blob: 630467b07288593ea317651150a5866787ee4427 [file] [log] [blame]
Willy Tarreauf89c1872009-10-01 11:19:37 +02001/*
2 * Fast Weighted Least Connection load balancing algorithm.
3 *
4 * Copyright 2000-2009 Willy Tarreau <w@1wt.eu>
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
Willy Tarreau1e56f922020-06-04 23:20:13 +020013#include <import/eb32tree.h>
Willy Tarreau4c7e4b72020-05-27 12:58:42 +020014#include <haproxy/api.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020015#include <haproxy/backend.h>
Willy Tarreaua55c4542020-06-04 22:59:39 +020016#include <haproxy/queue.h>
Willy Tarreau1e56f922020-06-04 23:20:13 +020017#include <haproxy/server-t.h>
Willy Tarreauf89c1872009-10-01 11:19:37 +020018
19
20/* Remove a server from a tree. It must have previously been dequeued. This
21 * function is meant to be called when a server is going down or has its
22 * weight disabled.
Willy Tarreau1b877482018-08-21 19:44:53 +020023 *
24 * The server's lock and the lbprm's lock must be held.
Willy Tarreauf89c1872009-10-01 11:19:37 +020025 */
26static inline void fwlc_remove_from_tree(struct server *s)
27{
28 s->lb_tree = NULL;
29}
30
Willy Tarreau1b877482018-08-21 19:44:53 +020031/* simply removes a server from a tree.
32 *
33 * The server's lock and the lbprm's lock must be held.
34 */
Willy Tarreauf89c1872009-10-01 11:19:37 +020035static inline void fwlc_dequeue_srv(struct server *s)
36{
37 eb32_delete(&s->lb_node);
38}
39
Christopher Faulet08818422020-12-11 15:36:01 +010040/* Queue a server in its associated tree, assuming the <eweight> is >0.
Willy Tarreau1eb6c552018-12-14 08:33:28 +010041 * Servers are sorted by (#conns+1)/weight. To ensure maximum accuracy,
42 * we use (#conns+1)*SRV_EWGHT_MAX/eweight as the sorting key. The reason
43 * for using #conns+1 is to sort by weights in case the server is picked
44 * and not before it is picked. This provides a better load accuracy for
45 * low connection counts when weights differ and makes sure the round-robin
Willy Tarreaued5ac9c2019-09-06 17:04:04 +020046 * applies between servers of highest weight first. However servers with no
47 * connection are always picked first so that under low loads, it's not
48 * always the single server with the highest weight that gets picked.
Willy Tarreau1b877482018-08-21 19:44:53 +020049 *
Christopher Faulet08818422020-12-11 15:36:01 +010050 * NOTE: Depending on the calling context, we use s->next_eweight or
51 * s->cur_eweight. The next value is used when the server state is updated
52 * (because the weight changed for instance). During this step, the server
53 * state is not yet committed. The current value is used to reposition the
54 * server in the tree. This happens when the server is used.
55 *
Willy Tarreau1b877482018-08-21 19:44:53 +020056 * The server's lock and the lbprm's lock must be held.
Willy Tarreauf89c1872009-10-01 11:19:37 +020057 */
Christopher Faulet08818422020-12-11 15:36:01 +010058static inline void fwlc_queue_srv(struct server *s, unsigned int eweight)
Willy Tarreauf89c1872009-10-01 11:19:37 +020059{
Willy Tarreau8c855f62020-10-22 17:41:45 +020060 unsigned int inflight = s->served + s->nbpend;
61
Christopher Faulet08818422020-12-11 15:36:01 +010062 s->lb_node.key = inflight ? (inflight + 1) * SRV_EWGHT_MAX / eweight : 0;
Willy Tarreauf89c1872009-10-01 11:19:37 +020063 eb32_insert(s->lb_tree, &s->lb_node);
64}
65
66/* Re-position the server in the FWLC tree after it has been assigned one
67 * connection or after it has released one. Note that it is possible that
68 * the server has been moved out of the tree due to failed health-checks.
Willy Tarreau1b877482018-08-21 19:44:53 +020069 *
70 * The server's lock must be held. The lbprm's lock will be used.
