Query planner strategy for record links (#3668)

core/src/dbs/iterator.rs

@@ -19,19 +19,18 @@ use crate::sql::thing::Thing;
 use crate::sql::value::Value;
 use async_recursion::async_recursion;
 use std::mem;
-use std::sync::Arc;
 
 #[derive(Clone)]
 pub(crate) enum Iterable {
 	Value(Value),
-	Table(Arc<Table>),
+	Table(Table),
 	Thing(Thing),
 	Range(Range),
 	Edges(Edges),
 	Defer(Thing),
 	Mergeable(Thing, Value),
 	Relatable(Thing, Thing, Thing),
-	Index(Arc<Table>, IteratorRef),
+	Index(Table, IteratorRef),
 }
 
 pub(crate) struct Processed {
@@ -118,7 +117,7 @@ impl Iterator {
 					}
 					_ => {
 						// Ingest the table for scanning
-						self.ingest(Iterable::Table(Arc::new(v)))
+						self.ingest(Iterable::Table(v))
 					}
 				},
 				// There is no data clause so create a record id
@@ -129,7 +128,7 @@ impl Iterator {
 					}
 					_ => {
 						// Ingest the table for scanning
-						self.ingest(Iterable::Table(Arc::new(v)))
+						self.ingest(Iterable::Table(v))
 					}
 				},
 			},

core/src/dbs/processor.rs

@@ -127,10 +127,10 @@ impl<'a> Processor<'a> {
 							// Avoiding search in the hashmap of the query planner for each doc
 							let mut ctx = Context::new(ctx);
 							ctx.set_query_executor(exe.clone());
-							return self.process_table(&ctx, opt, txn, stm, v.as_ref()).await;
+							return self.process_table(&ctx, opt, txn, stm, &v).await;
 						}
 					}
-					self.process_table(ctx, opt, txn, stm, v.as_ref()).await?
+					self.process_table(ctx, opt, txn, stm, &v).await?
 				}
 				Iterable::Range(v) => self.process_range(ctx, opt, txn, stm, v).await?,
 				Iterable::Edges(e) => self.process_edge(ctx, opt, txn, stm, e).await?,
@@ -141,10 +141,10 @@ impl<'a> Processor<'a> {
 							// Avoiding search in the hashmap of the query planner for each doc
 							let mut ctx = Context::new(ctx);
 							ctx.set_query_executor(exe.clone());
-							return self.process_index(&ctx, opt, txn, stm, t.as_ref(), ir).await;
+							return self.process_index(&ctx, opt, txn, stm, &t, ir).await;
 						}
 					}
-					self.process_index(ctx, opt, txn, stm, t.as_ref(), ir).await?
+					self.process_index(ctx, opt, txn, stm, &t, ir).await?
 				}
 				Iterable::Mergeable(v, o) => {
 					self.process_mergeable(ctx, opt, txn, stm, v, o).await?
@@ -563,7 +563,8 @@ impl<'a> Processor<'a> {
 		txn.lock().await.check_ns_db_tb(opt.ns(), opt.db(), &table.0, opt.strict).await?;
 		if let Some(exe) = ctx.get_query_executor() {
 			if let Some(mut iterator) = exe.new_iterator(opt, ir).await? {
-				let mut things = iterator.next_batch(txn, PROCESSOR_BATCH_SIZE).await?;
+				let mut things = Vec::new();
+				iterator.next_batch(txn, PROCESSOR_BATCH_SIZE, &mut things).await?;
 				while !things.is_empty() {
 					// Check if the context is finished
 					if ctx.is_done() {
@@ -601,7 +602,8 @@ impl<'a> Processor<'a> {
 					}
 
 					// Collect the next batch of ids
-					things = iterator.next_batch(txn, PROCESSOR_BATCH_SIZE).await?;
+					things = Vec::new();
+					iterator.next_batch(txn, PROCESSOR_BATCH_SIZE, &mut things).await?;
 				}
 				// Everything ok
 				return Ok(());

core/src/fnc/operate.rs

@@ -212,15 +212,19 @@ fn get_executor_option<'a>(
 
 pub(crate) async fn matches(
 	ctx: &Context<'_>,
+	opt: &Options,
 	txn: &Transaction,
 	doc: Option<&CursorDoc<'_>>,
 	exp: &Expression,
+	l: Value,
+	r: Value,
 ) -> Result<Value, Error> {
-	match get_executor_option(ctx, doc, exp) {
+	let res = match get_executor_option(ctx, doc, exp) {
-		ExecutorOption::PreMatch => Ok(Value::Bool(true)),
+		ExecutorOption::PreMatch => true,
-		ExecutorOption::None => Ok(Value::Bool(false)),
+		ExecutorOption::None => false,
-		ExecutorOption::Execute(exe, thg) => exe.matches(txn, thg, exp).await,
+		ExecutorOption::Execute(exe, thg) => exe.matches(ctx, opt, txn, thg, exp, l, r).await?,
-	}
+	};
+	Ok(res.into())
 }
 
 pub(crate) async fn knn(

core/src/idx/ft/analyzer/filter.rs

@@ -11,7 +11,6 @@ pub(super) enum FilteringStage {
 	Indexing,
 	Querying,
 }
-
 pub(super) enum Filter {
 	Stemmer(Stemmer),
 	Ascii,

