/* * __curjoin_init_next -- * Initialize the cursor join when the next function is first called. */ static int __curjoin_init_next(WT_SESSION_IMPL *session, WT_CURSOR_JOIN *cjoin, bool iterable) { WT_BLOOM *bloom; WT_CURSOR *origcur; WT_CURSOR_JOIN_ENDPOINT *end; WT_CURSOR_JOIN_ENTRY *je, *jeend, *je2; WT_DECL_RET; size_t size; uint32_t f, k; char *mainbuf; const char *def_cfg[] = { WT_CONFIG_BASE( session, WT_SESSION_open_cursor), NULL }; const char *raw_cfg[] = { WT_CONFIG_BASE( session, WT_SESSION_open_cursor), "raw", NULL }; const char **config, *proj, *urimain; mainbuf = NULL; if (cjoin->entries_next == 0) WT_RET_MSG(session, EINVAL, "join cursor has not yet been joined with any other " "cursors"); /* Get a consistent view of our subordinate cursors if appropriate. */ __wt_txn_cursor_op(session); if (F_ISSET((WT_CURSOR *)cjoin, WT_CURSTD_RAW)) config = &raw_cfg[0]; else config = &def_cfg[0]; urimain = cjoin->table->iface.name; if ((proj = cjoin->projection) != NULL) { size = strlen(urimain) + strlen(proj) + 1; WT_ERR(__wt_calloc(session, size, 1, &mainbuf)); WT_ERR(__wt_snprintf(mainbuf, size, "%s%s", urimain, proj)); urimain = mainbuf; } WT_ERR(__wt_open_cursor(session, urimain, (WT_CURSOR *)cjoin, config, &cjoin->main)); jeend = &cjoin->entries[cjoin->entries_next]; for (je = cjoin->entries; je < jeend; je++) { if (je->subjoin != NULL) { WT_ERR(__curjoin_init_next(session, je->subjoin, iterable)); continue; } __wt_stat_join_init_single(&je->stats); /* * For a single compare=le/lt endpoint in any entry that may * be iterated, construct a companion compare=ge endpoint * that will actually be iterated. */ if (iterable && je->ends_next == 1 && F_ISSET(&je->ends[0], WT_CURJOIN_END_LT)) { origcur = je->ends[0].cursor; WT_ERR(__curjoin_insert_endpoint(session, je, 0, &end)); WT_ERR(__wt_open_cursor(session, origcur->uri, (WT_CURSOR *)cjoin, F_ISSET(origcur, WT_CURSTD_RAW) ? raw_cfg : def_cfg, &end->cursor)); end->flags = WT_CURJOIN_END_GT | WT_CURJOIN_END_EQ | WT_CURJOIN_END_OWN_CURSOR; WT_ERR(end->cursor->next(end->cursor)); F_CLR(je, WT_CURJOIN_ENTRY_DISJUNCTION); } for (end = &je->ends[0]; end < &je->ends[je->ends_next]; end++) WT_ERR(__curjoin_endpoint_init_key(session, je, end)); /* * Do any needed Bloom filter initialization. Ignore Bloom * filters for entries that will be iterated. They won't * help since these entries either don't need an inclusion * check or are doing any needed check during the iteration. */ if (!iterable && F_ISSET(je, WT_CURJOIN_ENTRY_BLOOM)) { if (session->txn.isolation == WT_ISO_READ_UNCOMMITTED) WT_ERR_MSG(session, EINVAL, "join cursors with Bloom filters cannot be " "used with read-uncommitted isolation"); if (je->bloom == NULL) { /* * Look for compatible filters to be shared, * pick compatible numbers for bit counts * and number of hashes. */ f = je->bloom_bit_count; k = je->bloom_hash_count; for (je2 = je + 1; je2 < jeend; je2++) if (F_ISSET(je2, WT_CURJOIN_ENTRY_BLOOM) && je2->count == je->count) { f = WT_MAX( je2->bloom_bit_count, f); k = WT_MAX( je2->bloom_hash_count, k); } je->bloom_bit_count = f; je->bloom_hash_count = k; WT_ERR(__wt_bloom_create(session, NULL, NULL, je->count, f, k, &je->bloom)); F_SET(je, WT_CURJOIN_ENTRY_OWN_BLOOM); WT_ERR(__curjoin_init_bloom(session, cjoin, je, je->bloom)); /* * Share