Willy Tarreauf89c1872009-10-01 11:19:37 +020071 */
72static void fwlc_srv_reposition(struct server *s)
73{
Willy Tarreaucd10def2020-10-17 18:48:47 +020074 HA_RWLOCK_WRLOCK(LBPRM_LOCK, &s->proxy->lbprm.lock);
Christopher Faulet1ae2a882019-06-19 10:50:38 +020075 if (s->lb_tree) {
76 fwlc_dequeue_srv(s);
Christopher Faulet08818422020-12-11 15:36:01 +010077 fwlc_queue_srv(s, s->cur_eweight);
Christopher Faulet1ae2a882019-06-19 10:50:38 +020078 }
Willy Tarreaucd10def2020-10-17 18:48:47 +020079 HA_RWLOCK_WRUNLOCK(LBPRM_LOCK, &s->proxy->lbprm.lock);
Willy Tarreauf89c1872009-10-01 11:19:37 +020080}
81
82/* This function updates the server trees according to server <srv>'s new
83 * state. It should be called when server <srv>'s status changes to down.
84 * It is not important whether the server was already down or not. It is not
85 * important either that the new state is completely down (the caller may not
86 * know all the variables of a server's state).
Willy Tarreau1b877482018-08-21 19:44:53 +020087 *
88 * The server's lock must be held. The lbprm's lock will be used.
Willy Tarreauf89c1872009-10-01 11:19:37 +020089 */
90static void fwlc_set_server_status_down(struct server *srv)
91{
92 struct proxy *p = srv->proxy;
93
Willy Tarreauc5150da2014-05-13 19:27:31 +020094 if (!srv_lb_status_changed(srv))
Willy Tarreauf89c1872009-10-01 11:19:37 +020095 return;
96
Emeric Brun52a91d32017-08-31 14:41:55 +020097 if (srv_willbe_usable(srv))
Willy Tarreauf89c1872009-10-01 11:19:37 +020098 goto out_update_state;
Willy Tarreaucd10def2020-10-17 18:48:47 +020099 HA_RWLOCK_WRLOCK(LBPRM_LOCK, &p->lbprm.lock);
Willy Tarreau1b877482018-08-21 19:44:53 +0200100
Willy Tarreauf89c1872009-10-01 11:19:37 +0200101
Emeric Brun52a91d32017-08-31 14:41:55 +0200102 if (!srv_currently_usable(srv))
Willy Tarreauf89c1872009-10-01 11:19:37 +0200103 /* server was already down */
104 goto out_update_backend;
105
Willy Tarreauc93cd162014-05-13 15:54:22 +0200106 if (srv->flags & SRV_F_BACKUP) {
Emeric Brun52a91d32017-08-31 14:41:55 +0200107 p->lbprm.tot_wbck -= srv->cur_eweight;
Willy Tarreauf89c1872009-10-01 11:19:37 +0200108 p->srv_bck--;
109
110 if (srv == p->lbprm.fbck) {
111 /* we lost the first backup server in a single-backup
112 * configuration, we must search another one.
113 */
114 struct server *srv2 = p->lbprm.fbck;
115 do {
116 srv2 = srv2->next;
117 } while (srv2 &&
Willy Tarreauc93cd162014-05-13 15:54:22 +0200118 !((srv2->flags & SRV_F_BACKUP) &&
Emeric Brun52a91d32017-08-31 14:41:55 +0200119 srv_willbe_usable(srv2)));
Willy Tarreauf89c1872009-10-01 11:19:37 +0200120 p->lbprm.fbck = srv2;
121 }
122 } else {
Emeric Brun52a91d32017-08-31 14:41:55 +0200123 p->lbprm.tot_wact -= srv->cur_eweight;
Willy Tarreauf89c1872009-10-01 11:19:37 +0200124 p->srv_act--;
125 }
126
127 fwlc_dequeue_srv(srv);
128 fwlc_remove_from_tree(srv);
129
130out_update_backend:
131 /* check/update tot_used, tot_weight */
132 update_backend_weight(p);
Willy Tarreaucd10def2020-10-17 18:48:47 +0200133 HA_RWLOCK_WRUNLOCK(LBPRM_LOCK, &p->lbprm.lock);
Willy Tarreau1b877482018-08-21 19:44:53 +0200134
Willy Tarreauf89c1872009-10-01 11:19:37 +0200135 out_update_state:
Willy Tarreauc5150da2014-05-13 19:27:31 +0200136 srv_lb_commit_status(srv);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200137}
138
139/* This function updates the server trees according to server <srv>'s new
140 * state. It should be called when server <srv>'s status changes to up.