core/src/idx/ft/analyzer/mod.rs

@@ -6,7 +6,7 @@ use crate::idx::ft::analyzer::tokenizer::{Tokenizer, Tokens};
 use crate::idx::ft::doclength::DocLength;
 use crate::idx::ft::offsets::{Offset, OffsetRecords};
 use crate::idx::ft::postings::TermFrequency;
-use crate::idx::ft::terms::{TermId, Terms};
+use crate::idx::ft::terms::{TermId, TermLen, Terms};
 use crate::sql::statements::DefineAnalyzerStatement;
 use crate::sql::tokenizer::Tokenizer as SqlTokenizer;
 use crate::sql::Value;
@@ -34,31 +34,96 @@ impl From<DefineAnalyzerStatement> for Analyzer {
 		}
 	}
 }
+
+pub(in crate::idx) type TermsList = Vec<Option<(TermId, TermLen)>>;
+
+pub(in crate::idx) struct TermsSet {
+	set: HashSet<TermId>,
+	has_unknown_terms: bool,
+}
+
+impl TermsSet {
+	/// If the query TermsSet contains terms that are unknown in the index
+	/// of if there is no terms in the set then
+	/// we are sure that it does not match any document
+	pub(in crate::idx) fn is_matchable(&self) -> bool {
+		!(self.has_unknown_terms || self.set.is_empty())
+	}
+
+	pub(in crate::idx) fn is_subset(&self, other: &TermsSet) -> bool {
+		if self.has_unknown_terms {
+			return false;
+		}
+		self.set.is_subset(&other.set)
+	}
+}
+
 impl Analyzer {
-	pub(super) async fn extract_terms(
+	pub(super) async fn extract_querying_terms(
 		&self,
 		ctx: &Context<'_>,
 		opt: &Options,
 		txn: &Transaction,
 		t: &Terms,
-		query_string: String,
+		content: String,
-	) -> Result<Vec<Option<(TermId, u32)>>, Error> {
+	) -> Result<(TermsList, TermsSet), Error> {
-		let tokens =
+		let tokens = self.generate_tokens(ctx, opt, txn, FilteringStage::Querying, content).await?;
-			self.generate_tokens(ctx, opt, txn, FilteringStage::Querying, query_string).await?;
+		// We extract the term ids
-		// We first collect every unique terms
+		let mut list = Vec::with_capacity(tokens.list().len());
-		// as it can contains duplicates
+		let mut unique_tokens = HashSet::new();
-		let mut terms = HashSet::new();
+		let mut set = HashSet::new();
-		for token in tokens.list() {
-			terms.insert(token);
-		}
-		// Now we can extract the term ids
-		let mut res = Vec::with_capacity(terms.len());
 		let mut tx = txn.lock().await;
-		for term in terms {
+		let mut has_unknown_terms = false;
-			let opt_term_id = t.get_term_id(&mut tx, tokens.get_token_string(term)?).await?;
+		for token in tokens.list() {
-			res.push(opt_term_id.map(|tid| (tid, term.get_char_len())));
+			// Tokens can contains duplicated, not need to evaluate them again
+			if unique_tokens.insert(token) {
+				// Is the term known in the index?
+				let opt_term_id = t.get_term_id(&mut tx, tokens.get_token_string(token)?).await?;
+				list.push(opt_term_id.map(|tid| (tid, token.get_char_len())));
+				if let Some(term_id) = opt_term_id {
+					set.insert(term_id);
+				} else {
+					has_unknown_terms = true;
+				}
+			}
 		}
-		Ok(res)
+		Ok((
+			list,
+			TermsSet {
+				set,
+				has_unknown_terms,
+			},
+		))
+	}
+
+	pub(in crate::idx) async fn extract_indexing_terms(
+		&self,
+		ctx: &Context<'_>,
+		opt: &Options,
+		txn: &Transaction,
+		t: &Terms,
+		content: Value,
+	) -> Result<TermsSet, Error> {
+		let mut tv = Vec::new();
+		self.analyze_value(ctx, opt, txn, content, FilteringStage::Indexing, &mut tv).await?;
+		let mut set = HashSet::new();
+		let mut has_unknown_terms = false;
+		let mut tx = txn.lock().await;
+		for tokens in tv {
+			for token in tokens.list() {
+				if let Some(term_id) =
+					t.get_term_id(&mut tx, tokens.get_token_string(token)?).await?
+				{
+					set.insert(term_id);
+				} else {
+					has_unknown_terms = true;
+				}
+			}
+		}
+		Ok(TermsSet {
+			set,
+			has_unknown_terms,
+		})
 	}
 
 	/// This method is used for indexing.

core/src/idx/ft/analyzer/tokenizer.rs

@@ -5,7 +5,7 @@ use crate::idx::ft::offsets::{Offset, Position};
 use crate::sql::tokenizer::Tokenizer as SqlTokenizer;
 use crate::sql::Value;
 
-pub(super) struct Tokens {
+pub(in crate::idx) struct Tokens {
 	/// The input string
 	i: String,
 	/// The final list of tokens

core/src/idx/ft/mod.rs

@@ -11,14 +11,14 @@ use crate::ctx::Context;
 use crate::dbs::{Options, Transaction};
 use crate::err::Error;
 use crate::idx::docids::{DocId, DocIds};
-use crate::idx::ft::analyzer::Analyzer;
+use crate::idx::ft::analyzer::{Analyzer, TermsList, TermsSet};
 use crate::idx::ft::doclength::DocLengths;
 use crate::idx::ft::highlighter::{Highlighter, Offseter};
 use crate::idx::ft::offsets::Offsets;
 use crate::idx::ft::postings::Postings;
 use crate::idx::ft::scorer::BM25Scorer;
 use crate::idx::ft::termdocs::{TermDocs, TermsDocs};
-use crate::idx::ft::terms::{TermId, Terms};
+use crate::idx::ft::terms::{TermId, TermLen, Terms};
 use crate::idx::trees::btree::BStatistics;
 use crate::idx::trees::store::IndexStores;
 use crate::idx::{IndexKeyBase, VersionedSerdeState};
@@ -39,7 +39,7 @@ use tokio::sync::RwLock;
 pub(crate) type MatchRef = u8;
 
 pub(crate) struct FtIndex {
-	analyzer: Analyzer,
+	analyzer: Arc<Analyzer>,
 	state_key: Key,
 	index_key_base: IndexKeyBase,
 	state: State,
@@ -164,7 +164,7 @@ impl FtIndex {
 			index_key_base,
 			bm25,
 			highlighting: p.hl,
-			analyzer: az.into(),
+			analyzer: Arc::new(az.into()),
 			doc_ids,
 			doc_lengths,
 			postings,
@@ -178,6 +178,14 @@ impl FtIndex {
 		self.doc_ids.clone()
 	}
 
+	pub(super) fn terms(&self) -> Arc<RwLock<Terms>> {
+		self.terms.clone()
+	}
+
+	pub(super) fn analyzer(&self) -> Arc<Analyzer> {
+		self.analyzer.clone()
+	}
+
 	pub(crate) async fn remove_document(
 		&mut self,
 		txn: &Transaction,
@@ -326,22 +334,22 @@ impl FtIndex {
 		Ok(())
 	}
 
-	pub(super) async fn extract_terms(
+	pub(super) async fn extract_querying_terms(
 		&self,
 		ctx: &Context<'_>,
 		opt: &Options,
 		txn: &Transaction,
 		query_string: String,
-	) -> Result<Vec<Option<(TermId, u32)>>, Error> {
+	) -> Result<(TermsList, TermsSet), Error> {
 		let t = self.terms.read().await;
-		let terms = self.analyzer.extract_terms(ctx, opt, txn, &t, query_string).await?;
+		let res = self.analyzer.extract_querying_terms(ctx, opt, txn, &t, query_string).await?;
-		Ok(terms)
+		Ok(res)
 	}
 
 	pub(super) async fn get_terms_docs(
 		&self,
 		tx: &mut kvs::Transaction,
-		terms: &Vec<Option<(TermId, u32)>>,
+		terms: &TermsList,
 	) -> Result<Vec<Option<(TermId, RoaringTreemap)>>, Error> {
 		let mut terms_docs = Vec::with_capacity(terms.len());
 		for opt_term in terms {
@@ -402,7 +410,7 @@ impl FtIndex {
 		&self,
 		tx: &mut kvs::Transaction,
 		thg: &Thing,
-		terms: &[Option<(TermId, u32)>],
+		terms: &[Option<(TermId, TermLen)>],
 		prefix: Value,
 		suffix: Value,
 		partial: bool,
@@ -538,9 +546,10 @@ mod tests {
 		fti: &FtIndex,
 		qs: &str,
 	) -> (Option<HitsIterator>, BM25Scorer) {
-		let t = fti.extract_terms(ctx, opt, txn, qs.to_string()).await.unwrap();
+		let (term_list, _) =
+			fti.extract_querying_terms(ctx, opt, txn, qs.to_string()).await.unwrap();
 		let mut tx = txn.lock().await;
-		let td = Arc::new(fti.get_terms_docs(&mut tx, &t).await.unwrap());
+		let td = Arc::new(fti.get_terms_docs(&mut tx, &term_list).await.unwrap());
 		drop(tx);
 		let scr = fti.new_scorer(td.clone()).unwrap().unwrap();
 		let hits = fti.new_hits_iterator(td).unwrap();

core/src/idx/ft/terms.rs

@@ -9,8 +9,9 @@ use roaring::RoaringTreemap;
 use serde::{Deserialize, Serialize};
 
 pub(crate) type TermId = u64;
+pub(crate) type TermLen = u32;
 
