[packages/db4.7.git] / patch.4.7.25.4

*** dbinc/repmgr.h	2009-05-04 10:33:55.000000000 -0400
--- dbinc/repmgr.h	2009-05-04 10:27:26.000000000 -0400
***************
*** 374,379 ****
--- 374,380 ----
  #define	SITE_FROM_EID(eid)	(&db_rep->sites[eid])
  #define	EID_FROM_SITE(s)	((int)((s) - (&db_rep->sites[0])))
  #define	IS_VALID_EID(e)		((e) >= 0)
+ #define	IS_KNOWN_REMOTE_SITE(e)	((e) >= 0 && ((u_int)(e)) < db_rep->site_cnt)
  #define	SELF_EID		INT_MAX
  
  #define	IS_PEER_POLICY(p) ((p) == DB_REPMGR_ACKS_ALL_PEERS ||		\
*** rep/rep_elect.c	2009-05-04 10:35:50.000000000 -0400
--- rep/rep_elect.c	2009-05-04 10:31:24.000000000 -0400
***************
*** 33,39 ****
  static int __rep_fire_elected __P((ENV *, REP *, u_int32_t));
  static void __rep_elect_master __P((ENV *, REP *));
  static int __rep_tally __P((ENV *, REP *, int, u_int32_t *, u_int32_t, roff_t));
! static int __rep_wait __P((ENV *, db_timeout_t *, int *, int, u_int32_t));
  
  /*
   * __rep_elect --
--- 33,39 ----
  static int __rep_fire_elected __P((ENV *, REP *, u_int32_t));
  static void __rep_elect_master __P((ENV *, REP *));
  static int __rep_tally __P((ENV *, REP *, int, u_int32_t *, u_int32_t, roff_t));
! static int __rep_wait __P((ENV *, db_timeout_t *, int, u_int32_t));
  
  /*
   * __rep_elect --
***************
*** 55,61 ****
  	ENV *env;
  	LOG *lp;
  	REP *rep;
! 	int done, eid, elected, full_elect, locked, in_progress, need_req;
  	int ret, send_vote, t_ret;
  	u_int32_t ack, ctlflags, egen, nsites, orig_tally, priority, realpri;
  	u_int32_t tiebreaker;
--- 55,61 ----
  	ENV *env;
  	LOG *lp;
  	REP *rep;
! 	int done, elected, full_elect, locked, in_progress, need_req;
  	int ret, send_vote, t_ret;
  	u_int32_t ack, ctlflags, egen, nsites, orig_tally, priority, realpri;
  	u_int32_t tiebreaker;
***************
*** 181,188 ****
  			REP_SYSTEM_UNLOCK(env);
  			(void)__rep_send_message(env, DB_EID_BROADCAST,
  			    REP_MASTER_REQ, NULL, NULL, 0, 0);
! 			ret = __rep_wait(env, &to, &eid,
! 			    0, REP_F_EPHASE0);
  			REP_SYSTEM_LOCK(env);
  			F_CLR(rep, REP_F_EPHASE0);
  			switch (ret) {
--- 181,187 ----
  			REP_SYSTEM_UNLOCK(env);
  			(void)__rep_send_message(env, DB_EID_BROADCAST,
  			    REP_MASTER_REQ, NULL, NULL, 0, 0);
! 			ret = __rep_wait(env, &to, 0, REP_F_EPHASE0);
  			REP_SYSTEM_LOCK(env);
  			F_CLR(rep, REP_F_EPHASE0);
  			switch (ret) {
***************
*** 286,296 ****
  		REP_SYSTEM_LOCK(env);
  		goto vote;
  	}
! 	ret = __rep_wait(env, &to, &eid, full_elect, REP_F_EPHASE1);
  	switch (ret) {
  		case 0:
  			/* Check if election complete or phase complete. */
! 			if (eid != DB_EID_INVALID && !IN_ELECTION(rep)) {
  				RPRINT(env, DB_VERB_REP_ELECT,
  				    (env, "Ended election phase 1"));
  				goto edone;
--- 285,295 ----
  		REP_SYSTEM_LOCK(env);
