040-patch.4.7.25.4.patch (3907B)
1 --- a/dbinc/repmgr.h 2 +++ b/dbinc/repmgr.h 3 @@ -374,6 +374,7 @@ typedef struct { 4 #define SITE_FROM_EID(eid) (&db_rep->sites[eid]) 5 #define EID_FROM_SITE(s) ((int)((s) - (&db_rep->sites[0]))) 6 #define IS_VALID_EID(e) ((e) >= 0) 7 +#define IS_KNOWN_REMOTE_SITE(e) ((e) >= 0 && ((u_int)(e)) < db_rep->site_cnt) 8 #define SELF_EID INT_MAX 9 10 #define IS_PEER_POLICY(p) ((p) == DB_REPMGR_ACKS_ALL_PEERS || \ 11 --- a/rep/rep_elect.c 12 +++ b/rep/rep_elect.c 13 @@ -33,7 +33,7 @@ static int __rep_elect_init 14 static int __rep_fire_elected __P((ENV *, REP *, u_int32_t)); 15 static void __rep_elect_master __P((ENV *, REP *)); 16 static int __rep_tally __P((ENV *, REP *, int, u_int32_t *, u_int32_t, roff_t)); 17 -static int __rep_wait __P((ENV *, db_timeout_t *, int *, int, u_int32_t)); 18 +static int __rep_wait __P((ENV *, db_timeout_t *, int, u_int32_t)); 19 20 /* 21 * __rep_elect -- 22 @@ -55,7 +55,7 @@ __rep_elect(dbenv, given_nsites, nvotes, 23 ENV *env; 24 LOG *lp; 25 REP *rep; 26 - int done, eid, elected, full_elect, locked, in_progress, need_req; 27 + int done, elected, full_elect, locked, in_progress, need_req; 28 int ret, send_vote, t_ret; 29 u_int32_t ack, ctlflags, egen, nsites, orig_tally, priority, realpri; 30 u_int32_t tiebreaker; 31 @@ -181,8 +181,7 @@ __rep_elect(dbenv, given_nsites, nvotes, 32 REP_SYSTEM_UNLOCK(env); 33 (void)__rep_send_message(env, DB_EID_BROADCAST, 34 REP_MASTER_REQ, NULL, NULL, 0, 0); 35 - ret = __rep_wait(env, &to, &eid, 36 - 0, REP_F_EPHASE0); 37 + ret = __rep_wait(env, &to, 0, REP_F_EPHASE0); 38 REP_SYSTEM_LOCK(env); 39 F_CLR(rep, REP_F_EPHASE0); 40 switch (ret) { 41 @@ -286,11 +285,11 @@ restart: 42 REP_SYSTEM_LOCK(env); 43 goto vote; 44 } 45 - ret = __rep_wait(env, &to, &eid, full_elect, REP_F_EPHASE1); 46 + ret = __rep_wait(env, &to, full_elect, REP_F_EPHASE1); 47 switch (ret) { 48 case 0: 49 /* Check if election complete or phase complete. */ 50 - if (eid != DB_EID_INVALID && !IN_ELECTION(rep)) { 51 + if (!IN_ELECTION(rep)) { 52 RPRINT(env, DB_VERB_REP_ELECT, 53 (env, "Ended election phase 1")); 54 goto edone; 55 @@ -398,15 +397,12 @@ phase2: 56 REP_SYSTEM_LOCK(env); 57 goto i_won; 58 } 59 - ret = __rep_wait(env, &to, &eid, full_elect, REP_F_EPHASE2); 60 + ret = __rep_wait(env, &to, full_elect, REP_F_EPHASE2); 61 RPRINT(env, DB_VERB_REP_ELECT, 62 (env, "Ended election phase 2 %d", ret)); 63 switch (ret) { 64 case 0: 65 - if (eid != DB_EID_INVALID) 66 - goto edone; 67 - ret = DB_REP_UNAVAIL; 68 - break; 69 + goto edone; 70 case DB_REP_EGENCHG: 71 if (to > timeout) 72 to = timeout; 73 @@ -1050,13 +1046,6 @@ __rep_elect_master(env, rep) 74 ENV *env; 75 REP *rep; 76 { 77 - /* 78 - * We often come through here twice, sometimes even more. We mustn't 79 - * let the redundant calls affect stats counting. But rep_elect relies 80 - * on this first part for setting eidp. 81 - */ 82 - rep->master_id = rep->eid; 83 - 84 if (F_ISSET(rep, REP_F_MASTERELECT | REP_F_MASTER)) { 85 /* We've been through here already; avoid double counting. */ 86 return; 87 @@ -1093,10 +1082,10 @@ __rep_fire_elected(env, rep, egen) 88 (timeout > 5000000) ? 500000 : ((timeout >= 10) ? timeout / 10 : 1); 89 90 static int 91 -__rep_wait(env, timeoutp, eidp, full_elect, flags) 92 +__rep_wait(env, timeoutp, full_elect, flags) 93 ENV *env; 94 db_timeout_t *timeoutp; 95 - int *eidp, full_elect; 96 + int full_elect; 97 u_int32_t flags; 98 { 99 DB_REP *db_rep; 100 @@ -1174,7 +1163,6 @@ __rep_wait(env, timeoutp, eidp, full_ele 101 F_CLR(rep, REP_F_EGENUPDATE); 102 ret = DB_REP_EGENCHG; 103 } else if (phase_over) { 104 - *eidp = rep->master_id; 105 done = 1; 106 ret = 0; 107 } 108 --- a/repmgr/repmgr_net.c 109 +++ b/repmgr/repmgr_net.c 110 @@ -100,6 +100,8 @@ __repmgr_send(dbenv, control, rec, lsnp, 111 control, rec, &nsites_sent, &npeers_sent)) != 0) 112 goto out; 113 } else { 114 + DB_ASSERT(env, IS_KNOWN_REMOTE_SITE(eid)); 115 + 116 /* 117 * If this is a request that can be sent anywhere, then see if 118 * we can send it to our peer (to save load on the master), but