141 * It is not important whether the server was already down or not. It is not
142 * important either that the new state is completely UP (the caller may not
143 * know all the variables of a server's state). This function will not change
144 * the weight of a server which was already up.
Willy Tarreau1b877482018-08-21 19:44:53 +0200145 *
146 * The server's lock must be held. The lbprm's lock will be used.
Willy Tarreauf89c1872009-10-01 11:19:37 +0200147 */
148static void fwlc_set_server_status_up(struct server *srv)
149{
150 struct proxy *p = srv->proxy;
151
Willy Tarreauc5150da2014-05-13 19:27:31 +0200152 if (!srv_lb_status_changed(srv))
Willy Tarreauf89c1872009-10-01 11:19:37 +0200153 return;
154
Emeric Brun52a91d32017-08-31 14:41:55 +0200155 if (!srv_willbe_usable(srv))
Willy Tarreauf89c1872009-10-01 11:19:37 +0200156 goto out_update_state;
157
Willy Tarreaucd10def2020-10-17 18:48:47 +0200158 HA_RWLOCK_WRLOCK(LBPRM_LOCK, &p->lbprm.lock);
Willy Tarreau1b877482018-08-21 19:44:53 +0200159
Emeric Brun52a91d32017-08-31 14:41:55 +0200160 if (srv_currently_usable(srv))
Willy Tarreauf89c1872009-10-01 11:19:37 +0200161 /* server was already up */
162 goto out_update_backend;
163
Willy Tarreauc93cd162014-05-13 15:54:22 +0200164 if (srv->flags & SRV_F_BACKUP) {
Willy Tarreauf89c1872009-10-01 11:19:37 +0200165 srv->lb_tree = &p->lbprm.fwlc.bck;
Emeric Brun52a91d32017-08-31 14:41:55 +0200166 p->lbprm.tot_wbck += srv->next_eweight;
Willy Tarreauf89c1872009-10-01 11:19:37 +0200167 p->srv_bck++;
168
169 if (!(p->options & PR_O_USE_ALL_BK)) {
170 if (!p->lbprm.fbck) {
171 /* there was no backup server anymore */
172 p->lbprm.fbck = srv;
173 } else {
174 /* we may have restored a backup server prior to fbck,
175 * in which case it should replace it.
176 */
177 struct server *srv2 = srv;
178 do {
179 srv2 = srv2->next;
180 } while (srv2 && (srv2 != p->lbprm.fbck));
181 if (srv2)
182 p->lbprm.fbck = srv;
183 }
184 }
185 } else {
186 srv->lb_tree = &p->lbprm.fwlc.act;
Emeric Brun52a91d32017-08-31 14:41:55 +0200187 p->lbprm.tot_wact += srv->next_eweight;
Willy Tarreauf89c1872009-10-01 11:19:37 +0200188 p->srv_act++;
189 }
190
191 /* note that eweight cannot be 0 here */
Christopher Faulet08818422020-12-11 15:36:01 +0100192 fwlc_queue_srv(srv, srv->next_eweight);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200193
194 out_update_backend:
195 /* check/update tot_used, tot_weight */
196 update_backend_weight(p);
Willy Tarreaucd10def2020-10-17 18:48:47 +0200197 HA_RWLOCK_WRUNLOCK(LBPRM_LOCK, &p->lbprm.lock);
Willy Tarreau1b877482018-08-21 19:44:53 +0200198
Willy Tarreauf89c1872009-10-01 11:19:37 +0200199 out_update_state:
Willy Tarreauc5150da2014-05-13 19:27:31 +0200200 srv_lb_commit_status(srv);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200201}
202
203/* This function must be called after an update to server <srv>'s effective
204 * weight. It may be called after a state change too.
