void pmixp_dmdx_timeout_cleanup(void)
{
ListIterator it = list_iterator_create(_dmdx_requests);
dmdx_req_info_t *req = NULL;
time_t ts = time(NULL);
/* run through all requests and discard stale one's */
while (NULL != (req = list_next(it))) {
if ((ts - req->ts) > pmixp_info_timeout()) {
#ifndef NDEBUG
/* respond with the timeout to libpmix */
char *host = pmixp_nspace_resolve(req->nspace,
req->rank);
xassert(NULL != host);
PMIXP_ERROR("timeout: ns=%s, rank=%d," " host=%s, ts=%lu",
req->nspace, req->rank,
(NULL != host) ? host : "unknown", ts);
if (NULL != host) {
free(host);
}
#endif
req->cbfunc(PMIX_ERR_TIMEOUT, NULL, 0, req->cbdata,
NULL, NULL);
/* release tracker & list iterator */
list_delete_item(it);
}
}
list_iterator_destroy(it);
}
void pmixp_coll_ring_reset_if_to(pmixp_coll_t *coll, time_t ts) {
pmixp_coll_ring_ctx_t *coll_ctx;
int i;
/* lock the structure */
slurm_mutex_lock(&coll->lock);
for (i = 0; i < PMIXP_COLL_RING_CTX_NUM; i++) {
coll_ctx = &coll->state.ring.ctx_array[i];
if (!coll_ctx->in_use ||
(PMIXP_COLL_RING_SYNC == coll_ctx->state)) {
continue;
}
if (ts - coll->ts > pmixp_info_timeout()) {
/* respond to the libpmix */
if (coll_ctx->contrib_local && coll->cbfunc) {
pmixp_lib_modex_invoke(coll->cbfunc,
PMIXP_ERR_TIMEOUT, NULL,
0, coll->cbdata,
NULL, NULL);
}
/* report the timeout event */
PMIXP_ERROR("%p: collective timeout seq=%d",
coll, coll_ctx->seq);
pmixp_coll_log(coll);
/* drop the collective */
_reset_coll_ring(coll_ctx);
}
}
/* unlock the structure */
slurm_mutex_unlock(&coll->lock);
}
void pmixp_coll_reset_if_to(pmixp_coll_t *coll, time_t ts)
{
/* lock the */
slurm_mutex_lock(&coll->lock);
if (PMIXP_COLL_SYNC == coll->state) {
goto unlock;
}
if (ts - coll->ts > pmixp_info_timeout()) {
/* respond to the libpmix */
if (coll->contrib_local && coll->cbfunc) {
/* Call the callback only if:
* - we were asked to do that (coll->cbfunc != NULL);
* - local contribution was received.
* TODO: we may want to mark this event to respond with
* to the next local request immediately and with the
* proper (status == PMIX_ERR_TIMEOUT)
*/
coll->cbfunc(PMIX_ERR_TIMEOUT, NULL, 0, coll->cbdata,
NULL, NULL);
}
/* drop the collective */
_reset_coll(coll);
/* report the timeout event */
PMIXP_ERROR("Collective timeout!");
}
unlock:
/* unlock the structure */
slurm_mutex_unlock(&coll->lock);
}
void pmixp_coll_reset_if_to(pmixp_coll_t *coll, time_t ts)
{
/* lock the */
slurm_mutex_lock(&coll->lock);
if (PMIXP_COLL_SYNC == coll->state) {
goto unlock;
}
if (ts - coll->ts > pmixp_info_timeout()) {
/* respond to the libpmix */
coll->cbfunc(PMIX_ERR_TIMEOUT, NULL, 0, coll->cbdata, NULL,
NULL);
/* drop the collective */
_reset_coll(coll);
/* report the timeout event */
PMIXP_ERROR("Collective timeout!");
}
unlock:
/* unlock the structure */
slurm_mutex_unlock(&coll->lock);
}
