/* 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 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 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 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 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 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 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 recounts the number of usable active and backup servers for
* proxy <p>. These numbers are returned into the p->srv_act and p->srv_bck.
* This function also recomputes the total active and backup weights. However,
* it does not update tot_weight nor tot_used. Use update_backend_weight() for
* this.
*/
void recount_servers(struct proxy *px)
{
struct server *srv;
px->srv_act = px->srv_bck = 0;
px->lbprm.tot_wact = px->lbprm.tot_wbck = 0;
px->lbprm.fbck = NULL;
for (srv = px->srv; srv != NULL; srv = srv->next) {
if (!srv_is_usable(srv))
continue;
if (srv->flags & SRV_F_BACKUP) {
if (!px->srv_bck &&
!(px->options & PR_O_USE_ALL_BK))
px->lbprm.fbck = srv;
px->srv_bck++;
px->lbprm.tot_wbck += srv->eweight;
} else {
px->srv_act++;
px->lbprm.tot_wact += srv->eweight;
}
}
}
/* queues a server into the appropriate group and tree depending on its
* backup status, and ->npos. If the server is disabled, simply assign
* it to the NULL tree.
*/
static void fwrr_queue_srv(struct server *s)
{
struct proxy *p = s->proxy;
struct fwrr_group *grp;
grp = (s->state & SRV_BACKUP) ? &p->lbprm.fwrr.bck : &p->lbprm.fwrr.act;
/* Delay everything which does not fit into the window and everything
* which does not fit into the theorical new window.
*/
if (!srv_is_usable(s->state, s->eweight)) {
fwrr_remove_from_tree(s);
}
else if (s->eweight <= 0 ||
s->npos >= 2 * grp->curr_weight ||
s->npos >= grp->curr_weight + grp->next_weight) {
/* put into next tree, and readjust npos in case we could
* finally take this back to current. */
s->npos -= grp->curr_weight;
fwrr_queue_by_weight(grp->next, s);
}
else {
/* The sorting key is stored in units of s->npos * user_weight
* in order to avoid overflows. As stated in backend.h, the
* lower the scale, the rougher the weights modulation, and the
* higher the scale, the lower the number of servers without
* overflow. With this formula, the result is always positive,
* so we can use eb32_insert().
*/
s->lb_node.key = SRV_UWGHT_RANGE * s->npos +
(unsigned)(SRV_EWGHT_MAX + s->rweight - s->eweight) / BE_WEIGHT_SCALE;
eb32_insert(&grp->curr, &s->lb_node);
s->lb_tree = &grp->curr;
}
}
int assign_server(struct session *s)
{
struct connection *conn;
struct server *conn_slot;
struct server *srv, *prev_srv;
int err;
DPRINTF(stderr,"assign_server : s=%p\n",s);
err = SRV_STATUS_INTERNAL;
if (unlikely(s->pend_pos || s->flags & SN_ASSIGNED))
goto out_err;
prev_srv = objt_server(s->target);
conn_slot = s->srv_conn;
/* We have to release any connection slot before applying any LB algo,
* otherwise we may erroneously end up with no available slot.
*/
if (conn_slot)
sess_change_server(s, NULL);
/* We will now try to find the good server and store it into <objt_server(s->target)>.
* Note that <objt_server(s->target)> may be NULL in case of dispatch or proxy mode,
* as well as if no server is available (check error code).
*/
srv = NULL;
s->target = NULL;
conn = objt_conn(s->si[1].end);
if (conn &&
(conn->flags & CO_FL_CONNECTED) &&
objt_server(conn->target) && __objt_server(conn->target)->proxy == s->be &&
((s->txn.flags & TX_PREFER_LAST) ||
((s->be->options & PR_O_PREF_LAST) &&
(!s->be->max_ka_queue ||
server_has_room(__objt_server(conn->target)) ||
(__objt_server(conn->target)->nbpend + 1) < s->be->max_ka_queue))) &&
srv_is_usable(__objt_server(conn->target))) {
/* This session was relying on a server in a previous request
* and the proxy has "option prefer-last-server" set, so
* let's try to reuse the same server.
*/
srv = __objt_server(conn->target);
s->target = &srv->obj_type;
}
else if (s->be->lbprm.algo & BE_LB_KIND) {
/* we must check if we have at least one server available */
if (!s->be->lbprm.tot_weight) {
err = SRV_STATUS_NOSRV;
goto out;
}
/* First check whether we need to fetch some data or simply call
* the LB lookup function. Only the hashing functions will need
* some input data in fact, and will support multiple algorithms.