Willy Tarreau1b877482018-08-21 19:44:53 +0200205 *
206 * The server's lock must be held. The lbprm's lock will be used.
Willy Tarreauf89c1872009-10-01 11:19:37 +0200207 */
208static void fwlc_update_server_weight(struct server *srv)
209{
210 int old_state, new_state;
211 struct proxy *p = srv->proxy;
212
Willy Tarreauc5150da2014-05-13 19:27:31 +0200213 if (!srv_lb_status_changed(srv))
Willy Tarreauf89c1872009-10-01 11:19:37 +0200214 return;
215
216 /* If changing the server's weight changes its state, we simply apply
217 * the procedures we already have for status change. If the state
218 * remains down, the server is not in any tree, so it's as easy as
219 * updating its values. If the state remains up with different weights,
220 * there are some computations to perform to find a new place and
221 * possibly a new tree for this server.
222 */
223
Emeric Brun52a91d32017-08-31 14:41:55 +0200224 old_state = srv_currently_usable(srv);
225 new_state = srv_willbe_usable(srv);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200226
227 if (!old_state && !new_state) {
Willy Tarreauc5150da2014-05-13 19:27:31 +0200228 srv_lb_commit_status(srv);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200229 return;
230 }
231 else if (!old_state && new_state) {
232 fwlc_set_server_status_up(srv);
233 return;
234 }
235 else if (old_state && !new_state) {
236 fwlc_set_server_status_down(srv);
237 return;
238 }
239
Willy Tarreaucd10def2020-10-17 18:48:47 +0200240 HA_RWLOCK_WRLOCK(LBPRM_LOCK, &p->lbprm.lock);
Willy Tarreau1b877482018-08-21 19:44:53 +0200241
Willy Tarreauf89c1872009-10-01 11:19:37 +0200242 if (srv->lb_tree)
243 fwlc_dequeue_srv(srv);
244
Willy Tarreauc93cd162014-05-13 15:54:22 +0200245 if (srv->flags & SRV_F_BACKUP) {
Emeric Brun52a91d32017-08-31 14:41:55 +0200246 p->lbprm.tot_wbck += srv->next_eweight - srv->cur_eweight;
Willy Tarreauf89c1872009-10-01 11:19:37 +0200247 srv->lb_tree = &p->lbprm.fwlc.bck;
248 } else {
Emeric Brun52a91d32017-08-31 14:41:55 +0200249 p->lbprm.tot_wact += srv->next_eweight - srv->cur_eweight;
Willy Tarreauf89c1872009-10-01 11:19:37 +0200250 srv->lb_tree = &p->lbprm.fwlc.act;
251 }
252
Christopher Faulet08818422020-12-11 15:36:01 +0100253 fwlc_queue_srv(srv, srv->next_eweight);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200254
255 update_backend_weight(p);
Willy Tarreaucd10def2020-10-17 18:48:47 +0200256 HA_RWLOCK_WRUNLOCK(LBPRM_LOCK, &p->lbprm.lock);
Willy Tarreau1b877482018-08-21 19:44:53 +0200257
Willy Tarreauc5150da2014-05-13 19:27:31 +0200258 srv_lb_commit_status(srv);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200259}
260
261/* This function is responsible for building the trees in case of fast
262 * weighted least-conns. It also sets p->lbprm.wdiv to the eweight to
263 * uweight ratio. Both active and backup groups are initialized.
