/* This function is responsible for building the trees in case of fast
* weighted least-conns. It also sets p->lbprm.wdiv to the eweight to
* uweight ratio. Both active and backup groups are initialized.
*/
void fas_init_server_tree(struct proxy *p)
{
struct server *srv;
struct eb_root init_head = EB_ROOT;
p->lbprm.set_server_status_up = fas_set_server_status_up;
p->lbprm.set_server_status_down = fas_set_server_status_down;
p->lbprm.update_server_eweight = fas_update_server_weight;
p->lbprm.server_take_conn = fas_srv_reposition;
p->lbprm.server_drop_conn = fas_srv_reposition;
p->lbprm.wdiv = BE_WEIGHT_SCALE;
for (srv = p->srv; srv; srv = srv->next) {
srv->eweight = (srv->uweight * p->lbprm.wdiv + p->lbprm.wmult - 1) / p->lbprm.wmult;
srv_lb_commit_status(srv);
}
recount_servers(p);
update_backend_weight(p);
p->lbprm.fas.act = init_head;
p->lbprm.fas.bck = init_head;
/* queue active and backup servers in two distinct groups */
for (srv = p->srv; srv; srv = srv->next) {
if (!srv_is_usable(srv))
continue;
srv->lb_tree = (srv->state & SRV_BACKUP) ? &p->lbprm.fas.bck : &p->lbprm.fas.act;
fas_queue_srv(srv);
}
}
/* This function updates the server trees according to server <srv>'s new
* state. It should be called when server <srv>'s status changes to up.
* It is not important whether the server was already down or not. It is not
* important either that the new state is completely UP (the caller may not
* know all the variables of a server's state). This function will not change
* the weight of a server which was already up.
*/
static void fwrr_set_server_status_up(struct server *srv)
{
struct proxy *p = srv->proxy;
struct fwrr_group *grp;
if (srv->state == srv->prev_state &&
srv->eweight == srv->prev_eweight)
return;
if (!srv_is_usable(srv->state, srv->eweight))
goto out_update_state;
if (srv_is_usable(srv->prev_state, srv->prev_eweight))
/* server was already up */
goto out_update_backend;
grp = (srv->state & SRV_BACKUP) ? &p->lbprm.fwrr.bck : &p->lbprm.fwrr.act;
grp->next_weight += srv->eweight;
if (srv->state & SRV_BACKUP) {
p->lbprm.tot_wbck = p->lbprm.fwrr.bck.next_weight;
p->srv_bck++;
if (!(p->options & PR_O_USE_ALL_BK)) {
if (!p->lbprm.fbck) {
/* there was no backup server anymore */
p->lbprm.fbck = srv;
} else {
/* we may have restored a backup server prior to fbck,
* in which case it should replace it.
*/
struct server *srv2 = srv;
do {
srv2 = srv2->next;
} while (srv2 && (srv2 != p->lbprm.fbck));
if (srv2)
p->lbprm.fbck = srv;
}
}
} else {
p->lbprm.tot_wact = p->lbprm.fwrr.act.next_weight;
p->srv_act++;
}
/* note that eweight cannot be 0 here */
fwrr_get_srv(srv);
srv->npos = grp->curr_pos + (grp->next_weight + grp->curr_weight - grp->curr_pos) / srv->eweight;
fwrr_queue_srv(srv);
out_update_backend:
/* check/update tot_used, tot_weight */
update_backend_weight(p);
out_update_state:
srv->prev_state = srv->state;
srv->prev_eweight = srv->eweight;
}
/* This function updates the server trees according to server <srv>'s new
* state. It should be called when server <srv>'s status changes to up.
* It is not important whether the server was already down or not. It is not
* important either that the new state is completely UP (the caller may not
* know all the variables of a server's state). This function will not change
* the weight of a server which was already up.
*/
static void fas_set_server_status_up(struct server *srv)
{
struct proxy *p = srv->proxy;
if (!srv_lb_status_changed(srv))
return;
if (!srv_is_usable(srv))
goto out_update_state;
if (srv_was_usable(srv))
/* server was already up */
goto out_update_backend;
if (srv->state & SRV_BACKUP) {
srv->lb_tree = &p->lbprm.fas.bck;
p->lbprm.tot_wbck += srv->eweight;
p->srv_bck++;
if (!(p->options & PR_O_USE_ALL_BK)) {
if (!p->lbprm.fbck) {
/* there was no backup server anymore */
p->lbprm.fbck = srv;
} else {
/* we may have restored a backup server prior to fbck,
* in which case it should replace it.