-pub(super) struct Terms {
+pub(in crate::idx) struct Terms {
 	state_key: Key,
 	index_key_base: IndexKeyBase,
 	btree: BTree<FstKeys>,

core/src/idx/planner/executor.rs

@@ -3,19 +3,20 @@ use crate::dbs::{Options, Transaction};
 use crate::doc::CursorDoc;
 use crate::err::Error;
 use crate::idx::docids::{DocId, DocIds};
+use crate::idx::ft::analyzer::{Analyzer, TermsList, TermsSet};
 use crate::idx::ft::scorer::BM25Scorer;
 use crate::idx::ft::termdocs::TermsDocs;
-use crate::idx::ft::terms::TermId;
+use crate::idx::ft::terms::Terms;
 use crate::idx::ft::{FtIndex, MatchRef};
 use crate::idx::planner::iterators::{
-	DocIdsIterator, IndexEqualThingIterator, IndexRangeThingIterator, IndexUnionThingIterator,
+	DocIdsIterator, IndexEqualThingIterator, IndexJoinThingIterator, IndexRangeThingIterator,
-	MatchesThingIterator, ThingIterator, UniqueEqualThingIterator, UniqueRangeThingIterator,
+	IndexUnionThingIterator, MatchesThingIterator, ThingIterator, UniqueEqualThingIterator,
-	UniqueUnionThingIterator,
+	UniqueJoinThingIterator, UniqueRangeThingIterator, UniqueUnionThingIterator,
 };
 use crate::idx::planner::knn::KnnPriorityList;
 use crate::idx::planner::plan::IndexOperator::Matches;
 use crate::idx::planner::plan::{IndexOperator, IndexOption, RangeValue};
-use crate::idx::planner::tree::{IndexRef, IndexesMap};
+use crate::idx::planner::tree::{IdiomPosition, IndexRef, IndexesMap};
 use crate::idx::planner::{IterationStage, KnnSet};
 use crate::idx::trees::mtree::MTreeIndex;
 use crate::idx::IndexKeyBase;
@@ -28,9 +29,8 @@ use std::collections::{HashMap, HashSet, VecDeque};
 use std::sync::Arc;
 use tokio::sync::RwLock;
 
-pub(super) type KnnEntry = (KnnPriorityList, Arc<Idiom>, Arc<Vec<Number>>, Distance);
+pub(super) type KnnEntry = (KnnPriorityList, Idiom, Arc<Vec<Number>>, Distance);
-pub(super) type KnnExpressions =
+pub(super) type KnnExpressions = HashMap<Arc<Expression>, (u32, Idiom, Arc<Vec<Number>>, Distance)>;
-	HashMap<Arc<Expression>, (u32, Arc<Idiom>, Arc<Vec<Number>>, Distance)>;
 
 #[derive(Clone)]
 pub(crate) struct QueryExecutor(Arc<InnerQueryExecutor>);
@@ -60,16 +60,17 @@ pub(super) enum IteratorEntry {
 }
 
 impl IteratorEntry {
-	pub(super) fn explain(&self, e: &mut HashMap<&str, Value>) -> IndexRef {
+	pub(super) fn explain(&self, ix_def: &[DefineIndexStatement]) -> Value {
 		match self {
-			Self::Single(_, io) => {
+			Self::Single(_, io) => io.explain(ix_def),
-				io.explain(e);
-				io.ix_ref()
-			}
 			Self::Range(_, ir, from, to) => {
+				let mut e = HashMap::default();
+				if let Some(ix) = ix_def.get(*ir as usize) {
+					e.insert("index", Value::from(ix.name.0.to_owned()));
+				}
 				e.insert("from", Value::from(from));
 				e.insert("to", Value::from(to));
-				*ir
+				Value::from(Object::from(e))
 			}
 		}
 	}
@@ -227,7 +228,7 @@ impl QueryExecutor {
 		!self.0.knn_entries.is_empty()
 	}
 
-	/// Returns `true` if either the expression is matching the current iterator.
+	/// Returns `true` if the expression is matching the current iterator.
 	pub(crate) fn is_iterator_expression(&self, ir: IteratorRef, exp: &Expression) -> bool {
 		match self.0.it_entries.get(ir as usize) {
 			Some(IteratorEntry::Single(e, ..)) => exp.eq(e.as_ref()),
@@ -238,14 +239,7 @@ impl QueryExecutor {
 
 	pub(crate) fn explain(&self, itr: IteratorRef) -> Value {
 		match self.0.it_entries.get(itr as usize) {
-			Some(ie) => {
+			Some(ie) => ie.explain(self.0.index_definitions.as_slice()),
-				let mut e = HashMap::default();
-				let ir = ie.explain(&mut e);
-				if let Some(ix) = self.0.index_definitions.get(ir as usize) {
-					e.insert("index", Value::from(ix.name.0.to_owned()));
-				}
-				Value::from(Object::from(e))
-			}
 			None => Value::None,
 		}
 	}
@@ -282,10 +276,12 @@ impl QueryExecutor {
 		it_ref: IteratorRef,
 		io: &IndexOption,
 	) -> Result<Option<ThingIterator>, Error> {
-		if let Some(ix) = self.0.index_definitions.get(io.ix_ref() as usize) {
+		if let Some(ix) = self.get_index_def(io.ix_ref()) {
 			match ix.index {
-				Index::Idx => Ok(Self::new_index_iterator(opt, ix, io.clone())),
+				Index::Idx => Ok(self.new_index_iterator(opt, it_ref, ix, io.clone()).await?),
-				Index::Uniq => Ok(Self::new_unique_index_iterator(opt, ix, io.clone())),
+				Index::Uniq => {
+					Ok(self.new_unique_index_iterator(opt, it_ref, ix, io.clone()).await?)
+				}
 				Index::Search {
 					..
 				} => self.new_search_index_iterator(it_ref, io.clone()).await,
@@ -296,20 +292,27 @@ impl QueryExecutor {
 		}
 	}
 
