void _add_hportal_op(host_portal_t *hp, op_generic_t *hsop, int addtotop, int release_master)
{
command_op_t *hop = &(hsop->op->cmd);
tbx_stack_ele_t *ele;
hp->workload = hp->workload + hop->workload;
if (addtotop == 1) {
tbx_stack_push(hp->que, (void *)hsop);
} else {
tbx_stack_move_to_bottom(hp->que);
tbx_stack_insert_below(hp->que, (void *)hsop);
};
//** Since we've now added the op to the hp que we can release the master lock if needed
//** without fear of having the compact_hportals() coming in and destroying the hp
if (release_master == 1) apr_thread_mutex_unlock(hp->context->lock);
//** Check if we need a little pre-processing
if (hop->on_submit != NULL) {
ele = tbx_stack_get_current_ptr(hp->que);
hop->on_submit(hp->que, ele);
}
hportal_signal(hp); //** Send a signal for any tasks listening
}
void _tp_submit_op(void *arg, gop_op_generic_t *gop)
{
gop_thread_pool_op_t *op = gop_get_tp(gop);
apr_status_t aerr;
int running;
log_printf(15, "_tp_submit_op: gid=%d\n", gop_id(gop));
tbx_atomic_inc(op->tpc->n_submitted);
op->via_submit = 1;
running = tbx_atomic_inc(op->tpc->n_running) + 1;
if (running > op->tpc->max_concurrency) {
apr_thread_mutex_lock(_tp_lock);
tbx_atomic_inc(op->tpc->n_overflow);
if (op->depth >= op->tpc->recursion_depth) { //** Check if we hit the max recursion
log_printf(0, "GOP has a recursion depth >= max specified in the TP!!!! gop depth=%d TPC max=%d\n", op->depth, op->tpc->recursion_depth);
tbx_stack_push(op->tpc->reserve_stack[op->tpc->recursion_depth-1], gop); //** Need to do the push and overflow check
} else {
tbx_stack_push(op->tpc->reserve_stack[op->depth], gop); //** Need to do the push and overflow check
}
gop = _tpc_overflow_next(op->tpc); //** along with the submit or rollback atomically
if (gop) {
op = gop_get_tp(gop);
aerr = apr_thread_pool_push(op->tpc->tp,(void *(*)(apr_thread_t *, void *))thread_pool_exec_fn, gop, APR_THREAD_TASK_PRIORITY_NORMAL, NULL);
} else {
tbx_atomic_dec(op->tpc->n_running); //** We didn't actually submit anything
if (op->overflow_slot != -1) { //** Check if we need to undo our overflow slot
op->tpc->overflow_running_depth[op->overflow_slot] = -1;
}
aerr = APR_SUCCESS;
}
apr_thread_mutex_unlock(_tp_lock);
} else {
aerr = apr_thread_pool_push(op->tpc->tp, (void *(*)(apr_thread_t *, void *))thread_pool_exec_fn, gop, APR_THREAD_TASK_PRIORITY_NORMAL, NULL);
}
if (aerr != APR_SUCCESS) {
log_printf(0, "ERROR submiting task! aerr=%d gid=%d\n", aerr, gop_id(gop));
}
}
void rsrs_update_register(resource_service_fn_t *rs, mq_frame_t *fid, mq_msg_t *address, int timeout)
{
rs_remote_server_priv_t *rsrs = (rs_remote_server_priv_t *)rs->priv;
rsrs_update_handle_t *h;
tbx_type_malloc(h, rsrs_update_handle_t, 1);
//** Form the core message
h->msg = mq_msg_new();
mq_msg_append_mem(h->msg, MQF_VERSION_KEY, MQF_VERSION_SIZE, MQF_MSG_KEEP_DATA);
mq_msg_append_mem(h->msg, MQF_RESPONSE_KEY, MQF_RESPONSE_SIZE, MQF_MSG_KEEP_DATA);
mq_msg_append_frame(h->msg, fid);
mq_get_frame(fid, (void **)&(h->id), &(h->id_size));
//** Add the empty version frame and track it for filling in later
h->version_frame = mq_frame_new(NULL, 0, MQF_MSG_AUTO_FREE);