  		goto vote;
  	}
! 	ret = __rep_wait(env, &to, full_elect, REP_F_EPHASE1);
  	switch (ret) {
  		case 0:
  			/* Check if election complete or phase complete. */
! 			if (!IN_ELECTION(rep)) {
  				RPRINT(env, DB_VERB_REP_ELECT,
  				    (env, "Ended election phase 1"));
  				goto edone;
***************
*** 398,412 ****
  		REP_SYSTEM_LOCK(env);
  		goto i_won;
  	}
! 	ret = __rep_wait(env, &to, &eid, full_elect, REP_F_EPHASE2);
  	RPRINT(env, DB_VERB_REP_ELECT,
  	    (env, "Ended election phase 2 %d", ret));
  	switch (ret) {
  		case 0:
! 			if (eid != DB_EID_INVALID)
! 				goto edone;
! 			ret = DB_REP_UNAVAIL;
! 			break;
  		case DB_REP_EGENCHG:
  			if (to > timeout)
  				to = timeout;
--- 397,408 ----
  		REP_SYSTEM_LOCK(env);
  		goto i_won;
  	}
! 	ret = __rep_wait(env, &to, full_elect, REP_F_EPHASE2);
  	RPRINT(env, DB_VERB_REP_ELECT,
  	    (env, "Ended election phase 2 %d", ret));
  	switch (ret) {
  		case 0:
! 			goto edone;
  		case DB_REP_EGENCHG:
  			if (to > timeout)
  				to = timeout;
***************
*** 1050,1062 ****
  	ENV *env;
  	REP *rep;
  {
- 	/*
- 	 * We often come through here twice, sometimes even more.  We mustn't
- 	 * let the redundant calls affect stats counting.  But rep_elect relies
- 	 * on this first part for setting eidp.
- 	 */
- 	rep->master_id = rep->eid;
- 
  	if (F_ISSET(rep, REP_F_MASTERELECT | REP_F_MASTER)) {
  		/* We've been through here already; avoid double counting. */
  		return;
--- 1046,1051 ----
***************
*** 1093,1102 ****
  	(timeout > 5000000) ? 500000 : ((timeout >= 10) ? timeout / 10 : 1);
  
  static int
! __rep_wait(env, timeoutp, eidp, full_elect, flags)
  	ENV *env;
  	db_timeout_t *timeoutp;
! 	int *eidp, full_elect;
  	u_int32_t flags;
  {
  	DB_REP *db_rep;
--- 1082,1091 ----
  	(timeout > 5000000) ? 500000 : ((timeout >= 10) ? timeout / 10 : 1);
  
  static int
! __rep_wait(env, timeoutp, full_elect, flags)
  	ENV *env;
  	db_timeout_t *timeoutp;
! 	int full_elect;
  	u_int32_t flags;
  {
  	DB_REP *db_rep;
***************
*** 1174,1180 ****
  			F_CLR(rep, REP_F_EGENUPDATE);
  			ret = DB_REP_EGENCHG;
  		} else if (phase_over) {
- 			*eidp = rep->master_id;
  			done = 1;
  			ret = 0;
  		}
--- 1163,1168 ----
*** repmgr/repmgr_net.c	2009-05-04 10:34:46.000000000 -0400
--- repmgr/repmgr_net.c	2009-05-04 10:27:26.000000000 -0400
***************
*** 100,105 ****
--- 100,107 ----
  		    control, rec, &nsites_sent, &npeers_sent)) != 0)
  			goto out;
  	} else {
+ 		DB_ASSERT(env, IS_KNOWN_REMOTE_SITE(eid));
+ 
  		/*
  		 * If this is a request that can be sent anywhere, then see if
  		 * we can send it to our peer (to save load on the master), but
Commit	Line	Data
1825d9ec ER	1	*** dbinc/repmgr.h 2009-05-04 10:33:55.000000000 -0400