*/
switch (s->be->lbprm.algo & BE_LB_LKUP) {
case BE_LB_LKUP_RRTREE:
srv = fwrr_get_next_server(s->be, prev_srv);
break;
case BE_LB_LKUP_FSTREE:
srv = fas_get_next_server(s->be, prev_srv);
break;
case BE_LB_LKUP_LCTREE:
srv = fwlc_get_next_server(s->be, prev_srv);
break;
case BE_LB_LKUP_CHTREE:
case BE_LB_LKUP_MAP:
if ((s->be->lbprm.algo & BE_LB_KIND) == BE_LB_KIND_RR) {
if (s->be->lbprm.algo & BE_LB_LKUP_CHTREE)
srv = chash_get_next_server(s->be, prev_srv);
else
srv = map_get_server_rr(s->be, prev_srv);
break;
}
else if ((s->be->lbprm.algo & BE_LB_KIND) != BE_LB_KIND_HI) {
/* unknown balancing algorithm */
err = SRV_STATUS_INTERNAL;
goto out;
}
switch (s->be->lbprm.algo & BE_LB_PARM) {
case BE_LB_HASH_SRC:
conn = objt_conn(s->si[0].end);
if (conn && conn->addr.from.ss_family == AF_INET) {
srv = get_server_sh(s->be,
(void *)&((struct sockaddr_in *)&conn->addr.from)->sin_addr,
4);
}
else if (conn && conn->addr.from.ss_family == AF_INET6) {
srv = get_server_sh(s->be,
(void *)&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr,
16);
}
else {
/* unknown IP family */
err = SRV_STATUS_INTERNAL;
goto out;
}
break;
case BE_LB_HASH_URI:
/* URI hashing */
if (s->txn.req.msg_state < HTTP_MSG_BODY)
break;
srv = get_server_uh(s->be,
b_ptr(s->req.buf, -http_uri_rewind(&s->txn.req)),
s->txn.req.sl.rq.u_l);
break;
case BE_LB_HASH_PRM:
/* URL Parameter hashing */
if (s->txn.req.msg_state < HTTP_MSG_BODY)
break;
srv = get_server_ph(s->be,
b_ptr(s->req.buf, -http_uri_rewind(&s->txn.req)),
s->txn.req.sl.rq.u_l);
if (!srv && s->txn.meth == HTTP_METH_POST)
srv = get_server_ph_post(s);
break;
case BE_LB_HASH_HDR:
/* Header Parameter hashing */
if (s->txn.req.msg_state < HTTP_MSG_BODY)
break;
srv = get_server_hh(s);
break;
case BE_LB_HASH_RDP:
/* RDP Cookie hashing */
srv = get_server_rch(s);
break;
default:
/* unknown balancing algorithm */
err = SRV_STATUS_INTERNAL;
goto out;
}
/* If the hashing parameter was not found, let's fall
* back to round robin on the map.
*/
if (!srv) {
if (s->be->lbprm.algo & BE_LB_LKUP_CHTREE)
srv = chash_get_next_server(s->be, prev_srv);
else
srv = map_get_server_rr(s->be, prev_srv);
}
/* end of map-based LB */
break;
default:
/* unknown balancing algorithm */
err = SRV_STATUS_INTERNAL;
goto out;
}
if (!srv) {
err = SRV_STATUS_FULL;
goto out;
}
else if (srv != prev_srv) {
s->be->be_counters.cum_lbconn++;
srv->counters.cum_lbconn++;
}
s->target = &srv->obj_type;
}
else if (s->be->options & (PR_O_DISPATCH | PR_O_TRANSP)) {
s->target = &s->be->obj_type;
}
else if ((s->be->options & PR_O_HTTP_PROXY) &&
(conn = objt_conn(s->si[1].end)) &&
is_addr(&conn->addr.to)) {
/* in proxy mode, we need a valid destination address */
s->target = &s->be->obj_type;
}
else {
err = SRV_STATUS_NOSRV;
goto out;
}
s->flags |= SN_ASSIGNED;
err = SRV_STATUS_OK;
out:
/* Either we take back our connection slot, or we offer it to someone
* else if we don't need it anymore.
*/
if (conn_slot) {
if (conn_slot == srv) {
sess_change_server(s, srv);
} else {
if (may_dequeue_tasks(conn_slot, s->be))
process_srv_queue(conn_slot);
}
}
out_err:
return err;
}
/* 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;
}