mq_msg_append_frame(h->msg, h->version_frame);
//** Add the empty config frame and track it for filling in later
h->config_frame = mq_frame_new(NULL, 0, MQF_MSG_AUTO_FREE);
mq_msg_append_frame(h->msg, h->config_frame);
//** End with an empty frame
mq_msg_append_mem(h->msg, NULL, 0, MQF_MSG_KEEP_DATA);
//** Now address it
mq_apply_return_address_msg(h->msg, address, 0);
//** Figure out when we wake up if no change
if (timeout > 10) {
h->reply_time = apr_time_from_sec(timeout-10);
} else if (timeout > 5) {
h->reply_time = apr_time_from_sec(timeout-5);
} else {
h->reply_time = apr_time_from_sec(1);
}
h->reply_time += apr_time_now();
apr_thread_mutex_lock(rsrs->lock);
//** Add it to the queue
tbx_stack_push(rsrs->pending, h);
//** Check if we need to change when we wake up
if ((h->reply_time < rsrs->wakeup_time) || (rsrs->wakeup_time == 0)) rsrs->wakeup_time = h->reply_time;
log_printf(5, "timeout=%d now=" TT " reply_time=" TT " wakeup=" TT "\n", timeout, apr_time_now(), h->reply_time, rsrs->wakeup_time);
apr_thread_mutex_unlock(rsrs->lock);
}
int data_block_set_attr(lio_data_block_t *b, char *key, char *val)
{
lio_data_block_attr_t *attr;
//** See if the key exists
attr = db_find_key(b->attr_stack, key);
if (attr == NULL) { //** See if we need to add the attribute
tbx_type_malloc_clear(attr, lio_data_block_attr_t, 1);
attr->key = strdup(key);
}
if (attr->value != NULL) free(attr->value); //** Free the old value
attr->value = (val != NULL) ? strdup(val) : NULL; //** Store the new one
if (b->attr_stack == NULL) b->attr_stack = tbx_stack_new();
tbx_stack_push(b->attr_stack, attr);
return(0);
}
op_generic_t *rs_simple_request(resource_service_fn_t *arg, data_attr_t *da, rs_query_t *rsq, data_cap_set_t **caps, rs_request_t *req, int req_size, rs_hints_t *hints_list, int fixed_size, int n_rid, int ignore_fixed_err, int timeout)
{
rs_simple_priv_t *rss = (rs_simple_priv_t *)arg->priv;
rsq_base_t *query_global = (rsq_base_t *)rsq;
rsq_base_t *query_local;
kvq_table_t kvq_global, kvq_local, *kvq;
apr_hash_t *pick_from;
rid_change_entry_t *rid_change;
ex_off_t change;
op_status_t status;
opque_t *que;
rss_rid_entry_t *rse;
rsq_base_ele_t *q;
int slot, rnd_off, i, j, k, i_unique, i_pickone, found, err_cnt, loop, loop_end;
int state, *a, *b, *op_state, unique_size;
tbx_stack_t *stack;
log_printf(15, "rs_simple_request: START rss->n_rids=%d n_rid=%d req_size=%d fixed_size=%d\n", rss->n_rids, n_rid, req_size, fixed_size);
for (i=0; i<req_size; i++) req[i].rid_key = NULL; //** Clear the result in case of an error
apr_thread_mutex_lock(rss->lock);
i = _rs_simple_refresh(arg); //** Check if we need to refresh the data
if (i != 0) {
apr_thread_mutex_unlock(rss->lock);
return(gop_dummy(op_failure_status));
}
//** Determine the query sizes and make the processing arrays
memset(&kvq, 0, sizeof(kvq));
rs_query_count(arg, rsq, &i, &(kvq_global.n_unique), &(kvq_global.n_pickone));
log_printf(15, "rs_simple_request: n_unique=%d n_pickone=%d\n", kvq_global.n_unique, kvq_global.n_pickone);
tbx_log_flush();
//** Make space the for the uniq and pickone fields.