1825d9ec ER	2	--- dbinc/repmgr.h 2009-05-04 10:27:26.000000000 -0400
2dee7534 ER	3	***************
	4	* 374,379 **
	5	--- 374,380 ----
	6	#define SITE_FROM_EID(eid) (&db_rep->sites[eid])
	7	#define EID_FROM_SITE(s) ((int)((s) - (&db_rep->sites[0])))
	8	#define IS_VALID_EID(e) ((e) >= 0)
	9	+ #define IS_KNOWN_REMOTE_SITE(e) ((e) >= 0 && ((u_int)(e)) < db_rep->site_cnt)
	10	#define SELF_EID INT_MAX
	11
	12	#define IS_PEER_POLICY(p) ((p) == DB_REPMGR_ACKS_ALL_PEERS \|\| \
1825d9ec ER	13	*** rep/rep_elect.c 2009-05-04 10:35:50.000000000 -0400
1825d9ec ER	14	--- rep/rep_elect.c 2009-05-04 10:31:24.000000000 -0400
2dee7534 ER	15	***************
	16	* 33,39 **
	17	static int __rep_fire_elected __P((ENV , REP , u_int32_t));
	18	static void __rep_elect_master __P((ENV , REP ));
	19	static int __rep_tally __P((ENV , REP , int, u_int32_t *, u_int32_t, roff_t));
	20	! static int __rep_wait __P((ENV , db_timeout_t , int *, int, u_int32_t));
	21
	22	/*
	23	* __rep_elect --
	24	--- 33,39 ----
	25	static int __rep_fire_elected __P((ENV , REP , u_int32_t));
	26	static void __rep_elect_master __P((ENV , REP ));
	27	static int __rep_tally __P((ENV , REP , int, u_int32_t *, u_int32_t, roff_t));
	28	! static int __rep_wait __P((ENV , db_timeout_t , int, u_int32_t));
	29
	30	/*
	31	* __rep_elect --
	32	***************
	33	* 55,61 **
	34	ENV *env;
	35	LOG *lp;
	36	REP *rep;
	37	! int done, eid, elected, full_elect, locked, in_progress, need_req;
	38	int ret, send_vote, t_ret;
	39	u_int32_t ack, ctlflags, egen, nsites, orig_tally, priority, realpri;
	40	u_int32_t tiebreaker;
	41	--- 55,61 ----
	42	ENV *env;
	43	LOG *lp;
	44	REP *rep;
	45	! int done, elected, full_elect, locked, in_progress, need_req;
	46	int ret, send_vote, t_ret;
	47	u_int32_t ack, ctlflags, egen, nsites, orig_tally, priority, realpri;
	48	u_int32_t tiebreaker;
	49	***************
	50	* 181,188 **
	51	REP_SYSTEM_UNLOCK(env);
	52	(void)__rep_send_message(env, DB_EID_BROADCAST,
	53	REP_MASTER_REQ, NULL, NULL, 0, 0);
	54	! ret = __rep_wait(env, &to, &eid,
	55	! 0, REP_F_EPHASE0);
	56	REP_SYSTEM_LOCK(env);
	57	F_CLR(rep, REP_F_EPHASE0);
	58	switch (ret) {
	59	--- 181,187 ----
	60	REP_SYSTEM_UNLOCK(env);
	61	(void)__rep_send_message(env, DB_EID_BROADCAST,
	62	REP_MASTER_REQ, NULL, NULL, 0, 0);
	63	! ret = __rep_wait(env, &to, 0, REP_F_EPHASE0);
	64	REP_SYSTEM_LOCK(env);
	65	F_CLR(rep, REP_F_EPHASE0);
	66	switch (ret) {
	67	***************
	68	* 286,296 **
	69	REP_SYSTEM_LOCK(env);
	70	goto vote;
	71	}
	72	! ret = __rep_wait(env, &to, &eid, full_elect, REP_F_EPHASE1);
	73	switch (ret) {
	74	case 0:
	75	/* Check if election complete or phase complete. */
	76	! if (eid != DB_EID_INVALID && !IN_ELECTION(rep)) {
	77	RPRINT(env, DB_VERB_REP_ELECT,
	78	(env, "Ended election phase 1"));
79	goto edone;
80	--- 285,295 ----
81	REP_SYSTEM_LOCK(env);