-	fn new_index_iterator(
+	async fn new_index_iterator(
+		&self,
 		opt: &Options,
+		it_ref: IteratorRef,
 		ix: &DefineIndexStatement,
 		io: IndexOption,
-	) -> Option<ThingIterator> {
+	) -> Result<Option<ThingIterator>, Error> {
-		match io.op() {
+		Ok(match io.op() {
-			IndexOperator::Equality(value) => {
+			IndexOperator::Equality(value) => Some(ThingIterator::IndexEqual(
-				Some(ThingIterator::IndexEqual(IndexEqualThingIterator::new(opt, ix, value)))
+				IndexEqualThingIterator::new(opt.ns(), opt.db(), &ix.what, &ix.name, value),
-			}
+			)),
-			IndexOperator::Union(value) => {
+			IndexOperator::Union(value) => Some(ThingIterator::IndexUnion(
-				Some(ThingIterator::IndexUnion(IndexUnionThingIterator::new(opt, ix, value)))
+				IndexUnionThingIterator::new(opt.ns(), opt.db(), &ix.what, &ix.name, value),
+			)),
+			IndexOperator::Join(ios) => {
+				let iterators = self.build_iterators(opt, it_ref, ios).await?;
+				let index_join = Box::new(IndexJoinThingIterator::new(opt, ix, iterators));
+				Some(ThingIterator::IndexJoin(index_join))
 			}
 			_ => None,
-		}
+		})
 	}
 
 	fn new_range_iterator(
@@ -319,16 +322,26 @@ impl QueryExecutor {
 		from: &RangeValue,
 		to: &RangeValue,
 	) -> Option<ThingIterator> {
-		if let Some(ix) = self.0.index_definitions.get(ir as usize) {
+		if let Some(ix) = self.get_index_def(ir) {
 			match ix.index {
 				Index::Idx => {
 					return Some(ThingIterator::IndexRange(IndexRangeThingIterator::new(
-						opt, ix, from, to,
+						opt.ns(),
+						opt.db(),
+						&ix.what,
+						&ix.name,
+						from,
+						to,
 					)))
 				}
 				Index::Uniq => {
 					return Some(ThingIterator::UniqueRange(UniqueRangeThingIterator::new(
-						opt, ix, from, to,
+						opt.ns(),
+						opt.db(),
+						&ix.what,
+						&ix.name,
+						from,
+						to,
 					)))
 				}
 				_ => {}
@@ -337,20 +350,27 @@ impl QueryExecutor {
 		None
 	}
 
-	fn new_unique_index_iterator(
+	async fn new_unique_index_iterator(
+		&self,
 		opt: &Options,
+		it_ref: IteratorRef,
 		ix: &DefineIndexStatement,
 		io: IndexOption,
-	) -> Option<ThingIterator> {
+	) -> Result<Option<ThingIterator>, Error> {
-		match io.op() {
+		Ok(match io.op() {
-			IndexOperator::Equality(value) => {
+			IndexOperator::Equality(value) => Some(ThingIterator::UniqueEqual(
-				Some(ThingIterator::UniqueEqual(UniqueEqualThingIterator::new(opt, ix, value)))
+				UniqueEqualThingIterator::new(opt.ns(), opt.db(), &ix.what, &ix.name, value),
-			}
+			)),
 			IndexOperator::Union(value) => {
 				Some(ThingIterator::UniqueUnion(UniqueUnionThingIterator::new(opt, ix, value)))
 			}
+			IndexOperator::Join(ios) => {
+				let iterators = self.build_iterators(opt, it_ref, ios).await?;
+				let unique_join = Box::new(UniqueJoinThingIterator::new(opt, ix, iterators));
+				Some(ThingIterator::UniqueJoin(unique_join))
+			}
 			_ => None,
-		}
+		})
 	}
 
 	async fn new_search_index_iterator(
@@ -381,40 +401,43 @@ impl QueryExecutor {
 		None
 	}
 
+	async fn build_iterators(
+		&self,
+		opt: &Options,
+		it_ref: IteratorRef,
+		ios: &[IndexOption],
+	) -> Result<VecDeque<ThingIterator>, Error> {
+		let mut iterators = VecDeque::with_capacity(ios.len());
+		for io in ios {
+			if let Some(it) = Box::pin(self.new_single_iterator(opt, it_ref, io)).await? {
+				iterators.push_back(it);
+			}
+		}
+		Ok(iterators)
+	}
+
+	fn get_index_def(&self, ir: IndexRef) -> Option<&DefineIndexStatement> {
+		self.0.index_definitions.get(ir as usize)
+	}
+
+	#[allow(clippy::too_many_arguments)]
 	pub(crate) async fn matches(
 		&self,
+		ctx: &Context<'_>,
+		opt: &Options,
 		txn: &Transaction,
 		thg: &Thing,
 		exp: &Expression,
-	) -> Result<Value, Error> {
+		l: Value,
-		// Otherwise, we look for the first possible index options, and evaluate the expression
+		r: Value,
-		// Does the record id match this executor's table?
+	) -> Result<bool, Error> {
-		if thg.tb.eq(&self.0.table) {
+		if let Some(ft) = self.0.exp_entries.get(exp) {
-			if let Some(ft) = self.0.exp_entries.get(exp) {
+			if let Some(ix_def) = self.get_index_def(ft.0.index_option.ix_ref()) {
-				let mut run = txn.lock().await;
+				if self.0.table.eq(&ix_def.what.0) {
-				let doc_key: Key = thg.into();
+					return self.matches_with_doc_id(txn, thg, ft).await;
-				if let Some(doc_id) =
-					ft.0.doc_ids.read().await.get_doc_id(&mut run, doc_key).await?
-				{
-					let term_goals = ft.0.terms_docs.len();
-					// If there is no terms, it can't be a match
-					if term_goals == 0 {
-						return Ok(Value::Bool(false));
-					}
-					for opt_td in ft.0.terms_docs.iter() {
-						if let Some((_, docs)) = opt_td {
-							if !docs.contains(doc_id) {
-								return Ok(Value::Bool(false));
-							}
-						} else {
-							// If one of the term is missing, it can't be a match
-							return Ok(Value::Bool(false));
-						}
-					}
-					return Ok(Value::Bool(true));
 				}
-				return Ok(Value::Bool(false));
 			}
+			return self.matches_with_value(ctx, opt, txn, ft, l, r).await;
 		}
 
 		// If no previous case were successful, we end up with a user error
@@ -423,6 +446,60 @@ impl QueryExecutor {
 		})
 	}
 