//** Make sure we have space for at least 1 more than we need of each to pass to the routines even though they aren't used
j = (kvq_global.n_pickone == 0) ? 1 : kvq_global.n_pickone + 1;
tbx_type_malloc_clear(kvq_global.pickone, kvq_ele_t, j);
unique_size = kvq_global.n_unique + 1;
tbx_type_malloc_clear(kvq_global.unique, kvq_ele_t *, unique_size);
log_printf(15, "MALLOC j=%d\n", unique_size);
for (i=0; i<unique_size; i++) {
tbx_type_malloc_clear(kvq_global.unique[i], kvq_ele_t, n_rid);
}
//** We don't allow these on the local but make a temp space anyway
kvq_local.n_pickone = 0;
tbx_type_malloc_clear(kvq_local.pickone, kvq_ele_t, 1);
kvq_global.n_unique = 0;
tbx_type_malloc_clear(kvq_local.unique, kvq_ele_t *, 1);
tbx_type_malloc_clear(kvq_local.unique[0], kvq_ele_t, n_rid);
status = op_success_status;
que = new_opque();
stack = tbx_stack_new();
err_cnt = 0;
found = 0;
// max_size = (req_size > fixed_size) ? req_size : fixed_size;
for (i=0; i < n_rid; i++) {
found = 0;
loop_end = 1;
query_local = NULL;
rnd_off = tbx_random_get_int64(0, rss->n_rids-1);
//rnd_off = 0; //FIXME
if (hints_list != NULL) {
query_local = (rsq_base_t *)hints_list[i].local_rsq;
if (query_local != NULL) {
loop_end = 2;
rs_query_count(arg, query_local, &j, &(kvq_local.n_unique), &(kvq_local.n_pickone));
if ((kvq_local.n_unique != 0) && (kvq_local.n_pickone != 0)) {
log_printf(0, "Unsupported use of pickone/unique in local RSQ hints_list[%d]=%s!\n", i, hints_list[i].fixed_rid_key);
status.op_status = OP_STATE_FAILURE;
status.error_code = RS_ERROR_FIXED_NOT_FOUND;
hints_list[i].status = RS_ERROR_HINTS_INVALID_LOCAL;
err_cnt++;
continue;
}
}
if (i<fixed_size) { //** Use the fixed list for assignment
rse = tbx_list_search(rss->rid_table, hints_list[i].fixed_rid_key);
if (rse == NULL) {
log_printf(0, "Missing element in hints list[%d]=%s! Ignoring check.\n", i, hints_list[i].fixed_rid_key);
hints_list[i].status = RS_ERROR_FIXED_NOT_FOUND;
continue; //** Skip the check
}
rnd_off = rse->slot;
}
}
//** See if we use a restrictive list. Ususally used when rebalancing space
pick_from = (hints_list != NULL) ? hints_list[i].pick_from : NULL;
rid_change = NULL;
change = 0;
for (k=0; k<req_size; k++) {
if (req[k].rid_index == i) {
change += req[k].size;
}
}
for (j=0; j<rss->n_rids; j++) {
slot = (rnd_off+j) % rss->n_rids;
rse = rss->random_array[slot];
if (pick_from != NULL) {
rid_change = apr_hash_get(pick_from, rse->rid_key, APR_HASH_KEY_STRING);
log_printf(15, "PICK_FROM != NULL i=%d j=%d slot=%d rse->rid_key=%s rse->status=%d rid_change=%p\n", i, j, slot, rse->rid_key, rse->status, rid_change);
if (rid_change == NULL) continue; //** Not in our list so skip to the next
ex_off_t delta = rid_change->delta - change;
log_printf(15, "PICK_FROM != NULL i=%d j=%d slot=%d rse->rid_key=%s rse->status=%d rc->state=%d (" XOT ") > " XOT "????\n", i, j, slot, rse->rid_key, rse->status, rid_change->state, delta, rid_change->tolerance);
//** Make sure we don't overshoot the target
if (rid_change->state == 1) continue; //** Already converged RID
if (rid_change->delta <= 0) continue; //** Need to move data OFF this RID
if ((change - rid_change->delta) > rid_change->tolerance) continue; //**delta>0 if we made it here