*/
struct server *srv2 = srv;
do {
srv2 = srv2->next;
} while (srv2 && (srv2 != p->lbprm.fbck));
if (srv2)
p->lbprm.fbck = srv;
}
}
} else {
srv->lb_tree = &p->lbprm.fas.act;
p->lbprm.tot_wact += srv->eweight;
p->srv_act++;
}
/* note that eweight cannot be 0 here */
fas_queue_srv(srv);
out_update_backend:
/* check/update tot_used, tot_weight */
update_backend_weight(p);
out_update_state:
srv_lb_commit_status(srv);
}
/* This function updates the server trees according to server <srv>'s new
* state. It should be called when server <srv>'s status changes to down.
* It is not important whether the server was already down or not. It is not
* important either that the new state is completely down (the caller may not
* know all the variables of a server's state).
*/
static void fwrr_set_server_status_down(struct server *srv)
{
struct proxy *p = srv->proxy;
struct fwrr_group *grp;
if (srv->state == srv->prev_state &&
srv->eweight == srv->prev_eweight)
return;
if (srv_is_usable(srv->state, srv->eweight))
goto out_update_state;
if (!srv_is_usable(srv->prev_state, srv->prev_eweight))
/* server was already down */
goto out_update_backend;
grp = (srv->state & SRV_BACKUP) ? &p->lbprm.fwrr.bck : &p->lbprm.fwrr.act;
grp->next_weight -= srv->prev_eweight;
if (srv->state & SRV_BACKUP) {
p->lbprm.tot_wbck = p->lbprm.fwrr.bck.next_weight;
p->srv_bck--;
if (srv == p->lbprm.fbck) {
/* we lost the first backup server in a single-backup
* configuration, we must search another one.
*/
struct server *srv2 = p->lbprm.fbck;
do {
srv2 = srv2->next;
} while (srv2 &&
!((srv2->state & SRV_BACKUP) &&
srv_is_usable(srv2->state, srv2->eweight)));
p->lbprm.fbck = srv2;
}
} else {
p->lbprm.tot_wact = p->lbprm.fwrr.act.next_weight;
p->srv_act--;
}
fwrr_dequeue_srv(srv);
fwrr_remove_from_tree(srv);
out_update_backend:
/* check/update tot_used, tot_weight */
update_backend_weight(p);
out_update_state:
srv->prev_state = srv->state;
srv->prev_eweight = srv->eweight;
}
/* This function must be called after an update to server <srv>'s effective
* weight. It may be called after a state change too.
*/
static void fas_update_server_weight(struct server *srv)
{
int old_state, new_state;
struct proxy *p = srv->proxy;
if (!srv_lb_status_changed(srv))
return;
/* If changing the server's weight changes its state, we simply apply
* the procedures we already have for status change. If the state
* remains down, the server is not in any tree, so it's as easy as
* updating its values. If the state remains up with different weights,
* there are some computations to perform to find a new place and
* possibly a new tree for this server.
*/
old_state = srv_was_usable(srv);
new_state = srv_is_usable(srv);
if (!old_state && !new_state) {
srv_lb_commit_status(srv);
return;
}
else if (!old_state && new_state) {
fas_set_server_status_up(srv);
return;
}
else if (old_state && !new_state) {
fas_set_server_status_down(srv);
return;
}
if (srv->lb_tree)
fas_dequeue_srv(srv);
if (srv->state & SRV_BACKUP) {
p->lbprm.tot_wbck += srv->eweight - srv->prev_eweight;
srv->lb_tree = &p->lbprm.fas.bck;
} else {
p->lbprm.tot_wact += srv->eweight - srv->prev_eweight;
srv->lb_tree = &p->lbprm.fas.act;
}
fas_queue_srv(srv);
update_backend_weight(p);
srv_lb_commit_status(srv);
}
/* This function updates the server trees according to server <srv>'s new
* state. It should be called when server <srv>'s status changes to down.
* It is not important whether the server was already down or not. It is not
* important either that the new state is completely down (the caller may not
* know all the variables of a server's state).