+	async fn matches_with_doc_id(
+		&self,
+		txn: &Transaction,
+		thg: &Thing,
+		ft: &FtEntry,
+	) -> Result<bool, Error> {
+		let mut run = txn.lock().await;
+		let doc_key: Key = thg.into();
+		if let Some(doc_id) = ft.0.doc_ids.read().await.get_doc_id(&mut run, doc_key).await? {
+			let term_goals = ft.0.terms_docs.len();
+			// If there is no terms, it can't be a match
+			if term_goals == 0 {
+				return Ok(false);
+			}
+			for opt_td in ft.0.terms_docs.iter() {
+				if let Some((_, docs)) = opt_td {
+					if !docs.contains(doc_id) {
+						return Ok(false);
+					}
+				} else {
+					// If one of the term is missing, it can't be a match
+					return Ok(false);
+				}
+			}
+			return Ok(true);
+		}
+		Ok(false)
+	}
+
+	async fn matches_with_value(
+		&self,
+		ctx: &Context<'_>,
+		opt: &Options,
+		txn: &Transaction,
+		ft: &FtEntry,
+		l: Value,
+		r: Value,
+	) -> Result<bool, Error> {
+		// If the query terms contains terms that are unknown in the index
+		// of if there is not terms in the query
+		// we are sure that it does not match any document
+		if !ft.0.query_terms_set.is_matchable() {
+			return Ok(false);
+		}
+		let v = match ft.0.index_option.id_pos() {
+			IdiomPosition::Left => r,
+			IdiomPosition::Right => l,
+		};
+		let terms = ft.0.terms.read().await;
+		// Extract the terms set from the record
+		let t = ft.0.analyzer.extract_indexing_terms(ctx, opt, txn, &terms, v).await?;
+		Ok(ft.0.query_terms_set.is_subset(&t))
+	}
+
 	fn get_ft_entry(&self, match_ref: &Value) -> Option<&FtEntry> {
 		if let Some(mr) = Self::get_match_ref(match_ref) {
 			self.0.mr_entries.get(&mr)
@@ -457,7 +534,7 @@ impl QueryExecutor {
 				.highlight(
 					&mut run,
 					thg,
-					&e.0.terms,
+					&e.0.query_terms_list,
 					prefix,
 					suffix,
 					partial,
@@ -478,7 +555,7 @@ impl QueryExecutor {
 	) -> Result<Value, Error> {
 		if let Some((e, ft)) = self.get_ft_entry_and_index(&match_ref) {
 			let mut run = txn.lock().await;
-			return ft.extract_offsets(&mut run, thg, &e.0.terms, partial).await;
+			return ft.extract_offsets(&mut run, thg, &e.0.query_terms_list, partial).await;
 		}
 		Ok(Value::None)
 	}
@@ -515,7 +592,10 @@ struct FtEntry(Arc<Inner>);
 struct Inner {
 	index_option: IndexOption,
 	doc_ids: Arc<RwLock<DocIds>>,
-	terms: Vec<Option<(TermId, u32)>>,
+	analyzer: Arc<Analyzer>,
+	query_terms_set: TermsSet,
+	query_terms_list: TermsList,
+	terms: Arc<RwLock<Terms>>,
 	terms_docs: TermsDocs,
 	scorer: Option<BM25Scorer>,
 }
@@ -529,14 +609,18 @@ impl FtEntry {
 		io: IndexOption,
 	) -> Result<Option<Self>, Error> {
 		if let Matches(qs, _) = io.op() {
-			let terms = ft.extract_terms(ctx, opt, txn, qs.to_owned()).await?;
+			let (terms_list, terms_set) =
+				ft.extract_querying_terms(ctx, opt, txn, qs.to_owned()).await?;
 			let mut tx = txn.lock().await;
-			let terms_docs = Arc::new(ft.get_terms_docs(&mut tx, &terms).await?);
+			let terms_docs = Arc::new(ft.get_terms_docs(&mut tx, &terms_list).await?);
 			Ok(Some(Self(Arc::new(Inner {
 				index_option: io,
 				doc_ids: ft.doc_ids(),
+				analyzer: ft.analyzer(),
+				query_terms_set: terms_set,
+				query_terms_list: terms_list,
 				scorer: ft.new_scorer(terms_docs.clone())?,
-				terms,
+				terms: ft.terms(),
 				terms_docs,
 			}))))
 		} else {

core/src/idx/planner/iterators.rs

@@ -7,7 +7,8 @@ use crate::idx::planner::plan::RangeValue;
 use crate::key::index::Index;
 use crate::kvs::{Key, Limit, ScanPage};
 use crate::sql::statements::DefineIndexStatement;
-use crate::sql::{Array, Thing, Value};
+use crate::sql::{Array, Ident, Thing, Value};
+use radix_trie::Trie;
 use std::collections::VecDeque;
 use std::sync::Arc;
 use tokio::sync::RwLock;
@@ -16,54 +17,76 @@ pub(crate) enum ThingIterator {
 	IndexEqual(IndexEqualThingIterator),
 	IndexRange(IndexRangeThingIterator),
 	IndexUnion(IndexUnionThingIterator),
+	IndexJoin(Box<IndexJoinThingIterator>),
 	UniqueEqual(UniqueEqualThingIterator),
 	UniqueRange(UniqueRangeThingIterator),
 	UniqueUnion(UniqueUnionThingIterator),
+	UniqueJoin(Box<UniqueJoinThingIterator>),
 	Matches(MatchesThingIterator),
 	Knn(DocIdsIterator),
 }
 
 impl ThingIterator {
-	pub(crate) async fn next_batch(
+	pub(crate) async fn next_batch<T: ThingCollector>(
 		&mut self,
 		tx: &Transaction,
 		size: u32,
-	) -> Result<Vec<(Thing, Option<DocId>)>, Error> {
+		collector: &mut T,
+	) -> Result<usize, Error> {
 		match self {
-			ThingIterator::IndexEqual(i) => i.next_batch(tx, size).await,
+			Self::IndexEqual(i) => i.next_batch(tx, size, collector).await,
-			ThingIterator::UniqueEqual(i) => i.next_batch(tx).await,
+			Self::UniqueEqual(i) => i.next_batch(tx, collector).await,
-			ThingIterator::IndexRange(i) => i.next_batch(tx, size).await,
+			Self::IndexRange(i) => i.next_batch(tx, size, collector).await,
-			ThingIterator::UniqueRange(i) => i.next_batch(tx, size).await,
+			Self::UniqueRange(i) => i.next_batch(tx, size, collector).await,
-			ThingIterator::IndexUnion(i) => i.next_batch(tx, size).await,
+			Self::IndexUnion(i) => i.next_batch(tx, size, collector).await,
-			ThingIterator::UniqueUnion(i) => i.next_batch(tx, size).await,
+			Self::UniqueUnion(i) => i.next_batch(tx, size, collector).await,
-			ThingIterator::Matches(i) => i.next_batch(tx, size).await,
+			Self::Matches(i) => i.next_batch(tx, size, collector).await,
-			ThingIterator::Knn(i) => i.next_batch(tx, size).await,
+			Self::Knn(i) => i.next_batch(tx, size, collector).await,
+			Self::IndexJoin(i) => Box::pin(i.next_batch(tx, size, collector)).await,
+			Self::UniqueJoin(i) => Box::pin(i.next_batch(tx, size, collector)).await,
 		}
 	}
 }
 