}
log_printf(15, "i=%d j=%d slot=%d rse->rid_key=%s rse->status=%d\n", i, j, slot, rse->rid_key, rse->status);
if ((rse->status != RS_STATUS_UP) && (i>=fixed_size)) continue; //** Skip this if disabled and not in the fixed list
tbx_stack_empty(stack, 1);
q = query_global->head;
kvq = &kvq_global;
for (loop=0; loop<loop_end; loop++) {
i_unique = 0;
i_pickone = 0;
while (q != NULL) {
state = -1;
switch (q->op) {
case RSQ_BASE_OP_KV:
state = rss_test(q, rse, i, kvq->unique[i_unique], &(kvq->pickone[i_pickone]));
log_printf(0, "KV: key=%s val=%s i_unique=%d i_pickone=%d loop=%d rss_test=%d rse->rid_key=%s\n", q->key, q->val, i_unique, i_pickone, loop, state, rse->rid_key);
tbx_log_flush();
if ((q->key_op & RSQ_BASE_KV_UNIQUE) || (q->val_op & RSQ_BASE_KV_UNIQUE)) i_unique++;
if ((q->key_op & RSQ_BASE_KV_PICKONE) || (q->val_op & RSQ_BASE_KV_PICKONE)) i_pickone++;
break;
case RSQ_BASE_OP_NOT:
a = (int *)tbx_stack_pop(stack);
state = (*a == 0) ? 1 : 0;
//log_printf(0, "NOT(%d)=%d\n", *a, state);
free(a);
break;
case RSQ_BASE_OP_AND:
a = (int *)tbx_stack_pop(stack);
b = (int *)tbx_stack_pop(stack);
state = (*a) && (*b);
//log_printf(0, "%d AND %d = %d\n", *a, *b, state);
free(a);
free(b);
break;
case RSQ_BASE_OP_OR:
a = (int *)tbx_stack_pop(stack);
b = (int *)tbx_stack_pop(stack);
state = a || b;
//log_printf(0, "%d OR %d = %d\n", *a, *b, state);
free(a);
free(b);
break;
}
tbx_type_malloc(op_state, int, 1);
*op_state = state;
tbx_stack_push(stack, (void *)op_state);
log_printf(15, " stack_size=%d loop=%d push state=%d\n",tbx_stack_count(stack), loop, state);
tbx_log_flush();
q = q->next;
}
if (query_local != NULL) {
q = query_local->head;
kvq = &kvq_local;
}
}
op_state = (int *)tbx_stack_pop(stack);
state = -1;
if (op_state != NULL) {
state = *op_state;
free(op_state);
}
if (op_state == NULL) {
log_printf(1, "rs_simple_request: ERROR processing i=%d EMPTY STACK\n", i);
found = 0;
status.op_status = OP_STATE_FAILURE;
status.error_code = RS_ERROR_EMPTY_STACK;
} else if (state == 1) { //** Got one
log_printf(15, "rs_simple_request: processing i=%d ds_key=%s\n", i, rse->ds_key);
found = 1;
if ((i<fixed_size) && hints_list) hints_list[i].status = RS_ERROR_OK;
for (k=0; k<req_size; k++) {
if (req[k].rid_index == i) {
log_printf(15, "rs_simple_request: i=%d ds_key=%s, rid_key=%s size=" XOT "\n", i, rse->ds_key, rse->rid_key, req[k].size);
req[k].rid_key = strdup(rse->rid_key);
req[k].gop = ds_allocate(rss->ds, rse->ds_key, da, req[k].size, caps[k], timeout);
opque_add(que, req[k].gop);
}
}
if (rid_change != NULL) { //** Flag that I'm tweaking things. The caller does the source pending/delta half
rid_change->delta -= change;
rid_change->state = ((llabs(rid_change->delta) <= rid_change->tolerance) || (rid_change->tolerance == 0)) ? 1 : 0;
}
break; //** Got one so exit the RID scan and start the next one
} else if (i<fixed_size) { //** This should have worked so flag an error
if (hints_list) {
log_printf(1, "Match fail in fixed list[%d]=%s!\n", i, hints_list[i].fixed_rid_key);
hints_list[i].status = RS_ERROR_FIXED_MATCH_FAIL;
} else {
log_printf(1, "Match fail in fixed list and no hints are provided!\n");
}
status.op_status = OP_STATE_FAILURE;