the Bloom filter, making all * config info consistent. */ for (je2 = je + 1; je2 < jeend; je2++) if (F_ISSET(je2, WT_CURJOIN_ENTRY_BLOOM) && je2->count == je->count) { WT_ASSERT(session, je2->bloom == NULL); je2->bloom = je->bloom; je2->bloom_bit_count = f; je2->bloom_hash_count = k; } } else { /* * Create a temporary filter that we'll * merge into the shared one. The Bloom * parameters of the two filters must match. */ WT_ERR(__wt_bloom_create(session, NULL, NULL, je->count, je->bloom_bit_count, je->bloom_hash_count, &bloom)); WT_ERR(__curjoin_init_bloom(session, cjoin, je, bloom)); WT_ERR(__wt_bloom_intersection(je->bloom, bloom)); WT_ERR(__wt_bloom_close(bloom)); } } if (!F_ISSET(cjoin, WT_CURJOIN_DISJUNCTION)) iterable = false; } F_SET(cjoin, WT_CURJOIN_INITIALIZED); err: __wt_free(session, mainbuf); return (ret); }
/* * __curjoin_init_iter -- * Initialize before any iteration. */ static int __curjoin_init_iter(WT_SESSION_IMPL *session, WT_CURSOR_JOIN *cjoin) { WT_BLOOM *bloom; WT_DECL_RET; WT_CURSOR_JOIN_ENTRY *je, *jeend, *je2; WT_CURSOR_JOIN_ENDPOINT *end; uint32_t f, k; if (cjoin->entries_next == 0) WT_RET_MSG(session, EINVAL, "join cursor has not yet been joined with any other " "cursors"); je = &cjoin->entries[0]; WT_RET(__curjoin_entry_iter_init(session, cjoin, je, &cjoin->iter)); jeend = &cjoin->entries[cjoin->entries_next]; for (je = cjoin->entries; je < jeend; je++) { __wt_stat_join_init_single(&je->stats); for (end = &je->ends[0]; end < &je->ends[je->ends_next]; end++) WT_RET(__curjoin_endpoint_init_key(session, je, end)); /* * The first entry is iterated as the 'outermost' cursor. * For the common GE case, we don't have to test against * the left reference key, we know it will be true since * the btree is ordered. */ if (je == cjoin->entries && je->ends[0].flags == (WT_CURJOIN_END_GT | WT_CURJOIN_END_EQ)) F_SET(cjoin, WT_CURJOIN_SKIP_FIRST_LEFT); if (F_ISSET(je, WT_CURJOIN_ENTRY_BLOOM)) { if (je->bloom == NULL) { /* * Look for compatible filters to be shared, * pick compatible numbers for bit counts * and number of hashes. */ f = je->bloom_bit_count; k = je->bloom_hash_count; for (je2 = je + 1; je2 < jeend; je2++) if (F_ISSET(je2, WT_CURJOIN_ENTRY_BLOOM) && je2->count == je->count) { f = WT_MAX( je2->bloom_bit_count, f); k = WT_MAX( je2->bloom_hash_count, k); } je->bloom_bit_count = f; je->bloom_hash_count = k; WT_RET(__wt_bloom_create(session, NULL, NULL, je->count, f, k, &je->bloom)); F_SET(je, WT_CURJOIN_ENTRY_OWN_BLOOM); WT_RET(__curjoin_init_bloom(session, cjoin, je, je->bloom)); /* * Share the Bloom filter, making all * config info consistent. */ for (je2 = je + 1; je2 < jeend; je2++) if (F_ISSET(je2, WT_CURJOIN_ENTRY_BLOOM) && je2->count == je->count) { WT_ASSERT(session, je2->bloom == NULL); je2->bloom = je->bloom; je2->bloom_bit_count = f; je2->bloom_hash_count = k; } } else { /* * Create a temporary filter that we'll * merge into the shared one. The Bloom * parameters of the two filters must match. */ WT_RET(__wt_bloom_create(session, NULL, NULL, je->count, je->bloom_bit_count, je->bloom_hash_count, &bloom)); WT_RET(__curjoin_init_bloom(session, cjoin, je, bloom)); WT_RET(__wt_bloom_intersection(je->bloom, bloom)); WT_RET(__wt_bloom_close(bloom)); } } } F_SET(cjoin, WT_CURJOIN_INITIALIZED); return (ret); }