diff --git a/src/common/backend/catalog/builtin_funcs.ini b/src/common/backend/catalog/builtin_funcs.ini index 3bce48d3..db57ccec 100755 --- a/src/common/backend/catalog/builtin_funcs.ini +++ b/src/common/backend/catalog/builtin_funcs.ini @@ -7557,7 +7557,7 @@ ), AddFuncGroup( "pg_stat_get_wal_senders", 1, - AddBuiltinFunc(_0(3099), _1("pg_stat_get_wal_senders"), _2(0), _3(false), _4(true), _5(pg_stat_get_wal_senders), _6(2249), _7(PG_CATALOG_NAMESPACE), _8(BOOTSTRAP_SUPERUSERID), _9(INTERNALlanguageId), _10(1), _11(10), _12(0), _13(0), _14(false), _15(false), _16(false), _17(false), _18('s'), _19(0), _20(0), _21(21, 20, 23, 25, 25, 25, 25, 1184, 1184, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 23, 25, 25), _22(21, 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o'), _23(21, "pid", "sender_pid", "local_role", "peer_role", "peer_state", "state", "catchup_start", "catchup_end", "sender_sent_location", "sender_write_location", "sender_flush_location", "sender_replay_location", "receiver_received_location", "receiver_write_location", "receiver_flush_location", "receiver_replay_location", "sync_percent", "sync_state", "sync_priority", "sync_most_available", "channel"), _24(NULL), _25("pg_stat_get_wal_senders"), _26(NULL), _27(NULL), _28(NULL), _29(0), _30(false), _31(NULL), _32(false), _33(NULL), _34('f')) + AddBuiltinFunc(_0(3099), _1("pg_stat_get_wal_senders"), _2(0), _3(false), _4(true), _5(pg_stat_get_wal_senders), _6(2249), _7(PG_CATALOG_NAMESPACE), _8(BOOTSTRAP_SUPERUSERID), _9(INTERNALlanguageId), _10(1), _11(10), _12(0), _13(0), _14(false), _15(false), _16(false), _17(false), _18('s'), _19(0), _20(0), _21(22, 20, 23, 25, 25, 25, 25, 1184, 1184, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 23, 23, 25, 25), _22(22, 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o', 'o'), _23(22, "pid", "sender_pid", "local_role", "peer_role", "peer_state", "state", "catchup_start", "catchup_end", "sender_sent_location", "sender_write_location", "sender_flush_location", "sender_replay_location", "receiver_received_location", "receiver_write_location", "receiver_flush_location", "receiver_replay_location", "sync_percent", "sync_state", "sync_group", "sync_priority", "sync_most_available", "channel"), _24(NULL), _25("pg_stat_get_wal_senders"), _26(NULL), _27(NULL), _28(NULL), _29(0), _30(false), _31(NULL), _32(false), _33(NULL), _34('f')) ), AddFuncGroup( "pg_stat_get_wlm_ec_operator_info", 1, diff --git a/src/common/backend/utils/init/globals.cpp b/src/common/backend/utils/init/globals.cpp index 81716296..7db66384 100644 --- a/src/common/backend/utils/init/globals.cpp +++ b/src/common/backend/utils/init/globals.cpp @@ -59,7 +59,7 @@ bool open_join_children = true; bool will_shutdown = false; /* hard-wired binary version number */ -const uint32 GRAND_VERSION_NUM = 92300; +const uint32 GRAND_VERSION_NUM = 92301; const uint32 MATVIEW_VERSION_NUM = 92213; const uint32 PARTIALPUSH_VERSION_NUM = 92087; diff --git a/src/gausskernel/process/threadpool/knl_thread.cpp b/src/gausskernel/process/threadpool/knl_thread.cpp index 1c3e6698..26b897b9 100644 --- a/src/gausskernel/process/threadpool/knl_thread.cpp +++ b/src/gausskernel/process/threadpool/knl_thread.cpp @@ -759,7 +759,7 @@ static void knl_t_replscanner_init(knl_t_replscanner_context* replscanner_cxt) static void knl_t_syncgram_init(knl_t_syncrepgram_context* syncrepgram_cxt) { - syncrepgram_cxt->syncrep_parse_result = NULL; + syncrepgram_cxt->syncrep_parse_result = NIL; } static void knl_t_syncrepscanner_init(knl_t_syncrepscanner_context* syncrepscanner_cxt) @@ -772,6 +772,7 @@ static void knl_t_syncrepscanner_init(knl_t_syncrepscanner_context* syncrepscann static void knl_t_syncrep_init(knl_t_syncrep_context* syncrep_cxt) { syncrep_cxt->SyncRepConfig = NULL; + syncrep_cxt->SyncRepConfigGroups = 0; syncrep_cxt->announce_next_takeover = true; } diff --git a/src/gausskernel/storage/replication/syncrep.cpp b/src/gausskernel/storage/replication/syncrep.cpp index 5f757feb..976016e6 100644 --- a/src/gausskernel/storage/replication/syncrep.cpp +++ b/src/gausskernel/storage/replication/syncrep.cpp @@ -80,7 +80,7 @@ static int SyncRepWakeQueue(bool all, int mode); static void SyncRepWaitCompletionQueue(); static void SyncRepNotifyComplete(); -static int SyncRepGetStandbyPriority(void); +static void SyncRepGetStandbyGroupAndPriority(int* gid, int* prio); #ifndef ENABLE_MULTIPLE_NODES static bool SyncRepGetSyncLeftTime(XLogRecPtr XactCommitLSN, TimestampTz* leftTime); #endif @@ -89,15 +89,33 @@ static void SyncRepGetOldestSyncRecPtr(XLogRecPtr* receivePtr, XLogRecPtr* write static void SyncRepGetNthLatestSyncRecPtr(XLogRecPtr* receivePtr, XLogRecPtr* writePtr, XLogRecPtr* flushPtr, XLogRecPtr* replayPtr, List* sync_standbys, uint8 nth); - #ifdef USE_ASSERT_CHECKING static bool SyncRepQueueIsOrderedByLSN(int mode); #endif -static List *SyncRepGetSyncStandbysPriority(bool *am_sync, List** catchup_standbys = NULL); -static List *SyncRepGetSyncStandbysQuorum(bool *am_sync, List** catchup_standbys = NULL); +static List *SyncRepGetSyncStandbysPriority(bool *am_sync, int groupid, List** catchup_standbys = NULL); +static List *SyncRepGetSyncStandbysQuorum(bool *am_sync, int groupid, List** catchup_standbys = NULL); +static inline void free_sync_standbys_list(List* sync_standbys); +static bool allow_most_available_sync(const List* sync_standbys); static int cmp_lsn(const void *a, const void *b); +typedef struct TransContext { + /* for global */ + int group_nums; + List *existers; + + /* for each group*/ + SyncRepConfigData* conf; + bool is_star; +} TransContext; +static TransContext* create_transform_context(); +static void bind_transform_context(TransContext *tcxt, SyncRepConfigData *conf); +static void free_transform_context(TransContext *tcxt); +static bool analyze_star_and_num(TransContext *tcxt); +static bool analyze_duplicate_names(TransContext *tcxt); +static bool analyze_cluster_names(TransContext *tcxt); +static List* transform_synchronous_standby_names(TransContext *tcxt); + #define CATCHUP_XLOG_DIFF(ptr1, ptr2, amount) \ XLogRecPtrIsInvalid(ptr1) ? false : (XLByteDifference(ptr2, ptr1) < amount) @@ -109,6 +127,19 @@ static int cmp_lsn(const void *a, const void *b); } \ } while (0) +/* + * sync_standbys_list is a two-dimensional array means sync standby nodes in groups + * The structure is as follows: List[IntList[int, int,...], IntList[int, int, ...], ...] + */ +static inline void free_sync_standbys_list(List* sync_standbys) +{ + ListCell *lc = NULL; + foreach(lc, sync_standbys) { + list_free((List*)lfirst(lc)); + } + list_free(sync_standbys); +} + /* * Determine whether to wait for standby catching up, if requested by user. * @@ -495,15 +526,19 @@ void SyncRepCleanupAtProcExit(void) */ void SyncRepInitConfig(void) { + int group; int priority; /* * Determine if we are a potential sync standby and remember the result * for handling replies from standby. */ - priority = SyncRepGetStandbyPriority(); - if (t_thrd.walsender_cxt.MyWalSnd->sync_standby_priority != priority) { + SyncRepGetStandbyGroupAndPriority(&group, &priority); + if (t_thrd.walsender_cxt.MyWalSnd->sync_standby_group != group || + t_thrd.walsender_cxt.MyWalSnd->sync_standby_priority != priority) { LWLockAcquire(SyncRepLock, LW_EXCLUSIVE); + + t_thrd.walsender_cxt.MyWalSnd->sync_standby_group = group; t_thrd.walsender_cxt.MyWalSnd->sync_standby_priority = priority; /* @@ -513,8 +548,8 @@ void SyncRepInitConfig(void) SyncRepCheckSyncStandbyAlive(); LWLockRelease(SyncRepLock); - ereport(DEBUG1, (errmsg("standby \"%s\" now has synchronous standby priority %d", - u_sess->attr.attr_common.application_name, priority))); + ereport(DEBUG1, (errmsg("standby \"%s\" now has synchronous standby group and priority: %d %d", + u_sess->attr.attr_common.application_name, group, priority))); } } @@ -571,13 +606,15 @@ void SyncRepReleaseWaiters(void) */ if (t_thrd.syncrep_cxt.announce_next_takeover && am_sync) { t_thrd.syncrep_cxt.announce_next_takeover = false; - if (t_thrd.syncrep_cxt.SyncRepConfig->syncrep_method == SYNC_REP_PRIORITY) { - ereport(LOG, (errmsg("standby \"%s\" is now a synchronous standby with priority %d", + if (GetWalsndSyncRepConfig(t_thrd.walsender_cxt.MyWalSnd)->syncrep_method == SYNC_REP_PRIORITY) { + ereport(LOG, (errmsg("standby \"%s\" is now a synchronous standby with group and priority: %d %d", u_sess->attr.attr_common.application_name, + t_thrd.walsender_cxt.MyWalSnd->sync_standby_group, t_thrd.walsender_cxt.MyWalSnd->sync_standby_priority))); } else { - ereport(LOG, (errmsg("standby \"%s\" is now a candidate for quorum synchronous standby", - u_sess->attr.attr_common.application_name))); + ereport(LOG, (errmsg("standby \"%s\" is now a candidate for quorum synchronous standby in group: %d", + u_sess->attr.attr_common.application_name, + t_thrd.walsender_cxt.MyWalSnd->sync_standby_group))); } } @@ -620,6 +657,35 @@ void SyncRepReleaseWaiters(void) (uint32)writePtr, numflush, (uint32)(flushPtr >> 32), (uint32)flushPtr))); } +/* + * In a particular scenario, when most_available_sync is true, primary only wait + * the alive sync standbys, even if the quantity does not meet the configuration + * requirements. + */ +static bool allow_most_available_sync(const List* sync_standbys) +{ + ListCell* lc = NULL; + List* per_group = NIL; + int gid = 0; + int alive = 0; + + foreach(lc, sync_standbys) { + per_group = (List*)lfirst(lc); + if (!t_thrd.walsender_cxt.WalSndCtl->most_available_sync && + list_length(per_group) < t_thrd.syncrep_cxt.SyncRepConfig[gid]->num_sync) + return false; + + alive += list_length(per_group); + gid++; + } + + if (t_thrd.walsender_cxt.WalSndCtl->most_available_sync && alive == 0) { + return false; + } + + return true; +} + /* * Calculate the synced Receive, Write, Flush and Apply positions among sync standbys. * @@ -633,6 +699,8 @@ void SyncRepReleaseWaiters(void) bool SyncRepGetSyncRecPtr(XLogRecPtr *receivePtr, XLogRecPtr *writePtr, XLogRecPtr *flushPtr, XLogRecPtr* replayPtr, bool *am_sync, bool check_am_sync) { List *sync_standbys = NIL; + List *per_group = NIL; + ListCell *lc = NULL; *receivePtr = InvalidXLogRecPtr; *writePtr = InvalidXLogRecPtr; @@ -645,14 +713,12 @@ bool SyncRepGetSyncRecPtr(XLogRecPtr *receivePtr, XLogRecPtr *writePtr, XLogRecP /* * Quick exit if we are not managing a sync standby (or not check for check_am_sync is false) * or there are not enough synchronous standbys. - * but in a particular scenario, when most_available_sync is true, primary only wait the alive sync standbys - * if list_length(sync_standbys) doesn't satisfy t_thrd.syncrep_cxt.SyncRepConfig->num_sync. + * or most_available_sync is working and allow some standby nodes to be missing. */ - if ((!(*am_sync) && check_am_sync) || t_thrd.syncrep_cxt.SyncRepConfig == NULL || - (!t_thrd.walsender_cxt.WalSndCtl->most_available_sync && - list_length(sync_standbys) < t_thrd.syncrep_cxt.SyncRepConfig->num_sync) || - (t_thrd.walsender_cxt.WalSndCtl->most_available_sync && list_length((sync_standbys)) == 0)) { - list_free(sync_standbys); + if ((!(*am_sync) && check_am_sync) || + t_thrd.syncrep_cxt.SyncRepConfig == NULL || + !allow_most_available_sync(sync_standbys)) { + free_sync_standbys_list(sync_standbys); return false; } @@ -669,11 +735,20 @@ bool SyncRepGetSyncRecPtr(XLogRecPtr *receivePtr, XLogRecPtr *writePtr, XLogRecP * we can use SyncRepGetOldestSyncRecPtr() to calculate the synced * positions even in a quorum-based sync replication. */ - if (t_thrd.syncrep_cxt.SyncRepConfig->syncrep_method == SYNC_REP_PRIORITY) { - SyncRepGetOldestSyncRecPtr(receivePtr, writePtr, flushPtr, replayPtr, sync_standbys); - } else { - SyncRepGetNthLatestSyncRecPtr(receivePtr, writePtr, flushPtr, replayPtr, sync_standbys, - t_thrd.syncrep_cxt.SyncRepConfig->num_sync); + int i = 0; + foreach(lc, sync_standbys) { + per_group = (List*)lfirst(lc); + if (per_group == NIL) { + /* do nothing, this group of sync standbys are all offline or no sync standbys. */ + } else if (t_thrd.syncrep_cxt.SyncRepConfig[i]->syncrep_method == SYNC_REP_PRIORITY) { + SyncRepGetOldestSyncRecPtr(receivePtr, writePtr, flushPtr, replayPtr, per_group); + } else { + SyncRepGetNthLatestSyncRecPtr(receivePtr, writePtr, flushPtr, replayPtr, + per_group, t_thrd.syncrep_cxt.SyncRepConfig[i]->num_sync); + } + + list_free(per_group); + i++; } list_free(sync_standbys); @@ -700,7 +775,7 @@ static bool SyncRepGetSyncLeftTime(XLogRecPtr XactCommitLSN, TimestampTz* leftTi /* Skip here if there is at lease one sync standby, or no standby in catchup. */ if (list_length(sync_standbys) > 0 || list_length(catchup_standbys) == 0) { - list_free(sync_standbys); + free_sync_standbys_list(sync_standbys); list_free(catchup_standbys); return false; } @@ -723,7 +798,7 @@ static bool SyncRepGetSyncLeftTime(XLogRecPtr XactCommitLSN, TimestampTz* leftTi } } - list_free(sync_standbys); + free_sync_standbys_list(sync_standbys); list_free(catchup_standbys); return true; } @@ -822,10 +897,14 @@ static void SyncRepGetNthLatestSyncRecPtr(XLogRecPtr* receivePtr, XLogRecPtr* wr } /* Get Nth latest Write, Flush, Apply positions */ - *receivePtr = receive_array[nth - 1]; - *writePtr = write_array[nth - 1]; - *flushPtr = flush_array[nth - 1]; - *replayPtr = apply_array[nth - 1]; + if (XLogRecPtrIsInvalid(*writePtr) || XLByteLE(write_array[nth - 1], *writePtr)) + *writePtr = write_array[nth - 1]; + if (XLogRecPtrIsInvalid(*flushPtr) || XLByteLE(flush_array[nth - 1], *flushPtr)) + *flushPtr = flush_array[nth - 1]; + if (XLogRecPtrIsInvalid(*receivePtr) || XLByteLE(receive_array[nth - 1], *receivePtr)) + *receivePtr = receive_array[nth - 1]; + if (XLogRecPtrIsInvalid(*replayPtr) || XLByteLE(apply_array[nth - 1], *replayPtr)) + *replayPtr = apply_array[nth - 1]; pfree(receive_array); receive_array = NULL; @@ -855,47 +934,56 @@ static int cmp_lsn(const void *a, const void *b) /* * Check if we are in the list of sync standbys, and if so, determine - * priority sequence. Return priority if set, or zero to indicate that + * priority sequence. Return groupid and priority if set, or zero to indicate that * we are not a potential sync standby. * * Compare the parameter SyncRepStandbyNames against the application_name * for this WALSender, or allow any name if we find a wildcard "*". */ -static int SyncRepGetStandbyPriority(void) +static void SyncRepGetStandbyGroupAndPriority(int* gid, int* prio) { const char *standby_name = NULL; int priority; + int group; bool found = false; + *gid = 0; + *prio = 0; + /* * Since synchronous cascade replication is not allowed, we always set the * priority of cascading walsender to zero. */ if (AM_WAL_STANDBY_SENDER) - return 0; + return; if (!SyncStandbysDefined() || t_thrd.syncrep_cxt.SyncRepConfig == NULL || !SyncRepRequested()) - return 0; + return; - standby_name = t_thrd.syncrep_cxt.SyncRepConfig->member_names; - for (priority = 1; priority <= t_thrd.syncrep_cxt.SyncRepConfig->nmembers; priority++) { - if (pg_strcasecmp(standby_name, u_sess->attr.attr_common.application_name) == 0 || - strcmp(standby_name, "*") == 0) { - found = true; - break; + for (group = 0; group < t_thrd.syncrep_cxt.SyncRepConfigGroups && !found; group++) { + standby_name = t_thrd.syncrep_cxt.SyncRepConfig[group]->member_names; + for (priority = 1; priority <= t_thrd.syncrep_cxt.SyncRepConfig[group]->nmembers; priority++) { + if (pg_strcasecmp(standby_name, u_sess->attr.attr_common.application_name) == 0 || + strcmp(standby_name, "*") == 0) { + Assert(!