82	goto vote;
83	}
84	! ret = __rep_wait(env, &to, full_elect, REP_F_EPHASE1);
85	switch (ret) {
86	case 0:
87	/* Check if election complete or phase complete. */
88	! if (!IN_ELECTION(rep)) {
89	RPRINT(env, DB_VERB_REP_ELECT,
90	(env, "Ended election phase 1"));
91	goto edone;
92	***************
93	* 398,412 **
94	REP_SYSTEM_LOCK(env);
95	goto i_won;
96	}
97	! ret = __rep_wait(env, &to, &eid, full_elect, REP_F_EPHASE2);
98	RPRINT(env, DB_VERB_REP_ELECT,
99	(env, "Ended election phase 2 %d", ret));
100	switch (ret) {
101	case 0:
102	! if (eid != DB_EID_INVALID)
103	! goto edone;
104	! ret = DB_REP_UNAVAIL;
105	! break;
106	case DB_REP_EGENCHG:
107	if (to > timeout)
108	to = timeout;
109	--- 397,408 ----
110	REP_SYSTEM_LOCK(env);
111	goto i_won;
112	}
113	! ret = __rep_wait(env, &to, full_elect, REP_F_EPHASE2);
114	RPRINT(env, DB_VERB_REP_ELECT,
115	(env, "Ended election phase 2 %d", ret));
116	switch (ret) {
117	case 0:
118	! goto edone;
119	case DB_REP_EGENCHG:
120	if (to > timeout)
121	to = timeout;
122	***************
123	* 1050,1062 **
124	ENV *env;
125	REP *rep;
126	{
127	- /*
128	- * We often come through here twice, sometimes even more. We mustn't
129	- * let the redundant calls affect stats counting. But rep_elect relies
130	- * on this first part for setting eidp.
131	- */
132	- rep->master_id = rep->eid;
133	-
134	if (F_ISSET(rep, REP_F_MASTERELECT \| REP_F_MASTER)) {
135	/* We've been through here already; avoid double counting. */
136	return;
137	--- 1046,1051 ----
138	***************
139	* 1093,1102 **
140	(timeout > 5000000) ? 500000 : ((timeout >= 10) ? timeout / 10 : 1);
141
142	static int
143	! __rep_wait(env, timeoutp, eidp, full_elect, flags)
144	ENV *env;
145	db_timeout_t *timeoutp;
146	! int *eidp, full_elect;
147	u_int32_t flags;
148	{
149	DB_REP *db_rep;
150	--- 1082,1091 ----
151	(timeout > 5000000) ? 500000 : ((timeout >= 10) ? timeout / 10 : 1);
152
153	static int
154	! __rep_wait(env, timeoutp, full_elect, flags)
155	ENV *env;
156	db_timeout_t *timeoutp;
157	! int full_elect;
158	u_int32_t flags;
159	{
160	DB_REP *db_rep;
161	***************
162	* 1174,1180 **
163	F_CLR(rep, REP_F_EGENUPDATE);
164	ret = DB_REP_EGENCHG;
165	} else if (phase_over) {
166	- *eidp = rep->master_id;
167	done = 1;
168	ret = 0;
169	}
170	--- 1163,1168 ----
1825d9ec ER	171	*** repmgr/repmgr_net.c 2009-05-04 10:34:46.000000000 -0400
1825d9ec ER	172	--- repmgr/repmgr_net.c 2009-05-04 10:27:26.000000000 -0400
2dee7534 ER	173	***************
	174	* 100,105 **
	175	--- 100,107 ----
	176	control, rec, &nsites_sent, &npeers_sent)) != 0)
	177	goto out;
	178	} else {
	179	+ DB_ASSERT(env, IS_KNOWN_REMOTE_SITE(eid));
	180	+
	181	/*
	182	* If this is a request that can be sent anywhere, then see if
	183	* we can send it to our peer (to save load on the master), but