diff options
| -rw-r--r-- | src/backend/commands/analyze.c | 133 |
1 files changed, 103 insertions, 30 deletions
diff --git a/src/backend/commands/analyze.c b/src/backend/commands/analyze.c index ca48fca486..832d99a065 100644 --- a/src/backend/commands/analyze.c +++ b/src/backend/commands/analyze.c @@ -115,7 +115,8 @@ static Datum ind_fetch_func(VacAttrStatsP stats, int rownum, bool *isNull); #ifdef XCP static void analyze_rel_coordinator(Relation onerel, bool inh, int attr_cnt, - VacAttrStats **vacattrstats); + VacAttrStats **vacattrstats, int nindexes, + Relation *indexes, AnlIndexData *indexdata); #endif /* @@ -424,30 +425,6 @@ do_analyze_rel(Relation onerel, int options, VacuumParams *params, attr_cnt = tcnt; } -#ifdef XCP - if (IS_PGXC_COORDINATOR && onerel->rd_locator_info) - { - /* - * Fetch relation statistics from remote nodes and update - */ - vacuum_rel_coordinator(onerel, in_outer_xact); - - /* - * Fetch attribute statistics from remote nodes. - */ - analyze_rel_coordinator(onerel, inh, attr_cnt, vacattrstats); - - /* - * Skip acquiring local stats. Coordinator does not store data of - * distributed tables. - */ - nindexes = 0; - hasindex = false; - Irel = NULL; - goto cleanup; - } -#endif - /* * Open all indexes of the relation, and see if there are any analyzable * columns in the indexes. We do not analyze index columns if there was @@ -505,6 +482,28 @@ do_analyze_rel(Relation onerel, int options, VacuumParams *params, } } +#ifdef XCP + if (IS_PGXC_COORDINATOR && onerel->rd_locator_info) + { + /* + * Fetch relation statistics from remote nodes and update + */ + vacuum_rel_coordinator(onerel, in_outer_xact); + + /* + * Fetch attribute statistics from remote nodes. + */ + analyze_rel_coordinator(onerel, inh, attr_cnt, vacattrstats, + nindexes, Irel, indexdata); + + /* + * Skip acquiring local stats. Coordinator does not store data of + * distributed tables. + */ + goto cleanup; + } +#endif + /* * Determine how many rows we need to sample, using the worst case from * all analyzable columns. We use a lower bound of 100 rows to avoid @@ -2944,9 +2943,24 @@ compare_mcvs(const void *a, const void *b) #ifdef XCP +/* + * coord_collect_simple_stats + * Collect simple stats for a relation (pg_statistic contents). + * + * Collects statistics from the datanodes, and then keeps the one of the + * received statistics for each attribute (the first one we receive, but + * it's mostly random). + * + * XXX We do not try to build statistics covering data fro all the nodes, + * either by collecting fresh sample of rows or merging the statistics + * somehow. The current approach is very simple and cheap, but may have + * negative impact on estimate accuracy as the stats only covers data + * from a single node, and we may end up with stats from different node + * for each attribute. + */ static void -analyze_rel_coordinator(Relation onerel, bool inh, int attr_cnt, - VacAttrStats **vacattrstats) +coord_collect_simple_stats(Relation onerel, bool inh, int attr_cnt, + VacAttrStats **vacattrstats) { char *nspname; char *relname; @@ -2960,14 +2974,11 @@ analyze_rel_coordinator(Relation onerel, bool inh, int attr_cnt, int i; /* Number of data nodes from which attribute statistics are received. */ int *numnodes; - List *stat_oids; /* Get the relation identifier */ relname = RelationGetRelationName(onerel); nspname = get_namespace_name(RelationGetNamespace(onerel)); - elog(LOG, "Getting detailed statistics for %s.%s", nspname, relname); - /* Make up query string */ initStringInfo(&query); /* Generic statistic fields */ @@ -3367,6 +3378,38 @@ analyze_rel_coordinator(Relation onerel, bool inh, int attr_cnt, } } update_attstats(RelationGetRelid(onerel), inh, attr_cnt, vacattrstats); +} + +/* + * coord_collect_extended_stats + * Collect extended stats for a relation (pg_statistic_ext contents). + * + * Collects statistics from the datanodes, and then keeps the one of the + * received statistics for each attribute (the first one we receive, but + * it's mostly random). + * + * XXX This has similar issues as coord_collect_simple_stats. + */ +static void +coord_collect_extended_stats(Relation onerel, int attr_cnt) +{ + char *nspname; + char *relname; + /* Fields to run query to read statistics from data nodes */ + StringInfoData query; + EState *estate; + MemoryContext oldcontext; + RemoteQuery *step; + RemoteQueryState *node; + TupleTableSlot *result; + int i; + /* Number of data nodes from which attribute statistics are received. */ + int *numnodes; + List *stat_oids; + + /* Get the relation identifier */ + relname = RelationGetRelationName(onerel); + nspname = get_namespace_name(RelationGetNamespace(onerel)); /* * Build extended statistics on the coordinator. @@ -3521,4 +3564,34 @@ analyze_rel_coordinator(Relation onerel, bool inh, int attr_cnt, } ExecEndRemoteQuery(node); } + +/* + * analyze_rel_coordinator + * Collect all statistics for a particular relation. + * + * We collect three types of statistics for each table: + * + * - simple statistics (pg_statistic) + * - extended statistics (pg_statistic_ext) + * - index statistics (including expression indexes) + */ +static void +analyze_rel_coordinator(Relation onerel, bool inh, int attr_cnt, + VacAttrStats **vacattrstats, int nindexes, + Relation *indexes, AnlIndexData *indexdata) +{ + int i; + + /* simple statistics (pg_statistic) for the relation */ + coord_collect_simple_stats(onerel, inh, attr_cnt, vacattrstats); + + /* simple statistics (pg_statistic) for all indexes */ + for (i = 0; i < nindexes; i++) + coord_collect_simple_stats(indexes[i], false, + indexdata[i].attr_cnt, + indexdata[i].vacattrstats); + + /* extended statistics (pg_statistic) for the relation */ + coord_collect_extended_stats(onerel, attr_cnt); +} #endif |