*/
static void fas_set_server_status_down(struct server *srv)
{
struct proxy *p = srv->proxy;
if (!srv_lb_status_changed(srv))
return;
if (srv_is_usable(srv))
goto out_update_state;
if (!srv_was_usable(srv))
/* server was already down */
goto out_update_backend;
if (srv->state & SRV_BACKUP) {
p->lbprm.tot_wbck -= srv->prev_eweight;
p->srv_bck--;
if (srv == p->lbprm.fbck) {
/* we lost the first backup server in a single-backup
* configuration, we must search another one.
*/
struct server *srv2 = p->lbprm.fbck;
do {
srv2 = srv2->next;
} while (srv2 &&
!((srv2->state & SRV_BACKUP) &&
srv_is_usable(srv2)));
p->lbprm.fbck = srv2;
}
} else {
p->lbprm.tot_wact -= srv->prev_eweight;
p->srv_act--;
}
fas_dequeue_srv(srv);
fas_remove_from_tree(srv);
out_update_backend:
/* check/update tot_used, tot_weight */
update_backend_weight(p);
out_update_state:
srv_lb_commit_status(srv);
}
/* This function is responsible for building the weight trees in case of fast
* weighted round-robin. It also sets p->lbprm.wdiv to the eweight to uweight
* ratio. Both active and backup groups are initialized.
*/
void fwrr_init_server_groups(struct proxy *p)
{
struct server *srv;
struct eb_root init_head = EB_ROOT;
p->lbprm.set_server_status_up = fwrr_set_server_status_up;
p->lbprm.set_server_status_down = fwrr_set_server_status_down;
p->lbprm.update_server_eweight = fwrr_update_server_weight;
p->lbprm.wdiv = BE_WEIGHT_SCALE;
for (srv = p->srv; srv; srv = srv->next) {
srv->prev_eweight = srv->eweight = srv->uweight * BE_WEIGHT_SCALE;
srv->prev_state = srv->state;
}
recount_servers(p);
update_backend_weight(p);
/* prepare the active servers group */
p->lbprm.fwrr.act.curr_pos = p->lbprm.fwrr.act.curr_weight =
p->lbprm.fwrr.act.next_weight = p->lbprm.tot_wact;
p->lbprm.fwrr.act.curr = p->lbprm.fwrr.act.t0 =
p->lbprm.fwrr.act.t1 = init_head;
p->lbprm.fwrr.act.init = &p->lbprm.fwrr.act.t0;
p->lbprm.fwrr.act.next = &p->lbprm.fwrr.act.t1;
/* prepare the backup servers group */
p->lbprm.fwrr.bck.curr_pos = p->lbprm.fwrr.bck.curr_weight =
p->lbprm.fwrr.bck.next_weight = p->lbprm.tot_wbck;
p->lbprm.fwrr.bck.curr = p->lbprm.fwrr.bck.t0 =
p->lbprm.fwrr.bck.t1 = init_head;
p->lbprm.fwrr.bck.init = &p->lbprm.fwrr.bck.t0;
p->lbprm.fwrr.bck.next = &p->lbprm.fwrr.bck.t1;
/* queue active and backup servers in two distinct groups */
for (srv = p->srv; srv; srv = srv->next) {
if (!srv_is_usable(srv->state, srv->eweight))
continue;
fwrr_queue_by_weight((srv->state & SRV_BACKUP) ?
p->lbprm.fwrr.bck.init :
p->lbprm.fwrr.act.init,
srv);
}
}
/* This function updates the map according to server <srv>'s new state */
static void map_set_server_status_up(struct server *srv)
{
struct proxy *p = srv->proxy;
if (srv->state == srv->prev_state &&
srv->eweight == srv->prev_eweight)
return;
if (!srv_is_usable(srv->state, srv->eweight))
goto out_update_state;
/* FIXME: could be optimized since we know what changed */
recount_servers(p);
update_backend_weight(p);
p->lbprm.map.state |= LB_MAP_RECALC;
out_update_state:
srv->prev_state = srv->state;
srv->prev_eweight = srv->eweight;
}
/* This function is responsible of building the server MAP for map-based LB
* algorithms, allocating the map, and setting p->lbprm.wmult to the GCD of the
* weights if applicable. It should be called only once per proxy, at config
* time.