+pub(crate) trait ThingCollector {
+	fn add(&mut self, thing: Thing, doc_id: Option<DocId>);
+}
+
+impl ThingCollector for Vec<(Thing, Option<DocId>)> {
+	fn add(&mut self, thing: Thing, doc_id: Option<DocId>) {
+		self.push((thing, doc_id));
+	}
+}
+
+impl ThingCollector for VecDeque<(Thing, Option<DocId>)> {
+	fn add(&mut self, thing: Thing, doc_id: Option<DocId>) {
+		self.push_back((thing, doc_id));
+	}
+}
+
 pub(crate) struct IndexEqualThingIterator {
 	beg: Vec<u8>,
 	end: Vec<u8>,
 }
 
 impl IndexEqualThingIterator {
-	pub(super) fn new(opt: &Options, ix: &DefineIndexStatement, v: &Value) -> Self {
+	pub(super) fn new(ns: &str, db: &str, ix_what: &Ident, ix_name: &Ident, v: &Value) -> Self {
 		let a = Array::from(v.clone());
-		let beg = Index::prefix_ids_beg(opt.ns(), opt.db(), &ix.what, &ix.name, &a);
+		let beg = Index::prefix_ids_beg(ns, db, ix_what, ix_name, &a);
-		let end = Index::prefix_ids_end(opt.ns(), opt.db(), &ix.what, &ix.name, &a);
+		let end = Index::prefix_ids_end(ns, db, ix_what, ix_name, &a);
 		Self {
 			beg,
 			end,
 		}
 	}
 
-	async fn next_scan(
+	async fn next_scan<T: ThingCollector>(
 		txn: &Transaction,
 		beg: &mut Vec<u8>,
 		end: &[u8],
 		limit: u32,
-	) -> Result<Vec<(Thing, Option<DocId>)>, Error> {
+		collector: &mut T,
+	) -> Result<usize, Error> {
 		let min = beg.clone();
 		let max = end.to_owned();
 		let res = txn
@@ -83,16 +106,18 @@ impl IndexEqualThingIterator {
 			key.push(0x00);
 			*beg = key;
 		}
-		let res = res.iter().map(|(_, val)| (val.into(), None)).collect();
+		let count = res.len();
-		Ok(res)
+		res.into_iter().for_each(|(_, val)| collector.add(val.into(), None));
+		Ok(count)
 	}
 
-	async fn next_batch(
+	async fn next_batch<T: ThingCollector>(
 		&mut self,
 		txn: &Transaction,
 		limit: u32,
-	) -> Result<Vec<(Thing, Option<DocId>)>, Error> {
+		collector: &mut T,
-		Self::next_scan(txn, &mut self.beg, &self.end, limit).await
+	) -> Result<usize, Error> {
+		Self::next_scan(txn, &mut self.beg, &self.end, limit, collector).await
 	}
 }
 
@@ -146,47 +171,62 @@ pub(crate) struct IndexRangeThingIterator {
 
 impl IndexRangeThingIterator {
 	pub(super) fn new(
-		opt: &Options,
+		ns: &str,
-		ix: &DefineIndexStatement,
+		db: &str,
+		ix_what: &Ident,
+		ix_name: &Ident,
 		from: &RangeValue,
 		to: &RangeValue,
 	) -> Self {
-		let beg = Self::compute_beg(opt, ix, from);
+		let beg = Self::compute_beg(ns, db, ix_what, ix_name, from);
-		let end = Self::compute_end(opt, ix, to);
+		let end = Self::compute_end(ns, db, ix_what, ix_name, to);
 		Self {
 			r: RangeScan::new(beg, from.inclusive, end, to.inclusive),
 		}
 	}
 
-	fn compute_beg(opt: &Options, ix: &DefineIndexStatement, from: &RangeValue) -> Vec<u8> {
+	fn compute_beg(
+		ns: &str,
+		db: &str,
+		ix_what: &Ident,
+		ix_name: &Ident,
+		from: &RangeValue,
+	) -> Vec<u8> {
 		if from.value == Value::None {
-			return Index::prefix_beg(opt.ns(), opt.db(), &ix.what, &ix.name);
+			return Index::prefix_beg(ns, db, ix_what, ix_name);
 		}
 		let fd = Array::from(from.value.to_owned());
 		if from.inclusive {
-			Index::prefix_ids_beg(opt.ns(), opt.db(), &ix.what, &ix.name, &fd)
+			Index::prefix_ids_beg(ns, db, ix_what, ix_name, &fd)
 		} else {
-			Index::prefix_ids_end(opt.ns(), opt.db(), &ix.what, &ix.name, &fd)
+			Index::prefix_ids_end(ns, db, ix_what, ix_name, &fd)
 		}
 	}
 
-	fn compute_end(opt: &Options, ix: &DefineIndexStatement, to: &RangeValue) -> Vec<u8> {
+	fn compute_end(
+		ns: &str,
+		db: &str,
+		ix_what: &Ident,
+		ix_name: &Ident,
+		to: &RangeValue,
+	) -> Vec<u8> {
 		if to.value == Value::None {
-			return Index::prefix_end(opt.ns(), opt.db(), &ix.what, &ix.name);
+			return Index::prefix_end(ns, db, ix_what, ix_name);
 		}
 		let fd = Array::from(to.value.to_owned());
 		if to.inclusive {
-			Index::prefix_ids_end(opt.ns(), opt.db(), &ix.what, &ix.name, &fd)
+			Index::prefix_ids_end(ns, db, ix_what, ix_name, &fd)
 		} else {
-			Index::prefix_ids_beg(opt.ns(), opt.db(), &ix.what, &ix.name, &fd)
+			Index::prefix_ids_beg(ns, db, ix_what, ix_name, &fd)
 		}
 	}
 
-	async fn next_batch(
+	async fn next_batch<T: ThingCollector>(
 		&mut self,
 		txn: &Transaction,
 		limit: u32,
-	) -> Result<Vec<(Thing, Option<DocId>)>, Error> {
+		collector: &mut T,
+	) -> Result<usize, Error> {
 		let min = self.r.beg.clone();
 		let max = self.r.end.clone();
 		let res = txn
@@ -205,13 +245,14 @@ impl IndexRangeThingIterator {
 			self.r.beg.clone_from(key);
 			self.r.beg.push(0x00);
 		}
-		let mut r = Vec::with_capacity(res.len());
+		let mut count = 0;
 		for (k, v) in res {
 			if self.r.matches(&k) {
-				r.push((v.into(), None));
+				collector.add(v.into(), None);
+				count += 1;
 			}
 		}
-		Ok(r)
+		Ok(count)
 	}
 }
 
@@ -221,14 +262,14 @@ pub(crate) struct IndexUnionThingIterator {
 }
 
 impl IndexUnionThingIterator {
-	pub(super) fn new(opt: &Options, ix: &DefineIndexStatement, a: &Array) -> Self {
+	pub(super) fn new(ns: &str, db: &str, ix_what: &Ident, ix_name: &Ident, a: &Array) -> Self {
 		// We create a VecDeque to hold the prefix keys (begin and end) for each value in the array.
 		let mut values: VecDeque<(Vec<u8>, Vec<u8>)> =
 			a.0.iter()
 				.map(|v| {
 					let a = Array::from(v.clone());
-					let beg = Index::prefix_ids_beg(opt.ns(), opt.db(), &ix.what, &ix.name, &a);
+					let beg = Index::prefix_ids_beg(ns, db, ix_what, ix_name, &a);
-					let end = Index::prefix_ids_end(opt.ns(), opt.db(), &ix.what, &ix.name, &a);
+					let end = Index::prefix_ids_end(ns, db, ix_what, ix_name, &a);
 					(beg, end)
 				})
 				.collect();
@@ -239,19 +280,148 @@ impl IndexUnionThingIterator {
 		}
 	}
 