264 */
265void fwlc_init_server_tree(struct proxy *p)
266{
267 struct server *srv;
268 struct eb_root init_head = EB_ROOT;
269
270 p->lbprm.set_server_status_up = fwlc_set_server_status_up;
271 p->lbprm.set_server_status_down = fwlc_set_server_status_down;
272 p->lbprm.update_server_eweight = fwlc_update_server_weight;
273 p->lbprm.server_take_conn = fwlc_srv_reposition;
274 p->lbprm.server_drop_conn = fwlc_srv_reposition;
275
276 p->lbprm.wdiv = BE_WEIGHT_SCALE;
277 for (srv = p->srv; srv; srv = srv->next) {
Emeric Brun52a91d32017-08-31 14:41:55 +0200278 srv->next_eweight = (srv->uweight * p->lbprm.wdiv + p->lbprm.wmult - 1) / p->lbprm.wmult;
Willy Tarreauc5150da2014-05-13 19:27:31 +0200279 srv_lb_commit_status(srv);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200280 }
281
282 recount_servers(p);
283 update_backend_weight(p);
284
285 p->lbprm.fwlc.act = init_head;
286 p->lbprm.fwlc.bck = init_head;
287
288 /* queue active and backup servers in two distinct groups */
289 for (srv = p->srv; srv; srv = srv->next) {
Emeric Brun52a91d32017-08-31 14:41:55 +0200290 if (!srv_currently_usable(srv))
Willy Tarreauf89c1872009-10-01 11:19:37 +0200291 continue;
Willy Tarreauc93cd162014-05-13 15:54:22 +0200292 srv->lb_tree = (srv->flags & SRV_F_BACKUP) ? &p->lbprm.fwlc.bck : &p->lbprm.fwlc.act;
Christopher Faulet08818422020-12-11 15:36:01 +0100293 fwlc_queue_srv(srv, srv->next_eweight);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200294 }
295}
296
297/* Return next server from the FWLC tree in backend <p>. If the tree is empty,
298 * return NULL. Saturated servers are skipped.
Willy Tarreau1b877482018-08-21 19:44:53 +0200299 *
300 * The server's lock must be held. The lbprm's lock will be used.
Willy Tarreauf89c1872009-10-01 11:19:37 +0200301 */
302struct server *fwlc_get_next_server(struct proxy *p, struct server *srvtoavoid)
303{
304 struct server *srv, *avoided;
305 struct eb32_node *node;
306
307 srv = avoided = NULL;
308
Willy Tarreau58bc9c12020-10-17 19:32:09 +0200309 HA_RWLOCK_RDLOCK(LBPRM_LOCK, &p->lbprm.lock);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200310 if (p->srv_act)
311 node = eb32_first(&p->lbprm.fwlc.act);
Christopher Faulet5b517552017-06-09 14:17:53 +0200312 else if (p->lbprm.fbck) {
313 srv = p->lbprm.fbck;
314 goto out;
315 }
Willy Tarreauf89c1872009-10-01 11:19:37 +0200316 else if (p->srv_bck)
317 node = eb32_first(&p->lbprm.fwlc.bck);
Christopher Faulet5b517552017-06-09 14:17:53 +0200318 else {
319 srv = NULL;
320 goto out;
321 }
Willy Tarreauf89c1872009-10-01 11:19:37 +0200322
323 while (node) {
324 /* OK, we have a server. However, it may be saturated, in which
325 * case we don't want to reconsider it for now, so we'll simply
326 * skip it. Same if it's the server we try to avoid, in which
327 * case we simply remember it for later use if needed.
328 */
329 struct server *s;
330
331 s = eb32_entry(node, struct server, lb_node);
Willy Tarreau8ae8c482020-10-22 17:19:07 +0200332 if (!s->maxconn || s->served + s->nbpend < srv_dynamic_maxconn(s) + s->maxqueue) {
Willy Tarreauf89c1872009-10-01 11:19:37 +0200333 if (s != srvtoavoid) {
334 srv = s;
335 break;
336 }
337 avoided = s;
338 }
339 node = eb32_next(node);
340 }
341
342 if (!srv)
343 srv = avoided;
Christopher Faulet5b517552017-06-09 14:17:53 +0200344 out:
Willy Tarreau58bc9c12020-10-17 19:32:09 +0200345 HA_RWLOCK_RDUNLOCK(LBPRM_LOCK, &p->lbprm.lock);
Willy Tarreauf89c1872009-10-01 11:19:37 +0200346 return srv;
347}
348
349
350/*
351 * Local variables:
352 * c-indent-level: 8
353 * c-basic-offset: 8
354 * End:
355 */