/* Wait up to sleep_time for RPC from slurmctld indicating resource allocation
* has occured.
* IN sleep_time: delay in seconds (0 means unbounded wait)
* OUT resp: resource allocation response message
* RET 1 if resp is filled in, 0 otherwise */
static int
_wait_for_alloc_rpc(const listen_t *listen, int sleep_time,
resource_allocation_response_msg_t **resp)
{
struct pollfd fds[1];
int rc;
int timeout_ms;
if (listen == NULL) {
error("Listening port not found");
sleep(MAX(sleep_time, 1));
return SLURM_ERROR;
}
fds[0].fd = listen->fd;
fds[0].events = POLLIN;
if (sleep_time != 0) {
timeout_ms = sleep_time * 1000;
} else {
timeout_ms = -1;
}
while ((rc = poll(fds, 1, timeout_ms)) < 0) {
switch (errno) {
case EAGAIN:
case EINTR:
*resp = NULL;
return -1;
case EBADF:
case ENOMEM:
case EINVAL:
case EFAULT:
error("poll: %m");
*resp = NULL;
return -1;
default:
error("poll: %m. Continuing...");
}
}
if (rc == 0) { /* poll timed out */
errno = ETIMEDOUT;
} else if (fds[0].revents & POLLIN) {
return (_accept_msg_connection(listen->fd, resp));
}
return 0;
}
static void _wait_for_allocation_response(uint32_t job_id,
const listen_t *listen,
uint16_t msg_type, int timeout,
void **resp)
{
int errnum, rc;
info("job %u queued and waiting for resources", job_id);
*resp = NULL;
if ((rc = _wait_for_alloc_rpc(listen, timeout)) == 1)
rc = _accept_msg_connection(listen->fd, msg_type, resp);
if (rc <= 0) {
errnum = errno;
/* Maybe the resource allocation response RPC got lost
* in the mail; surely it should have arrived by now.
* Let's see if the controller thinks that the allocation
* has been granted.
*/
if (msg_type == RESPONSE_RESOURCE_ALLOCATION) {
if (slurm_allocation_lookup(job_id,
(resource_allocation_response_msg_t **)
resp) >= 0)
return;
} else if (msg_type == RESPONSE_JOB_PACK_ALLOCATION) {
if (slurm_pack_job_lookup(job_id, (List *) resp) >= 0)
return;
} else {
error("%s: Invalid msg_type (%u)", __func__, msg_type);
}
if (slurm_get_errno() == ESLURM_JOB_PENDING) {
debug3("Still waiting for allocation");
errno = errnum;
return;
} else {
debug3("Unable to confirm allocation for job %u: %m",
job_id);
return;
}
}
info("job %u has been allocated resources", job_id);
return;
}
