, sizeof(int));
safe_write(fd, &errnum, sizeof(int));
return SLURM_SUCCESS;
}
safe_read(fd, &version, sizeof(int));
safe_read(fd, &first, sizeof(int));
safe_read(fd, &last, sizeof(int));
safe_read(fd, &step_rc, sizeof(int));
/*
* We must not use getinfo over a pipe with slurmd here
* Indeed, slurmstepd does a large use of setinfo over a pipe
* with slurmd and doing the reverse can result in a deadlock
* scenario with slurmd :
* slurmd(lockforread,write)/slurmstepd(write,lockforread)
* Do pack/unpack instead to be sure of independances of
* slurmd and slurmstepd
*/
safe_read(fd, &len, sizeof(int));
buf = xmalloc(len);
safe_read(fd, buf, len);
buffer = create_buf(buf, len);
jobacctinfo_unpack(&jobacct, SLURM_PROTOCOL_VERSION,
PROTOCOL_TYPE_SLURM, buffer, 1);
free_buf(buffer);
/*
* Record the completed nodes
*/
pthread_mutex_lock(&step_complete.lock);
lock_set = true;
if (! step_complete.wait_children) {
rc = -1;
errnum = ETIMEDOUT; /* not used anyway */
goto timeout;
}
/* SlurmUser or root can craft a launch without a valid credential
* ("srun --no-alloc ...") and no tree information can be built
* without the hostlist from the credential. */
if (step_complete.rank >= 0) {
#if 0
char bits_string[128];
debug2("Setting range %d (bit %d) through %d(bit %d)",
first, first-(step_complete.rank+1),
last, last-(step_complete.rank+1));
bit_fmt(bits_string, sizeof(bits_string), step_complete.bits);
debug2(" before bits: %s", bits_string);
#endif
bit_nset(step_complete.bits,
first - (step_complete.rank+1),
last - (step_complete.rank+1));
#if 0
bit_fmt(bits_string, sizeof(bits_string), step_complete.bits);
debug2(" after bits: %s", bits_string);
#endif
}
step_complete.step_rc = MAX(step_complete.step_rc, step_rc);
/************* acct stuff ********************/
jobacctinfo_aggregate(step_complete.jobacct, jobacct);
timeout:
jobacctinfo_destroy(jobacct);
/*********************************************/
/* Send the return code and errno, we do this within the locked
* region to ensure that the stepd doesn't exit before we can
* perform this send. */
safe_write(fd, &rc, sizeof(int));
safe_write(fd, &errnum, sizeof(int));
pthread_cond_signal(&step_complete.cond);
pthread_mutex_unlock(&step_complete.lock);
return SLURM_SUCCESS;
rwfail: if (lock_set) {
pthread_cond_signal(&step_complete.cond);
pthread_mutex_unlock(&step_complete.lock);
}
return SLURM_FAILURE;
}