(group > 1 && strcmp(standby_name, "*") == 0)); + found = true; + break; + } + standby_name += strlen(standby_name) + 1; } - standby_name += strlen(standby_name) + 1; } if (!found) { - return 0; + return; } /* * In quorum-based sync replication, all the standbys in the list * have the same priority, one. */ - return (t_thrd.syncrep_cxt.SyncRepConfig->syncrep_method == SYNC_REP_PRIORITY) ? priority : 1; + *gid = group - 1; + *prio = (t_thrd.syncrep_cxt.SyncRepConfig[group - 1]->syncrep_method == SYNC_REP_PRIORITY) ? priority : 1; + return; } /* @@ -1116,9 +1204,16 @@ List *SyncRepGetSyncStandbys(bool *am_sync, List** catchup_standbys) if (t_thrd.syncrep_cxt.SyncRepConfig == NULL) return NIL; - return (t_thrd.syncrep_cxt.SyncRepConfig->syncrep_method == SYNC_REP_PRIORITY) - ? SyncRepGetSyncStandbysPriority(am_sync, catchup_standbys) - : SyncRepGetSyncStandbysQuorum(am_sync, catchup_standbys); + List* results = NIL; + for(int i = 0; i < t_thrd.syncrep_cxt.SyncRepConfigGroups; i++) { + if (t_thrd.syncrep_cxt.SyncRepConfig[i]->syncrep_method == SYNC_REP_PRIORITY) { + results = lappend(results, SyncRepGetSyncStandbysPriority(am_sync, i,catchup_standbys)); + } else { + results = lappend(results, SyncRepGetSyncStandbysQuorum(am_sync, i, catchup_standbys)); + } + } + + return results; } /* @@ -1131,13 +1226,13 @@ List *SyncRepGetSyncStandbys(bool *am_sync, List** catchup_standbys) * On return, *am_sync is set to true if this walsender is connecting to * sync standby. Otherwise it's set to false. */ -static List *SyncRepGetSyncStandbysQuorum(bool *am_sync, List** catchup_standbys) +static List *SyncRepGetSyncStandbysQuorum(bool *am_sync, int groupid, List **catchup_standbys) { List *result = NIL; int i; volatile WalSnd *walsnd = NULL; /* Use volatile pointer to prevent code rearrangement */ - Assert(t_thrd.syncrep_cxt.SyncRepConfig->syncrep_method == SYNC_REP_QUORUM); + Assert(t_thrd.syncrep_cxt.SyncRepConfig[groupid]->syncrep_method == SYNC_REP_QUORUM); for (i = 0; i < g_instance.attr.attr_storage.max_wal_senders; i++) { walsnd = &t_thrd.walsender_cxt.WalSndCtl->walsnds[i]; @@ -1147,7 +1242,7 @@ static List *SyncRepGetSyncStandbysQuorum(bool *am_sync, List** catchup_standbys continue; /* Must be synchronous */ - if (walsnd->sync_standby_priority == 0) + if (walsnd->sync_standby_priority == 0 || walsnd->sync_standby_group != groupid) continue; if ((walsnd->state == WALSNDSTATE_CATCHUP || walsnd->peer_state == CATCHUP_STATE) && @@ -1188,7 +1283,7 @@ static List *SyncRepGetSyncStandbysQuorum(bool *am_sync, List** catchup_standbys * On return, *am_sync is set to true if this walsender is connecting to * sync standby. Otherwise it's set to false. */ -static List *SyncRepGetSyncStandbysPriority(bool *am_sync, List** catchup_standbys) +static List *SyncRepGetSyncStandbysPriority(bool *am_sync, int groupid, List **catchup_standbys) { List *result = NIL; List *pending = NIL; @@ -1200,9 +1295,9 @@ static List *SyncRepGetSyncStandbysPriority(bool *am_sync, List** catchup_standb bool am_in_pending = false; volatile WalSnd *walsnd = NULL; /* Use volatile pointer to prevent code rearrangement */ - Assert(t_thrd.syncrep_cxt.SyncRepConfig->syncrep_method == SYNC_REP_PRIORITY); + Assert(t_thrd.syncrep_cxt.SyncRepConfig[groupid]->syncrep_method == SYNC_REP_PRIORITY); - lowest_priority = t_thrd.syncrep_cxt.SyncRepConfig->nmembers; + lowest_priority = t_thrd.syncrep_cxt.SyncRepConfig[groupid]->nmembers; next_highest_priority = lowest_priority + 1; /* @@ -1219,7 +1314,7 @@ static List *SyncRepGetSyncStandbysPriority(bool *am_sync, List** catchup_standb /* Must be synchronous */ this_priority = walsnd->sync_standby_priority; - if (this_priority == 0) + if (this_priority == 0 || walsnd->sync_standby_group != groupid) continue; if ((walsnd->state == WALSNDSTATE_CATCHUP || walsnd->peer_state == CATCHUP_STATE) && @@ -1244,7 +1339,7 @@ static List *SyncRepGetSyncStandbysPriority(bool *am_sync, List** catchup_standb result = lappend_int(result, i); if (am_sync != NULL && walsnd == t_thrd.walsender_cxt.MyWalSnd) *am_sync = true; - if (list_length(result) == t_thrd.syncrep_cxt.SyncRepConfig->num_sync) { + if (list_length(result) == t_thrd.syncrep_cxt.SyncRepConfig[groupid]->num_sync) { list_free(pending); return result; /* Exit if got enough sync standbys */ } @@ -1269,7 +1364,7 @@ static List *SyncRepGetSyncStandbysPriority(bool *am_sync, List** catchup_standb * Consider all pending standbys as sync if the number of them plus * already-found sync ones is lower than the configuration requests. */ - if (list_length(result) + list_length(pending) <= t_thrd.syncrep_cxt.SyncRepConfig->num_sync) { + if (list_length(result) + list_length(pending) <= t_thrd.syncrep_cxt.SyncRepConfig[groupid]->num_sync) { bool needfree = (result != NIL && pending != NIL); /* @@ -1315,7 +1410,7 @@ static List *SyncRepGetSyncStandbysPriority(bool *am_sync, List** catchup_standb * starts because we know that the list has enough elements to * reach SyncRepConfig->num_sync. */ - if (list_length(result) == t_thrd.syncrep_cxt.SyncRepConfig->num_sync) { + if (list_length(result) == t_thrd.syncrep_cxt.SyncRepConfig[groupid]->num_sync) { list_free(pending); return result; /* Exit if got enough sync standbys */ } @@ -1489,6 +1584,166 @@ int init_gauss_cluster_config(void) return 0; } +/* + * ================================================================= + * Analyze and transform GUC param synchronous_standby_names content + * Create result in session MEMORY_CONTEXT_STORAGE session. + * ================================================================= + */ +static TransContext* create_transform_context() +{ + TransContext* tcxt = (TransContext*)palloc(sizeof(TransContext)); + + tcxt->group_nums = list_length(t_thrd.syncrepgram_cxt.syncrep_parse_result); + Assert(tcxt->group_nums > 0); + tcxt->existers = NIL; + tcxt->conf = NULL; + tcxt->is_star = false; + + return tcxt; +} + +static void bind_transform_context(TransContext *tcxt, SyncRepConfigData *conf) +{ + tcxt->conf = conf; + tcxt->is_star = false; +} + +static void free_transform_context(TransContext *tcxt) +{ + list_free(tcxt->existers); + pfree(tcxt); +} + +static bool analyze_star_and_num(TransContext *tcxt) +{ + SyncRepConfigData *conf = tcxt->conf; + + char *p = conf->member_names; + for (int i = 0; i < conf->nmembers; i++) { + if (strcmp(p, "*") != 0) { + p += strlen(p) + 1; + continue; + } + + if (tcxt->group_nums > 1) { + GUC_check_errdetail("'*' is not support when sync standby strategy is combinated."); + return false; + } + if (conf->nmembers > 1) { + GUC_check_errdetail("Please use the '*' separately, otherwise, it is semantic ambiguity."); + return false; + } + tcxt->is_star = true; + } + + if (!tcxt->is_star && conf->num_sync > conf->nmembers) { + GUC_check_errdetail("The sync number must less or equals to the number of standby node names."); + return false; + } + + return true; +} + +static bool analyze_duplicate_names(TransContext *tcxt) +{ + if (tcxt->is_star) + return true; + + SyncRepConfigData *conf = tcxt->conf; + char *p = conf->member_names; + char *exister = NULL; + + ListCell *lc = NULL; + for (int i = 0; i < conf->nmembers; i++) { + foreach(lc, tcxt->existers) { + exister = (char*)lfirst(lc); + if (pg_strcasecmp(p, exister) == 0) { + GUC_check_errdetail("Duplicate standby node name: %s", p); + return false; + } + } + tcxt->existers = lappend(tcxt->existers, p); + p += strlen(p) + 1; + } + return true; +} + +static bool analyze_cluster_names(TransContext *tcxt) +{ + if (tcxt->is_star) + return true; + + SyncRepConfigData *conf = tcxt->conf; + + if (t_thrd.role != WORKER || + strcmp(u_sess->attr.attr_common.application_name, "gsql") != 0 || + !has_static_config()) { + return true; + } + + if (0 != init_gauss_cluster_config()) { + return true; + } + + char* p = conf->member_names; + for (int i = 1; i <= conf->nmembers; i++) { + if (!CheckDataNameValue(p, t_thrd.proc_cxt.DataDir)) { + GUC_check_errdetail("invalid standby name: %s", p); + return false; + } + p += strlen(p) + 1; + } + return true; +} + +static List* transform_synchronous_standby_names(TransContext* tcxt) +{ + errno_t rc = EOK; + ListCell* lc = NULL; + List* results = NIL; + + if (tcxt->group_nums > SYNC_REP_MAX_GROUPS) { + GUC_check_errdetail("Too much groups. Please no more than %d.", SYNC_REP_MAX_GROUPS); + return NIL; + } + + foreach(lc, t_thrd.syncrepgram_cxt.syncrep_parse_result) { + SyncRepConfigData* conf = (SyncRepConfigData*)lfirst(lc); + bind_transform_context(tcxt, conf); + + if (conf->syncrep_method == SYNC_REP_PRIORITY && tcxt->group_nums > 1) { + GUC_check_errdetail("FIRST rule is not support when sync standby strategy is combinated."); + return NIL; + } + + if (!analyze_star_and_num(tcxt)) { + return NIL; + } + + if (!analyze_duplicate_names(tcxt)) { + return NIL; + } + + if (!analyze_cluster_names(tcxt)) { + return NIL; + } + } + + // append into results + MemoryContext old_context = MemoryContextSwitchTo(SESS_GET_MEM_CXT_GROUP(MEMORY_CONTEXT_STORAGE)); + foreach(lc, t_thrd.syncrepgram_cxt.syncrep_parse_result) { + SyncRepConfigData* conf = (SyncRepConfigData*)lfirst(lc); + SyncRepConfigData* result = (SyncRepConfigData*)palloc(conf->config_size); + rc = memcpy_s(result, conf->config_size, conf, conf->config_size); + securec_check(rc, "", ""); + results = lappend(results, result); + } + (void)MemoryContextSwitchTo(old_context); + + return results; +} + /* * =========================================================== * Synchronous Replication functions executed by any process @@ -1496,70 +1751,42 @@ int init_gauss_cluster_config(void) */ bool check_synchronous_standby_names(char **newval, void **extra, GucSource source) { - errno_t rc = EOK; - if (*newval != NULL && (*newval)[0] != '\0') { int parse_rc; - SyncRepConfigData *pconf = NULL; syncrep_scanner_yyscan_t yyscanner; - char* data_dir = t_thrd.proc_cxt.DataDir; - char* p = NULL; + List* pconf = NIL; /* Reset communication variables to ensure a fresh start */ - t_thrd.syncrepgram_cxt.syncrep_parse_result = NULL; + t_thrd.syncrepgram_cxt.syncrep_parse_result = NIL; /* Parse the synchronous_standby_names string */ yyscanner = syncrep_scanner_init(*newval); parse_rc = syncrep_yyparse(yyscanner); syncrep_scanner_finish(yyscanner); - if (parse_rc != 0 || t_thrd.syncrepgram_cxt.syncrep_parse_result == NULL) { + if (parse_rc != 0 || t_thrd.syncrepgram_cxt.syncrep_parse_result == NIL) { GUC_check_errcode(ERRCODE_SYNTAX_ERROR); GUC_check_errdetail("synchronous_standby_names parser failed"); return false; } - /* GUC extra value must be malloc'd, not palloc'd */ - pconf = (SyncRepConfigData *)MemoryContextAlloc(SESS_GET_MEM_CXT_GROUP(MEMORY_CONTEXT_STORAGE), - t_thrd.syncrepgram_cxt.syncrep_parse_result->config_size); - if (pconf == NULL) - return false; - rc = memcpy_s(pconf, t_thrd.syncrepgram_cxt.syncrep_parse_result->config_size, - t_thrd.syncrepgram_cxt.syncrep_parse_result, - t_thrd.syncrepgram_cxt.syncrep_parse_result->config_size); - securec_check(rc, "", ""); + /* + * analyze and transform synchronous_standby_names content, + * make result in session MEMORY_CONTEXT_STORAGE. + */ + TransContext* tcxt = create_transform_context(); + pconf = transform_synchronous_standby_names(tcxt); + free_transform_context(tcxt); + tcxt = NULL; - if (strcmp(pconf->member_names, "*") == 0) { - goto pass; - } - - if (pconf->num_sync > pconf->nmembers) { - // The sync number must less or equals to the number of standby node names. + if (pconf == NIL) { return false; } - /* get current cluster information from cluster_staic_config */ - if (t_thrd.role == WORKER && strcmp(u_sess->attr.attr_common.application_name, "gsql") == 0 - && has_static_config()) { - if (0 != init_gauss_cluster_config()) { - goto pass; - } - } else { - goto pass; - } - p = pconf->member_names; - for (int i = 1; i <= pconf->nmembers; i++) { - if (!CheckDataNameValue(p, data_dir)) { - return false; - } - p += strlen(p) + 1; - } - -pass: *extra = (void *)pconf; if (t_thrd.syncrepgram_cxt.syncrep_parse_result) { - pfree(t_thrd.syncrepgram_cxt.syncrep_parse_result); - t_thrd.syncrepgram_cxt.syncrep_parse_result = NULL; + list_free_deep(t_thrd.syncrepgram_cxt.syncrep_parse_result); + t_thrd.syncrepgram_cxt.syncrep_parse_result = NIL; } /* @@ -1582,16 +1809,35 @@ void assign_synchronous_standby_names(const char *newval, void *extra) * it should be safe to know the latest rep config ASAP for all sessions. * If this assumption no longer holds, please move it to session level. */ - pfree_ext(t_thrd.syncrep_cxt.SyncRepConfig); - - if (extra != NULL) { - SyncRepConfigData *pconf = (SyncRepConfigData *)extra; - errno_t rc = EOK; - t_thrd.syncrep_cxt.SyncRepConfig = (SyncRepConfigData *)MemoryContextAlloc( - THREAD_GET_MEM_CXT_GROUP(MEMORY_CONTEXT_STORAGE), pconf->config_size); - rc = memcpy_s(t_thrd.syncrep_cxt.SyncRepConfig, pconf->config_size, pconf, pconf->config_size); - securec_check(rc, "", ""); + for (int i = 0; i < t_thrd.syncrep_cxt.SyncRepConfigGroups; i++) { + pfree_ext(t_thrd.syncrep_cxt.SyncRepConfig[i]); } + pfree_ext(t_thrd.syncrep_cxt.SyncRepConfig); + t_thrd.syncrep_cxt.SyncRepConfigGroups = 0; + + if (extra == NULL) + return; + + List* pconf_list = (List*)extra; + ListCell* lc = NULL; + int i = 0; + errno_t rc = EOK; + MemoryContext old_context = MemoryContextSwitchTo(THREAD_GET_MEM_CXT_GROUP(MEMORY_CONTEXT_STORAGE)); + + t_thrd.syncrep_cxt.SyncRepConfigGroups = list_length(pconf_list); + t_thrd.syncrep_cxt.SyncRepConfig = (SyncRepConfigData **)palloc( + t_thrd.syncrep_cxt.SyncRepConfigGroups * sizeof(SyncRepConfigData*)); + + foreach(lc, pconf_list) { + SyncRepConfigData* pconf = (SyncRepConfigData*)lfirst(lc); + + t_thrd.syncrep_cxt.SyncRepConfig[i] = (SyncRepConfigData*)palloc(pconf->config_size); + rc = memcpy_s(t_thrd.syncrep_cxt.SyncRepConfig[i], pconf->config_size, pconf, pconf->config_size); + securec_check(rc, "", ""); + + i++; + } + (void)MemoryContextSwitchTo(old_context); } void assign_synchronous_commit(int newval, void *extra) diff --git a/src/gausskernel/storage/replication/syncrep_gram.y b/src/gausskernel/storage/replication/syncrep_gram.y index ad5cb4dc..211108be 100644 --- a/src/gausskernel/storage/replication/syncrep_gram.y +++ b/src/gausskernel/storage/replication/syncrep_gram.y @@ -66,8 +66,8 @@ static void syncrep_yyerror(YYLTYPE *yylloc, %token NAME NUM JUNK ANY FIRST -%type result standby_config -%type standby_list +%type standby_config_simplify standby_config_complete +%type result standby_config standby_config_combination standby_list %type standby_name %start result @@ -78,9 +78,22 @@ result: ; standby_config: + standby_config_simplify { $$ = list_make1($1); } + | standby_config_combination { $$ = $1; } + ; + +standby_config_combination: + standby_config_complete { $$ = list_make1($1); } + | standby_config_combination ',' standby_config_complete { $$ = lappend($1, $3); } + ; + +standby_config_simplify: standby_list { $$ = create_syncrep_config("1", $1, SYNC_REP_PRIORITY); } | NUM '(' standby_list ')' { $$ = create_syncrep_config($1, $3, SYNC_REP_PRIORITY); pfree($1); } - | ANY NUM '(' standby_list ')' { $$ = create_syncrep_config($2, $4, SYNC_REP_QUORUM); pfree($2); } + ; + +standby_config_complete: + ANY NUM '(' standby_list ')' { $$ = create_syncrep_config($2, $4, SYNC_REP_QUORUM); pfree($2); } | FIRST NUM '(' standby_list ')' { $$ = create_syncrep_config($2, $4, SYNC_REP_PRIORITY); pfree($2); } ; diff --git a/src/gausskernel/storage/replication/walsender.cpp b/src/gausskernel/storage/replication/walsender.cpp index abfafc0d..151182ab 100644 --- a/src/gausskernel/storage/replication/walsender.cpp +++ b/src/gausskernel/storage/replication/walsender.cpp @@ -3638,6 +3638,7 @@ static void InitWalSnd(void) securec_check(rc, "", ""); rc = memset_s((void *)&walsnd->wal_sender_channel, sizeof(ReplConnInfo), 0, sizeof(ReplConnInfo)); securec_check(rc, "", ""); + walsnd->sync_standby_group = 0; walsnd->sync_standby_priority = 0; walsnd->index = i; walsnd->log_ctrl.sleep_time = 0; @@ -4411,8 +4412,7 @@ bool WalSndInProgress(int type) bool WalSndQuorumInProgress(int type) { int i; - int num = 0; - int num_sync = t_thrd.syncrep_cxt.SyncRepConfig->num_sync; + int* nums = (int*)palloc0(t_thrd.syncrep_cxt.SyncRepConfigGroups * sizeof(int)); for (i = 0; i < g_instance.attr.attr_storage.max_wal_senders; i++) { /* use volatile pointer to prevent code rearrangement */ @@ -4420,15 +4420,19 @@ bool WalSndQuorumInProgress(int type) SpinLockAcquire(&walsnd->mutex); if (walsnd->pid != 0 && walsnd->pid != t_thrd.proc_cxt.MyProcPid && ((walsnd->sendRole & type) == walsnd->sendRole)) { - num++; + nums[walsnd->sync_standby_group]++; } SpinLockRelease(&walsnd->mutex); } - if (num_sync <= num) { - return true; - } else { - return false; + + for (i = 0; i < t_thrd.syncrep_cxt.SyncRepConfigGroups; i++) { + if (t_thrd.syncrep_cxt.SyncRepConfig[i]->num_sync > nums[i]) { + pfree(nums); + return false; + } } + pfree(nums); + return true; } /* check if there is all wal sender in type status. */ @@ -4601,7 +4605,7 @@ static void set_xlog_location(ServerMode local_role, XLogRecPtr* sndWrite, XLogR */ Datum pg_stat_get_wal_senders(PG_FUNCTION_ARGS) { -#define PG_STAT_GET_WAL_SENDERS_COLS 21 +#define PG_STAT_GET_WAL_SENDERS_COLS 22 TupleDesc tupdesc; Tuplestorestate *tupstore = NULL; @@ -4609,6 +4613,7 @@ Datum pg_stat_get_wal_senders(PG_FUNCTION_ARGS) int i = 0; volatile HaShmemData *hashmdata = t_thrd.postmaster_cxt.HaShmData; List *sync_standbys = NIL; + ListCell *lc = NULL; tupstore = BuildTupleResult(fcinfo, &tupdesc); @@ -4667,6 +4672,7 @@ Datum pg_stat_get_wal_senders(PG_FUNCTION_ARGS) int j = 0; errno_t rc = 0; int ret = 0; + int group = 0; int priority = 0; SpinLockAcquire(&hashmdata->mutex); @@ -4699,8 +4705,10 @@ Datum pg_stat_get_wal_senders(PG_FUNCTION_ARGS) syncStart = walsnd->syncPercentCountStart; catchup_time[0] = walsnd->catchupTime[0]; catchup_time[1] = walsnd->catchupTime[1]; - if (IS_DN_MULTI_STANDYS_MODE()) + if (IS_DN_MULTI_STANDYS_MODE()) { + group = walsnd->sync_standby_group; priority = walsnd->sync_standby_priority; + } SpinLockRelease(&walsnd->mutex); set_xlog_location(local_role, &sndWrite, &sndFlush, &sndReplay); @@ -4822,9 +4830,11 @@ Datum pg_stat_get_wal_senders(PG_FUNCTION_ARGS) /* sync_state and sync_prority */ if (!SyncRepRequested()) { values[j++] = CStringGetTextDatum("Async"); + nulls[j++] = true; values[j++] = Int32GetDatum(0); } else { values[j++] = CStringGetTextDatum("Sync"); + nulls[j++] = true; values[j++] = Int32GetDatum(sync_priority[i]); } } else { @@ -4844,14 +4854,19 @@ Datum pg_stat_get_wal_senders(PG_FUNCTION_ARGS) * basically useless to report "sync" or "potential" as their sync * states. We report just "quorum" for them. */ - if (priority == 0) + if (priority == 0) { values[j++] = CStringGetTextDatum("Async"); - else if (list_member_int(sync_standbys, i)) { - values[j++] = t_thrd.syncrep_cxt.SyncRepConfig->syncrep_method == SYNC_REP_PRIORITY + nulls[j++] = true; + } else if (list_member_int((List*)list_nth(sync_standbys, group), i)) { + values[j++] = GetWalsndSyncRepConfig(walsnd)->syncrep_method == SYNC_REP_PRIORITY ? CStringGetTextDatum("Sync") : CStringGetTextDatum("Quorum"); - } else + values[j++] = Int32GetDatum(group); + } else { values[j++] = CStringGetTextDatum("Potential"); + values[j++] = Int32GetDatum(group); + } + values[j++] = Int32GetDatum(priority); } @@ -4869,6 +4884,9 @@ Datum pg_stat_get_wal_senders(PG_FUNCTION_ARGS) tuplestore_putvalues(tupstore, tupdesc, values, nulls); } + foreach(lc, sync_standbys) { + list_free((List*)lfirst(lc)); + } list_free(sync_standbys); if (sync_priority != NULL) { pfree(sync_priority); diff --git a/src/include/catalog/upgrade_sql/rollback_catalog_maindb/rollback_catalog_maindb_92_301.sql b/src/include/catalog/upgrade_sql/rollback_catalog_maindb/rollback_catalog_maindb_92_301.sql new file mode 100644 index 00000000..1461f7fe --- /dev/null +++ b/src/include/catalog/upgrade_sql/rollback_catalog_maindb/rollback_catalog_maindb_92_301.sql @@ -0,0 +1,29 @@ +DROP FUNCTION IF EXISTS pg_catalog.pg_stat_get_wal_senders() CASCADE; +SET LOCAL inplace_upgrade_next_system_object_oids = IUO_PROC, 3099; +CREATE FUNCTION pg_catalog.pg_stat_get_wal_senders( + OUT pid bigint, + OUT sender_pid integer, + OUT local_role text, + OUT peer_role text, + OUT peer_state text, + OUT state text, + OUT catchup_start timestamp with time zone, + OUT catchup_end timestamp with time zone, + OUT sender_sent_location text, + OUT sender_write_location text, + OUT sender_flush_location text, + OUT sender_replay_location text, + OUT receiver_received_location text, + OUT receiver_write_location text, + OUT receiver_flush_location text, + OUT receiver_replay_location text, + OUT sync_percent text, + OUT sync_state text, + OUT sync_priority integer, + OUT sync_most_available text, + OUT channel text +) RETURNS SETOF record +STABLE NOT FENCED NOT