*/
void init_server_map(struct proxy *p)
{
struct server *srv;
int pgcd;
int act, bck;
p->lbprm.set_server_status_up = map_set_server_status_up;
p->lbprm.set_server_status_down = map_set_server_status_down;
p->lbprm.update_server_eweight = NULL;
if (!p->srv)
return;
/* We will factor the weights to reduce the table,
* using Euclide's largest common divisor algorithm.
* Since we may have zero weights, we have to first
* find a non-zero weight server.
*/
pgcd = 1;
srv = p->srv;
while (srv && !srv->uweight)
srv = srv->next;
if (srv) {
pgcd = srv->uweight; /* note: cannot be zero */
while (pgcd > 1 && (srv = srv->next)) {
int w = srv->uweight;
while (w) {
int t = pgcd % w;
pgcd = w;
w = t;
}
}
}
/* It is sometimes useful to know what factor to apply
* to the backend's effective weight to know its real
* weight.
*/
p->lbprm.wmult = pgcd;
act = bck = 0;
for (srv = p->srv; srv; srv = srv->next) {
srv->eweight = srv->uweight / pgcd;
srv->prev_eweight = srv->eweight;
srv->prev_state = srv->state;
if (srv->state & SRV_BACKUP)
bck += srv->eweight;
else
act += srv->eweight;
}
/* this is the largest map we will ever need for this servers list */
if (act < bck)
act = bck;
if (!act)
act = 1;
p->lbprm.map.srv = (struct server **)calloc(act, sizeof(struct server *));
/* recounts servers and their weights */
p->lbprm.map.state = LB_MAP_RECALC;
recount_servers(p);
update_backend_weight(p);
recalc_server_map(p);
}
/* This function must be called after an update to server <srv>'s effective
* weight. It may be called after a state change too.
*/
static void fwrr_update_server_weight(struct server *srv)
{
int old_state, new_state;
struct proxy *p = srv->proxy;
struct fwrr_group *grp;
if (srv->state == srv->prev_state &&
srv->eweight == srv->prev_eweight)
return;
/* If changing the server's weight changes its state, we simply apply
* the procedures we already have for status change. If the state
* remains down, the server is not in any tree, so it's as easy as
* updating its values. If the state remains up with different weights,
* there are some computations to perform to find a new place and
* possibly a new tree for this server.
*/
old_state = srv_is_usable(srv->prev_state, srv->prev_eweight);
new_state = srv_is_usable(srv->state, srv->eweight);
if (!old_state && !new_state) {
srv->prev_state = srv->state;
srv->prev_eweight = srv->eweight;
return;
}
else if (!old_state && new_state) {
fwrr_set_server_status_up(srv);
return;
}
else if (old_state && !new_state) {
fwrr_set_server_status_down(srv);
return;
}
grp = (srv->state & SRV_BACKUP) ? &p->lbprm.fwrr.bck : &p->lbprm.fwrr.act;
grp->next_weight = grp->next_weight - srv->prev_eweight + srv->eweight;
p->lbprm.tot_wact = p->lbprm.fwrr.act.next_weight;
p->lbprm.tot_wbck = p->lbprm.fwrr.bck.next_weight;
if (srv->lb_tree == grp->init) {
fwrr_dequeue_srv(srv);
fwrr_queue_by_weight(grp->init, srv);
}
else if (!srv->lb_tree) {
/* FIXME: server was down. This is not possible right now but
* may be needed soon for slowstart or graceful shutdown.
*/
fwrr_dequeue_srv(srv);
fwrr_get_srv(srv);
srv->npos = grp->curr_pos + (grp->next_weight + grp->curr_weight - grp->curr_pos) / srv->eweight;
fwrr_queue_srv(srv);
} else {
/* The server is either active or in the next queue. If it's
* still in the active queue and it has not consumed all of its
* places, let's adjust its next position.
*/
fwrr_get_srv(srv);
if (srv->eweight > 0) {
int prev_next = srv->npos;
int step = grp->next_weight / srv->eweight;
srv->npos = srv->lpos + step;
srv->rweight = 0;
if (srv->npos > prev_next)
srv->npos = prev_next;
if (srv->npos < grp->curr_pos + 2)
srv->npos = grp->curr_pos + step;
} else {
/* push it into the next tree */
srv->npos = grp->curr_pos + grp->curr_weight;
}
fwrr_dequeue_srv(srv);
fwrr_queue_srv(srv);
}
update_backend_weight(p);
srv->prev_state = srv->state;
srv->prev_eweight = srv->eweight;
}