status.error_code = RS_ERROR_FIXED_MATCH_FAIL;
if (ignore_fixed_err == 0) err_cnt++;
break; //** Skip to the next in the list
} else {
found = 0;
}
}
if ((found == 0) && (i>=fixed_size)) break;
}
//** Clean up
log_printf(15, "FREE j=%d\n", unique_size);
for (i=0; i<unique_size; i++) {
free(kvq_global.unique[i]);
}
free(kvq_global.unique);
free(kvq_global.pickone);
free(kvq_local.unique[0]);
free(kvq_local.unique);
free(kvq_local.pickone);
tbx_stack_free(stack, 1);
log_printf(15, "rs_simple_request: END n_rid=%d\n", n_rid);
//callback_t *cb = (callback_t *)que->qd.list->top->data;
//op_generic_t *gop = (op_generic_t *)cb->priv;
//log_printf(15, "top gid=%d reg=%d\n", gop_id(gop), gop_id(req[0].gop));
apr_thread_mutex_unlock(rss->lock);
if ((found == 0) || (err_cnt>0)) {
opque_free(que, OP_DESTROY);
if (status.error_code == 0) {
log_printf(1, "rs_simple_request: Can't find enough RIDs! requested=%d found=%d err_cnt=%d\n", n_rid, found, err_cnt);
status.op_status = OP_STATE_FAILURE;
status.error_code = RS_ERROR_NOT_ENOUGH_RIDS;
}
return(gop_dummy(status));
}
return(opque_get_gop(que));
}
lio_data_block_t *data_block_deserialize_text(lio_service_manager_t *sm, ex_id_t id, lio_exnode_exchange_t *exp)
{
int bufsize=1024;
char capgrp[bufsize];
char *text, *etext;
int i;
lio_data_block_t *b;
lio_data_service_fn_t *ds;
tbx_inip_file_t *cfd;
tbx_inip_group_t *cg;
tbx_inip_element_t *ele;
char *key;
lio_data_block_attr_t *attr;
//** Parse the ini text
cfd = exp->text.fd;
//** Find the cooresponding cap
snprintf(capgrp, bufsize, "block-" XIDT, id);
cg = tbx_inip_group_find(cfd, capgrp);
if (cg == NULL) {
log_printf(0, "data_block_deserialize_text: id=" XIDT " not found!\n", id);
return(NULL);
}
//** Determine the type and make a blank block
text = tbx_inip_get_string(cfd, capgrp, "type", "");
ds = lio_lookup_service(sm, DS_SM_RUNNING, text);
if (ds == NULL) {
log_printf(0, "data_block_deserialize_text: b->id=" XIDT " Unknown data service tpye=%s!\n", id, text);
return(NULL);;
}
free(text);
//** Make the space
b = data_block_create_with_id(ds, id);
//** and parse the fields
b->rid_key = tbx_inip_get_string(cfd, capgrp, "rid_key", "");
b->size = tbx_inip_get_integer(cfd, capgrp, "size", b->size);
b->max_size = tbx_inip_get_integer(cfd, capgrp, "max_size", b->max_size);
i = tbx_inip_get_integer(cfd, capgrp, "ref_count", b->ref_count);
tbx_atomic_set(b->ref_count, 0);
tbx_atomic_set(b->initial_ref_count, i);
etext = tbx_inip_get_string(cfd, capgrp, "read_cap", "");
ds_set_cap(b->ds, b->cap, DS_CAP_READ, tbx_stk_unescape_text('\\', etext));
free(etext);
etext = tbx_inip_get_string(cfd, capgrp, "write_cap", "");
ds_set_cap(b->ds, b->cap, DS_CAP_WRITE, tbx_stk_unescape_text('\\', etext));
free(etext);
etext = tbx_inip_get_string(cfd, capgrp, "manage_cap", "");
ds_set_cap(b->ds, b->cap, DS_CAP_MANAGE, tbx_stk_unescape_text('\\', etext));
free(etext);
//** Now cycle through any misc attributes set
ele = tbx_inip_ele_first(tbx_inip_group_find(cfd, capgrp));
while (ele != NULL) {
key = tbx_inip_ele_get_key(ele);
//** Ignore the builtin commands
if ((strcmp("rid_key", key) != 0) && (strcmp("size", key) != 0) && (strcmp("max_size", key) != 0) && (strcmp("type", key) != 0) &&