SHIPPABLE ROWS 10 +LANGUAGE internal AS 'pg_stat_get_wal_senders'; + +SET LOCAL inplace_upgrade_next_system_object_oids = IUO_PROC, 0; \ No newline at end of file diff --git a/src/include/catalog/upgrade_sql/rollback_catalog_otherdb/rollback_catalog_otherdb_92_301.sql b/src/include/catalog/upgrade_sql/rollback_catalog_otherdb/rollback_catalog_otherdb_92_301.sql new file mode 100644 index 00000000..1461f7fe --- /dev/null +++ b/src/include/catalog/upgrade_sql/rollback_catalog_otherdb/rollback_catalog_otherdb_92_301.sql @@ -0,0 +1,29 @@ +DROP FUNCTION IF EXISTS pg_catalog.pg_stat_get_wal_senders() CASCADE; +SET LOCAL inplace_upgrade_next_system_object_oids = IUO_PROC, 3099; +CREATE FUNCTION pg_catalog.pg_stat_get_wal_senders( + OUT pid bigint, + OUT sender_pid integer, + OUT local_role text, + OUT peer_role text, + OUT peer_state text, + OUT state text, + OUT catchup_start timestamp with time zone, + OUT catchup_end timestamp with time zone, + OUT sender_sent_location text, + OUT sender_write_location text, + OUT sender_flush_location text, + OUT sender_replay_location text, + OUT receiver_received_location text, + OUT receiver_write_location text, + OUT receiver_flush_location text, + OUT receiver_replay_location text, + OUT sync_percent text, + OUT sync_state text, + OUT sync_priority integer, + OUT sync_most_available text, + OUT channel text +) RETURNS SETOF record +STABLE NOT FENCED NOT SHIPPABLE ROWS 10 +LANGUAGE internal AS 'pg_stat_get_wal_senders'; + +SET LOCAL inplace_upgrade_next_system_object_oids = IUO_PROC, 0; \ No newline at end of file diff --git a/src/include/catalog/upgrade_sql/upgrade_catalog_maindb/upgrade-post_catalog_maindb_92_301.sql b/src/include/catalog/upgrade_sql/upgrade_catalog_maindb/upgrade-post_catalog_maindb_92_301.sql new file mode 100644 index 00000000..6324361e --- /dev/null +++ b/src/include/catalog/upgrade_sql/upgrade_catalog_maindb/upgrade-post_catalog_maindb_92_301.sql @@ -0,0 +1,30 @@ +DROP FUNCTION IF EXISTS pg_catalog.pg_stat_get_wal_senders() CASCADE; +SET LOCAL inplace_upgrade_next_system_object_oids = IUO_PROC, 3099; +CREATE FUNCTION pg_catalog.pg_stat_get_wal_senders( + OUT pid bigint, + OUT sender_pid integer, + OUT local_role text, + OUT peer_role text, + OUT peer_state text, + OUT state text, + OUT catchup_start timestamp with time zone, + OUT catchup_end timestamp with time zone, + OUT sender_sent_location text, + OUT sender_write_location text, + OUT sender_flush_location text, + OUT sender_replay_location text, + OUT receiver_received_location text, + OUT receiver_write_location text, + OUT receiver_flush_location text, + OUT receiver_replay_location text, + OUT sync_percent text, + OUT sync_state text, + OUT sync_group integer, + OUT sync_priority integer, + OUT sync_most_available text, + OUT channel text +) RETURNS SETOF record +STABLE NOT FENCED NOT SHIPPABLE ROWS 10 +LANGUAGE internal AS 'pg_stat_get_wal_senders'; + +SET LOCAL inplace_upgrade_next_system_object_oids = IUO_PROC, 0; \ No newline at end of file diff --git a/src/include/catalog/upgrade_sql/upgrade_catalog_otherdb/upgrade-post_catalog_otherdb_92_301.sql b/src/include/catalog/upgrade_sql/upgrade_catalog_otherdb/upgrade-post_catalog_otherdb_92_301.sql new file mode 100644 index 00000000..6324361e --- /dev/null +++ b/src/include/catalog/upgrade_sql/upgrade_catalog_otherdb/upgrade-post_catalog_otherdb_92_301.sql @@ -0,0 +1,30 @@ +DROP FUNCTION IF EXISTS pg_catalog.pg_stat_get_wal_senders() CASCADE; +SET LOCAL inplace_upgrade_next_system_object_oids = IUO_PROC, 3099; +CREATE FUNCTION pg_catalog.pg_stat_get_wal_senders( + OUT pid bigint, + OUT sender_pid integer, + OUT local_role text, + OUT peer_role text, + OUT peer_state text, + OUT state text, + OUT catchup_start timestamp with time zone, + OUT catchup_end timestamp with time zone, + OUT sender_sent_location text, + OUT sender_write_location text, + OUT sender_flush_location text, + OUT sender_replay_location text, + OUT receiver_received_location text, + OUT receiver_write_location text, + OUT receiver_flush_location text, + OUT receiver_replay_location text, + OUT sync_percent text, + OUT sync_state text, + OUT sync_group integer, + OUT sync_priority integer, + OUT sync_most_available text, + OUT channel text +) RETURNS SETOF record +STABLE NOT FENCED NOT SHIPPABLE ROWS 10 +LANGUAGE internal AS 'pg_stat_get_wal_senders'; + +SET LOCAL inplace_upgrade_next_system_object_oids = IUO_PROC, 0; \ No newline at end of file diff --git a/src/include/knl/knl_thread.h b/src/include/knl/knl_thread.h index 80057fe0..b65483e6 100644 --- a/src/include/knl/knl_thread.h +++ b/src/include/knl/knl_thread.h @@ -2220,7 +2220,7 @@ typedef struct knl_t_replscanner_context { typedef struct knl_t_syncrepgram_context { /* Result of parsing is returned in one of these two variables */ - struct SyncRepConfigData* syncrep_parse_result; + List* syncrep_parse_result; } knl_t_syncrepgram_context; typedef struct knl_t_syncrepscanner_context { @@ -2230,7 +2230,8 @@ typedef struct knl_t_syncrepscanner_context { } knl_t_syncrepscanner_context; typedef struct knl_t_syncrep_context { - struct SyncRepConfigData* SyncRepConfig; + struct SyncRepConfigData** SyncRepConfig; + int SyncRepConfigGroups; bool announce_next_takeover; } knl_t_syncrep_context; diff --git a/src/include/replication/syncrep.h b/src/include/replication/syncrep.h index 2ecadff2..4c5a45f9 100644 --- a/src/include/replication/syncrep.h +++ b/src/include/replication/syncrep.h @@ -38,11 +38,16 @@ #define SYNC_REP_PRIORITY 0 #define SYNC_REP_QUORUM 1 +#define SYNC_REP_MAX_GROUPS 256 + extern volatile bool most_available_sync; #define SyncStandbysDefined() \ (u_sess->attr.attr_storage.SyncRepStandbyNames != NULL && u_sess->attr.attr_storage.SyncRepStandbyNames[0] != '\0') +#define GetWalsndSyncRepConfig(walsnder) \ + (t_thrd.syncrep_cxt.SyncRepConfig[(walsnder)->sync_standby_group]) + /* * Struct for the configuration of synchronous replication. * diff --git a/src/include/replication/walsender_private.h b/src/include/replication/walsender_private.h index a097fb21..9d865b21 100644 --- a/src/include/replication/walsender_private.h +++ b/src/include/replication/walsender_private.h @@ -109,10 +109,11 @@ typedef struct WalSnd { Latch latch; /* - * The priority order of the standby managed by this WALSender, as listed - * in synchronous_standby_names, or 0 if not-listed. Protected by - * SyncRepLock. + * The strategy group and priority order of the standby managed by this WALSender, + * as listed in synchronous_standby_names, or 0 if not-listed. + * Protected by SyncRepLock. */ + uint8 sync_standby_group; int sync_standby_priority; int index; XLogRecPtr arch_task_lsn; diff --git a/src/test/regress/expected/sync_standy_names.out b/src/test/regress/expected/sync_standy_names.out new file mode 100644 index 00000000..69999272 --- /dev/null +++ b/src/test/regress/expected/sync_standy_names.out @@ -0,0 +1,150 @@ +show synchronous_standby_names; + synchronous_standby_names +--------------------------- + * +(1 row) + +-- single gram +alter system set synchronous_standby_names = '*'; -- suc +alter system set synchronous_standby_names = 'd1, *'; -- err, semantic ambiguity +ERROR: invalid value for parameter "synchronous_standby_names": "d1, *" +DETAIL: Please use