-	async fn next_batch(
+	async fn next_batch<T: ThingCollector>(
 		&mut self,
 		txn: &Transaction,
 		limit: u32,
-	) -> Result<Vec<(Thing, Option<DocId>)>, Error> {
+		collector: &mut T,
+	) -> Result<usize, Error> {
 		while let Some(r) = &mut self.current {
-			let res = IndexEqualThingIterator::next_scan(txn, &mut r.0, &r.1, limit).await?;
+			let count =
-			if !res.is_empty() {
+				IndexEqualThingIterator::next_scan(txn, &mut r.0, &r.1, limit, collector).await?;
-				return Ok(res);
+			if count != 0 {
+				return Ok(count);
 			}
 			self.current = self.values.pop_front();
 		}
-		Ok(vec![])
+		Ok(0)
+	}
+}
+
+struct JoinThingIterator {
+	ns: String,
+	db: String,
+	ix_what: Ident,
+	ix_name: Ident,
+	remote_iterators: VecDeque<ThingIterator>,
+	current_remote: Option<ThingIterator>,
+	current_remote_batch: VecDeque<(Thing, Option<DocId>)>,
+	current_local: Option<ThingIterator>,
+	distinct: Trie<Key, bool>,
+}
+
+impl JoinThingIterator {
+	pub(super) fn new(
+		opt: &Options,
+		ix: &DefineIndexStatement,
+		remote_iterators: VecDeque<ThingIterator>,
+	) -> Self {
+		Self {
+			ns: opt.ns().to_string(),
+			db: opt.db().to_string(),
+			ix_what: ix.what.clone(),
+			ix_name: ix.name.clone(),
+			current_remote: None,
+			current_remote_batch: VecDeque::with_capacity(0),
+			remote_iterators,
+			current_local: None,
+			distinct: Default::default(),
+		}
+	}
+}
+
+impl JoinThingIterator {
+	async fn next_current_remote_batch(
+		&mut self,
+		tx: &Transaction,
+		limit: u32,
+	) -> Result<bool, Error> {
+		loop {
+			if let Some(it) = &mut self.current_remote {
+				self.current_remote_batch.clear();
+				if it.next_batch(tx, limit, &mut self.current_remote_batch).await? > 0 {
+					return Ok(true);
+				}
+			}
+			self.current_remote = self.remote_iterators.pop_front();
+			if self.current_remote.is_none() {
+				return Ok(false);
+			}
+		}
+	}
+
+	async fn next_current_local<F>(
+		&mut self,
+		tx: &Transaction,
+		limit: u32,
+		new_iter: F,
+	) -> Result<bool, Error>
+	where
+		F: Fn(&str, &str, &Ident, &Ident, Value) -> ThingIterator,
+	{
+		loop {
+			while let Some((thing, _)) = self.current_remote_batch.pop_front() {
+				let k: Key = (&thing).into();
+				let value = Value::from(thing);
+				if self.distinct.insert(k, true).is_none() {
+					self.current_local =
+						Some(new_iter(&self.ns, &self.db, &self.ix_what, &self.ix_name, value));
+					return Ok(true);
+				}
+			}
+			if !self.next_current_remote_batch(tx, limit).await? {
+				break;
+			}
+		}
+		Ok(false)
+	}
+
+	async fn next_batch<T: ThingCollector, F>(
+		&mut self,
+		tx: &Transaction,
+		limit: u32,
+		collector: &mut T,
+		new_iter: F,
+	) -> Result<usize, Error>
+	where
+		F: Fn(&str, &str, &Ident, &Ident, Value) -> ThingIterator + Copy,
+	{
+		loop {
+			if let Some(current_local) = &mut self.current_local {
+				let n = current_local.next_batch(tx, limit, collector).await?;
+				if n > 0 {
+					return Ok(n);
+				}
+			}
+			if !self.next_current_local(tx, limit, new_iter).await? {
+				return Ok(0);
+			}
+		}
+	}
+}
+
+pub(crate) struct IndexJoinThingIterator(JoinThingIterator);
+
+impl IndexJoinThingIterator {
+	pub(super) fn new(
+		opt: &Options,
+		ix: &DefineIndexStatement,
+		remote_iterators: VecDeque<ThingIterator>,
+	) -> Self {
+		Self(JoinThingIterator::new(opt, ix, remote_iterators))
+	}
+
+	async fn next_batch<T: ThingCollector>(
+		&mut self,
+		tx: &Transaction,
+		limit: u32,
+		collector: &mut T,
+	) -> Result<usize, Error> {
+		let new_iter = |ns: &str, db: &str, ix_what: &Ident, ix_name: &Ident, value: Value| {
+			let it = IndexEqualThingIterator::new(ns, db, ix_what, ix_name, &value);
+			ThingIterator::IndexEqual(it)
+		};
+		self.0.next_batch(tx, limit, collector, new_iter).await
 	}
 }
 
@@ -260,24 +430,27 @@ pub(crate) struct UniqueEqualThingIterator {
 }
 
 impl UniqueEqualThingIterator {
-	pub(super) fn new(opt: &Options, ix: &DefineIndexStatement, v: &Value) -> Self {
+	pub(super) fn new(ns: &str, db: &str, ix_what: &Ident, ix_name: &Ident, v: &Value) -> Self {
 		let a = Array::from(v.to_owned());
-		let key = Index::new(opt.ns(), opt.db(), &ix.what, &ix.name, &a, None).into();
+		let key = Index::new(ns, db, ix_what, ix_name, &a, None).into();
 		Self {
 			key: Some(key),
 		}
 	}
 
-	async fn next_batch(
+	async fn next_batch<T: ThingCollector>(
 		&mut self,
 		txn: &Transaction,
-	) -> Result<Vec<(Thing, Option<DocId>)>, Error> {
+		collector: &mut T,
+	) -> Result<usize, Error> {
+		let mut count = 0;
 		if let Some(key) = self.key.take() {
 			if let Some(val) = txn.lock().await.get(key).await? {
-				return Ok(vec![(val.into(), None)]);
+				collector.add(val.into(), None);
+				count += 1;
 			}
 		}
-		Ok(vec![])
+		Ok(count)
 	}
 }
 
@@ -288,51 +461,59 @@ pub(crate) struct UniqueRangeThingIterator {
 
 impl UniqueRangeThingIterator {
 	pub(super) fn new(
-		opt: &Options,
+		ns: &str,
-		ix: &DefineIndexStatement,
+		db: &str,
+		ix_what: &Ident,
+		ix_name: &Ident,
 		from: &RangeValue,
 		to: &RangeValue,
 	) -> Self {
-		let beg = Self::compute_beg(opt, ix, from);
+		let beg = Self::compute_beg(ns, db, ix_what, ix_name, from);
-		let end = Self::compute_end(opt, ix, to);
+		let end = Self::compute_end(ns, db, ix_what, ix_name, to);
 		Self {
 			r: RangeScan::new(beg, from.inclusive, end, to.inclusive),
 			done: false,
 		}
 	}
 