(strcmp("ref_count", key) != 0) && (strcmp("read_cap", key) != 0) && (strcmp("write_cap", key) != 0) && (strcmp("manage_cap", key) != 0)) {
tbx_type_malloc(attr, lio_data_block_attr_t, 1);
attr->key = tbx_stk_unescape_text('\\', tbx_inip_ele_get_key(ele));
attr->value = tbx_stk_unescape_text('\\', tbx_inip_ele_get_value(ele));
if (b->attr_stack == NULL) b->attr_stack = tbx_stack_new();
tbx_stack_push(b->attr_stack, attr);
}
ele = tbx_inip_ele_next(ele);
}
return(b);
}
int submit_hp_direct_op(portal_context_t *hpc, op_generic_t *op)
{
int status;
host_portal_t *hp, *shp;
host_connection_t *hc;
command_op_t *hop = &(op->op->cmd);
apr_thread_mutex_lock(hpc->lock);
//** Check if we should do a garbage run **
if (hpc->next_check < time(NULL)) {
hpc->next_check = time(NULL) + hpc->compact_interval;
apr_thread_mutex_unlock(hpc->lock);
compact_hportals(hpc);
apr_thread_mutex_lock(hpc->lock);
}
//** Find it in the list or make a new one
hp = _lookup_hportal(hpc, hop->hostport);
if (hp == NULL) {
log_printf(15, "submit_hp_direct_op: New host: %s\n", hop->hostport);
hp = create_hportal(hpc, hop->connect_context, hop->hostport, 1, 1, apr_time_from_sec(1));
if (hp == NULL) {
log_printf(15, "submit_hp_direct_op: create_hportal failed!\n");
apr_thread_mutex_unlock(hpc->lock);
return(-1);
}
apr_hash_set(hpc->table, hp->skey, APR_HASH_KEY_STRING, (const void *)hp);
}
apr_thread_mutex_unlock(hpc->lock);
log_printf(15, "submit_hp_direct_op: start opid=%d\n", op->base.id);
//** Scan the direct list for a free connection
hportal_lock(hp);
tbx_stack_move_to_top(hp->direct_list);
while ((shp = (host_portal_t *)tbx_stack_get_current_data(hp->direct_list)) != NULL) {
if (hportal_trylock(shp) == 0) {
log_printf(15, "submit_hp_direct_op: opid=%d shp->wl=" I64T " stack_size=%d\n", op->base.id, shp->workload, tbx_stack_count(shp->que));
if (tbx_stack_count(shp->que) == 0) {
if (tbx_stack_count(shp->conn_list) > 0) {
tbx_stack_move_to_top(shp->conn_list);
hc = (host_connection_t *)tbx_stack_get_current_data(shp->conn_list);
if (trylock_hc(hc) == 0) {
if ((tbx_stack_count(hc->pending_stack) == 0) && (hc->curr_workload == 0)) {
log_printf(15, "submit_hp_direct_op(A): before submit ns=%d opid=%d wl=%d\n",tbx_ns_getid(hc->ns), op->base.id, hc->curr_workload);
unlock_hc(hc);
hportal_unlock(shp);
status = submit_hportal(shp, op, 1, 0);
log_printf(15, "submit_hp_direct_op(A): after submit ns=%d opid=%d\n",tbx_ns_getid(hc->ns), op->base.id);
hportal_unlock(hp);
return(status);
}
unlock_hc(hc);
}
} else {
hportal_unlock(shp);
log_printf(15, "submit_hp_direct_op(B): opid=%d\n", op->base.id);
status = submit_hportal(shp, op, 1, 0);
hportal_unlock(hp);
return(status);
}
}
hportal_unlock(shp);
}
tbx_stack_move_down(hp->direct_list); //** Move to the next hp in the list
}
//** If I made it here I have to add a new hportal
shp = create_hportal(hpc, hop->connect_context, hop->hostport, 1, 1, apr_time_from_sec(1));
if (shp == NULL) {
log_printf(15, "submit_hp_direct_op: create_hportal failed!\n");
hportal_unlock(hp);
return(-1);
}
tbx_stack_push(hp->direct_list, (void *)shp);
status = submit_hportal(shp, op, 1, 0);
hportal_unlock(hp);
return(status);
}