the '*' separately, otherwise, it is semantic ambiguity. +alter system set synchronous_standby_names = '*, *'; -- err, semantic ambiguity +ERROR: invalid value for parameter "synchronous_standby_names": "*, *" +DETAIL: Please use the '*' separately, otherwise, it is semantic ambiguity. +alter system set synchronous_standby_names = 'd1'; -- suc +alter system set synchronous_standby_names = 'd1, d2, d3, d4'; -- suc +alter system set synchronous_standby_names = 'd1, d2, d1, d4'; -- err, duplicate name +ERROR: invalid value for parameter "synchronous_standby_names": "d1, d2, d1, d4" +DETAIL: Duplicate standby node name: d1 +alter system set synchronous_standby_names = '2 (*)'; -- suc +alter system set synchronous_standby_names = '2 (d1, *)'; -- err, semantic ambiguity +ERROR: invalid value for parameter "synchronous_standby_names": "2 (d1, *)" +DETAIL: Please use the '*' separately, otherwise, it is semantic ambiguity. +alter system set synchronous_standby_names = '2 (*, *)'; -- err, semantic ambiguity +ERROR: invalid value for parameter "synchronous_standby_names": "2 (*, *)" +DETAIL: Please use the '*' separately, otherwise, it is semantic ambiguity. +alter system set synchronous_standby_names = '2 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = '2 (d1, d2, d1, d4)'; -- err, duplicate name +ERROR: invalid value for parameter "synchronous_standby_names": "2 (d1, d2, d1, d4)" +DETAIL: Duplicate standby node name: d1 +alter system set synchronous_standby_names = '5 (d1, d2, d3, d4)'; -- err, requriement more than have +ERROR: invalid value for parameter "synchronous_standby_names": "5 (d1, d2, d3, d4)" +DETAIL: The sync number must less or equals to the number of standby node names. +alter system set synchronous_standby_names = '0 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = '-1 (d1, d2, d3, d4)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "-1 (d1, d2, d3, d4)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = 'first 2 (*)'; -- suc +alter system set synchronous_standby_names = 'first 2 (d1, *)'; -- err, semantic ambiguity +ERROR: invalid value for parameter "synchronous_standby_names": "first 2 (d1, *)" +DETAIL: Please use the '*' separately, otherwise, it is semantic ambiguity. +alter system set synchronous_standby_names = 'first 2 (*, *)'; -- err, semantic ambiguity +ERROR: invalid value for parameter "synchronous_standby_names": "first 2 (*, *)" +DETAIL: Please use the '*' separately, otherwise, it is semantic ambiguity. +alter system set synchronous_standby_names = 'first 2 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = 'first 2 (d1, d2, d1, d4)'; -- err, duplicate name +ERROR: invalid value for parameter "synchronous_standby_names": "first 2 (d1, d2, d1, d4)" +DETAIL: Duplicate standby node name: d1 +alter system set synchronous_standby_names = 'first 5 (d1, d2, d3, d4)'; -- err, requriement more than have +ERROR: invalid value for parameter "synchronous_standby_names": "first 5 (d1, d2, d3, d4)" +DETAIL: The sync number must less or equals to the number of standby node names. +alter system set synchronous_standby_names = 'first 0 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = 'first -1 (d1, d2, d3, d4)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "first -1 (d1, d2, d3, d4)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = 'any 2 (*)'; -- suc +alter system set synchronous_standby_names = 'any 2 (d1, *'; -- err, semantic ambiguity +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d1, *" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = 'any 2 (*, *)'; -- err, semantic ambiguity +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (*, *)" +DETAIL: Please use the '*' separately, otherwise, it is semantic ambiguity. +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = 'any 2 (d1, d2, d1, d4)'; -- err, duplicate name +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d1, d2, d1, d4)" +DETAIL: Duplicate standby node name: d1 +alter system set synchronous_standby_names = 'any 5 (d1, d2, d3, d4)'; -- err, requriement more than have +ERROR: invalid value for parameter "synchronous_standby_names": "any 5 (d1, d2, d3, d4)" +DETAIL: The sync number must less or equals to the number of standby node names. +alter system set synchronous_standby_names = 'any 0 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = 'any -1 (d1, d2, d3, d4)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "any -1 (d1, d2, d3, d4)" +DETAIL: synchronous_standby_names parser failed +-- combinate gram +alter system set synchronous_standby_names = 'd1, d2, d3, 1 (d5, d6)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "d1, d2, d3, 1 (d5, d6)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = '1 (d1, d2, d3), 1 (d5, d6)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "1 (d1, d2, d3), 1 (d5, d6)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = '*, 1 (d5, d6)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "*, 1 (d5, d6)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = '1 (d1, d2, d3), 1 (*)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "1 (d1, d2, d3), 1 (*)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = '1 (d1, d2, d3), 1 (*)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "1 (d1, d2, d3), 1 (*)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = 'any 2 (*), any 2 (d1, d2, d3, d4), any 2 (d5, d6, d7)'; -- err, semantic ambiguity +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (*), any 2 (d1, d2, d3, d4), any 2 (d5, d6, d7)" +DETAIL: '*' is not support when sync standby strategy is combinated. +alter system set synchronous_standby_names = 'any 2 (d, dd), any 2 (d1, *, d3, d4), any 2 (d5, d6, d7)'; -- err, semantic ambiguity +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d, dd), any 2 (d1, *, d3, d4), any 2 (d5, d6, d7)" +DETAIL: '*' is not support when sync standby strategy is combinated. +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), d5, d6, d7'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d1, d2, d3, d4), d5, d6, d7" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), 2 (d5, d1, d7)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d1, d2, d3, d4), 2 (d5, d1, d7)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), *'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d1, d2, d3, d4), *" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = '2 (d1, d2, d3, d4), first 2 (d5, d6, d7)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "2 (d1, d2, d3, d4), first 2 (d5, d6, d7)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = '2 (d1, d2, d3, d4), any 4 (d5, d6, d7)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "2 (d1, d2, d3, d4), any 4 (d5, d6, d7)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = '*, any 0 (d5, d6, d7)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "*, any 0 (d5, d6, d7)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = 'd1, d2, d3, d4, any 2 (d5, d6, d7)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "d1, d2, d3, d4, any 2 (d5, d6, d7)" +DETAIL: synchronous_standby_names parser failed +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), any 2 (d5, d6, d7)'; -- suc +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), any 2 (d5, d1, d7)'; -- err, duplicate name +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d1, d2, d3, d4), any 2 (d5, d1, d7)" +DETAIL: Duplicate standby node name: d1 +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), any 2 (d5, d6, d6)'; -- err, duplicate name +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d1, d2, d3, d4), any 2 (d5, d6, d6)" +DETAIL: Duplicate standby node name: d6 +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), first 2 (d5, d6, d7)'; -- err, not support first +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d1, d2, d3, d4), first 2 (d5, d6, d7)" +DETAIL: FIRST rule is not support when sync standby strategy is combinated. +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), any 4 (d5, d6, d7)'; -- err, requriement more than have +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d1, d2, d3, d4), any 4 (d5, d6, d7)" +DETAIL: The sync number must less or equals to the number of standby node names. +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), any 0 (d5, d6, d7)'; -- suc +alter system set synchronous_standby_names = 'first 2 (d1, d2, d3, d4), first 2 (d5, d6, d7)'; -- err, not support first +ERROR: invalid value for parameter "synchronous_standby_names": "first 2 (d1, d2, d3, d4), first 2 (d5, d6, d7)" +DETAIL: FIRST rule is not support when sync standby strategy is combinated. +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), , any 0 (d5, d6, d7)'; -- err, parser err +ERROR: invalid value for parameter "synchronous_standby_names": "any 2 (d1, d2, d3, d4), , any 0 (d5, d6, d7)" +DETAIL: synchronous_standby_names parser failed +-- recover +alter system set synchronous_standby_names = '*'; +select pg_sleep(3); + pg_sleep +---------- + +(1 row) + +show synchronous_standby_names; + synchronous_standby_names +--------------------------- + * +(1 row) + diff --git a/src/test/regress/parallel_schedule0 b/src/test/regress/parallel_schedule0 index 1243e5a8..c0ea599b 100644 --- a/src/test/regress/parallel_schedule0 +++ b/src/test/regress/parallel_schedule0 @@ -10,7 +10,7 @@ test: analyze_commands test: single_node_ddl test: single_node_sqlbypass test: median - +test: sync_standy_names # run tablespace by itself, and first, because it forces a checkpoint; # we'd prefer not to have checkpoints later in the tests because that diff --git a/src/test/regress/sql/sync_standy_names.sql b/src/test/regress/sql/sync_standy_names.sql new file mode 100644 index 00000000..992423ed --- /dev/null +++ b/src/test/regress/sql/sync_standy_names.sql @@ -0,0 +1,62 @@ +show synchronous_standby_names; + +-- single gram +alter system set synchronous_standby_names = '*'; -- suc +alter system set synchronous_standby_names = 'd1, *'; -- err, semantic ambiguity +alter system set synchronous_standby_names = '*, *'; -- err, semantic ambiguity +alter system set synchronous_standby_names = 'd1'; -- suc +alter system set synchronous_standby_names = 'd1, d2, d3, d4'; -- suc +alter system set synchronous_standby_names = 'd1, d2, d1, d4'; -- err, duplicate name +alter system set synchronous_standby_names = '2 (*)'; -- suc +alter system set synchronous_standby_names = '2 (d1, *)'; -- err, semantic ambiguity +alter system set synchronous_standby_names = '2 (*, *)'; -- err, semantic ambiguity +alter system set synchronous_standby_names = '2 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = '2 (d1, d2, d1, d4)'; -- err, duplicate name +alter system set synchronous_standby_names = '5 (d1, d2, d3, d4)'; -- err, requriement more than have +alter system set synchronous_standby_names = '0 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = '-1 (d1, d2, d3, d4)'; -- err, parser err +alter system set synchronous_standby_names = 'first 2 (*)'; -- suc +alter system set synchronous_standby_names = 'first 2 (d1, *)'; -- err, semantic ambiguity +alter system set synchronous_standby_names = 'first 2 (*, *)'; -- err, semantic ambiguity +alter system set synchronous_standby_names = 'first 2 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = 'first 2 (d1, d2, d1, d4)'; -- err, duplicate name +alter system set synchronous_standby_names = 'first 5 (d1, d2, d3, d4)'; -- err, requriement more than have +alter system set synchronous_standby_names = 'first 0 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = 'first -1 (d1, d2, d3, d4)'; -- err, parser err +alter system set synchronous_standby_names = 'any 2 (*)'; -- suc +alter system set synchronous_standby_names = 'any 2 (d1, *'; -- err, semantic ambiguity +alter system set synchronous_standby_names = 'any 2 (*, *)'; -- err, semantic ambiguity +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = 'any 2 (d1, d2, d1, d4)'; -- err, duplicate name +alter system set synchronous_standby_names = 'any 5 (d1, d2, d3, d4)'; -- err, requriement more than have +alter system set synchronous_standby_names = 'any 0 (d1, d2, d3, d4)'; -- suc +alter system set synchronous_standby_names = 'any -1 (d1, d2, d3, d4)'; -- err, parser err + +-- combinate gram +alter system set synchronous_standby_names = 'd1, d2, d3, 1 (d5, d6)'; -- err, parser err +alter system set synchronous_standby_names = '1 (d1, d2, d3), 1 (d5, d6)'; -- err, parser err +alter system set synchronous_standby_names = '*, 1 (d5, d6)'; -- err, parser err +alter system set synchronous_standby_names = '1 (d1, d2, d3), 1 (*)'; -- err, parser err +alter system set synchronous_standby_names = '1 (d1, d2, d3), 1 (*)'; -- err, parser err +alter system set synchronous_standby_names = 'any 2 (*), any 2 (d1, d2, d3, d4), any 2 (d5, d6, d7)'; -- err, semantic ambiguity +alter system set synchronous_standby_names = 'any 2 (d, dd), any 2 (d1, *, d3, d4), any 2 (d5, d6, d7)'; -- err, semantic ambiguity +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), d5, d6, d7'; -- err, parser err +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), 2 (d5, d1, d7)'; -- err, parser err +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), *'; -- err, parser err +alter system set synchronous_standby_names = '2 (d1, d2, d3, d4), first 2 (d5, d6, d7)'; -- err, parser err +alter system set synchronous_standby_names = '2 (d1, d2, d3, d4), any 4 (d5, d6, d7)'; -- err, parser err +alter system set synchronous_standby_names = '*, any 0 (d5, d6, d7)'; -- err, parser err +alter system set synchronous_standby_names = 'd1, d2, d3, d4, any 2 (d5, d6, d7)'; -- err, parser err +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), any 2 (d5, d6, d7)'; -- suc +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), any 2 (d5, d1, d7)'; -- err, duplicate name +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), any 2 (d5, d6, d6)'; -- err, duplicate name +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), first 2 (d5, d6, d7)'; -- err, not support first +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), any 4 (d5, d6, d7)'; -- err, requriement more than have +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), any 0 (d5, d6, d7)'; -- suc +alter system set synchronous_standby_names = 'first 2 (d1, d2, d3, d4), first 2 (d5, d6, d7)'; -- err, not support first +alter system set synchronous_standby_names = 'any 2 (d1, d2, d3, d4), , any 0 (d5, d6, d7)'; -- err, parser err + +-- recover +alter system set synchronous_standby_names = '*'; +select pg_sleep(3); +show synchronous_standby_names;