-	fn compute_beg(opt: &Options, ix: &DefineIndexStatement, from: &RangeValue) -> Vec<u8> {
+	fn compute_beg(
+		ns: &str,
+		db: &str,
+		ix_what: &Ident,
+		ix_name: &Ident,
+		from: &RangeValue,
+	) -> Vec<u8> {
 		if from.value == Value::None {
-			return Index::prefix_beg(opt.ns(), opt.db(), &ix.what, &ix.name);
+			return Index::prefix_beg(ns, db, ix_what, ix_name);
 		}
-		Index::new(
+		Index::new(ns, db, ix_what, ix_name, &Array::from(from.value.to_owned()), None)
-			opt.ns(),
-			opt.db(),
-			&ix.what,
-			&ix.name,
-			&Array::from(from.value.to_owned()),
-			None,
-		)
-		.encode()
-		.unwrap()
-	}
-
-	fn compute_end(opt: &Options, ix: &DefineIndexStatement, to: &RangeValue) -> Vec<u8> {
-		if to.value == Value::None {
-			return Index::prefix_end(opt.ns(), opt.db(), &ix.what, &ix.name);
-		}
-		Index::new(opt.ns(), opt.db(), &ix.what, &ix.name, &Array::from(to.value.to_owned()), None)
 			.encode()
 			.unwrap()
 	}
 
-	async fn next_batch(
+	fn compute_end(
+		ns: &str,
+		db: &str,
+		ix_what: &Ident,
+		ix_name: &Ident,
+		to: &RangeValue,
+	) -> Vec<u8> {
+		if to.value == Value::None {
+			return Index::prefix_end(ns, db, ix_what, ix_name);
+		}
+		Index::new(ns, db, ix_what, ix_name, &Array::from(to.value.to_owned()), None)
+			.encode()
+			.unwrap()
+	}
+
+	async fn next_batch<T: ThingCollector>(
 		&mut self,
 		txn: &Transaction,
 		mut limit: u32,
-	) -> Result<Vec<(Thing, Option<DocId>)>, Error> {
+		collector: &mut T,
+	) -> Result<usize, Error> {
 		if self.done {
-			return Ok(vec![]);
+			return Ok(0);
 		}
 		let min = self.r.beg.clone();
 		let max = self.r.end.clone();
@@ -347,26 +528,27 @@ impl UniqueRangeThingIterator {
 				limit,
 			)
 			.await?;
-		let res = res.values;
+		let mut count = 0;
-		let mut r = Vec::with_capacity(res.len());
+		for (k, v) in res.values {
-		for (k, v) in res {
 			limit -= 1;
 			if limit == 0 {
 				self.r.beg = k;
-				return Ok(r);
+				return Ok(count);
 			}
 			if self.r.matches(&k) {
-				r.push((v.into(), None));
+				collector.add(v.into(), None);
+				count += 1;
 			}
 		}
 		let end = self.r.end.clone();
 		if self.r.matches(&end) {
 			if let Some(v) = tx.get(end).await? {
-				r.push((v.into(), None));
+				collector.add(v.into(), None);
+				count += 1;
 			}
 		}
 		self.done = true;
-		Ok(r)
+		Ok(count)
 	}
 }
 
@@ -390,22 +572,49 @@ impl UniqueUnionThingIterator {
 		}
 	}
 
-	async fn next_batch(
+	async fn next_batch<T: ThingCollector>(
 		&mut self,
 		txn: &Transaction,
 		limit: u32,
-	) -> Result<Vec<(Thing, Option<DocId>)>, Error> {
+		collector: &mut T,
+	) -> Result<usize, Error> {
 		let mut run = txn.lock().await;
-		let mut res = vec![];
+		let mut count = 0;
 		while let Some(key) = self.keys.pop_front() {
 			if let Some(val) = run.get(key).await? {
-				res.push((val.into(), None));
+				collector.add(val.into(), None);
-			}
+				count += 1;
-			if res.len() >= limit as usize {
+				if count >= limit {
-				return Ok(res);
+					break;
+				}
 			}
 		}
-		Ok(res)
+		Ok(count as usize)
+	}
+}
+
+pub(crate) struct UniqueJoinThingIterator(JoinThingIterator);
+
+impl UniqueJoinThingIterator {
+	pub(super) fn new(
+		opt: &Options,
+		ix: &DefineIndexStatement,
+		remote_iterators: VecDeque<ThingIterator>,
+	) -> Self {
+		Self(JoinThingIterator::new(opt, ix, remote_iterators))
+	}
+
+	async fn next_batch<T: ThingCollector>(
+		&mut self,
+		tx: &Transaction,
+		limit: u32,
+		collector: &mut T,
+	) -> Result<usize, Error> {
+		let new_iter = |ns: &str, db: &str, ix_what: &Ident, ix_name: &Ident, value: Value| {
+			let it = UniqueEqualThingIterator::new(ns, db, ix_what, ix_name, &value);
+			ThingIterator::UniqueEqual(it)
+		};
+		self.0.next_batch(tx, limit, collector, new_iter).await
 	}
 }
 
@@ -421,24 +630,25 @@ impl MatchesThingIterator {
 		})
 	}
 
-	async fn next_batch(
+	async fn next_batch<T: ThingCollector>(
 		&mut self,
 		txn: &Transaction,
-		mut limit: u32,
+		limit: u32,
-	) -> Result<Vec<(Thing, Option<DocId>)>, Error> {
+		collector: &mut T,
-		let mut res = vec![];
+	) -> Result<usize, Error> {
+		let mut count = 0;
 		if let Some(hits) = &mut self.hits {
 			let mut run = txn.lock().await;
-			while limit > 0 {
+			while limit > count {
 				if let Some((thg, doc_id)) = hits.next(&mut run).await? {
-					res.push((thg, Some(doc_id)));
+					collector.add(thg, Some(doc_id));
+					count += 1;
 				} else {
 					break;
 				}
-				limit -= 1;
 			}
 		}
-		Ok(res)
+		Ok(count as usize)
 	}
 }
 
@@ -454,25 +664,26 @@ impl DocIdsIterator {
 			res,
 		}
 	}
-	async fn next_batch(
+	async fn next_batch<T: ThingCollector>(
 		&mut self,
 		txn: &Transaction,
-		mut limit: u32,
+		limit: u32,
-	) -> Result<Vec<(Thing, Option<DocId>)>, Error> {
+		collector: &mut T,
-		let mut res = vec![];
+	) -> Result<usize, Error> {
 		let mut tx = txn.lock().await;
-		while limit > 0 {
+		let mut count = 0;
+		while limit > count {
 			if let Some(doc_id) = self.res.pop_front() {
 				if let Some(doc_key) =
 					self.doc_ids.read().await.get_doc_key(&mut tx, doc_id).await?
 				{
-					res.push((doc_key.into(), Some(doc_id)));
+					collector.add(doc_key.into(), Some(doc_id));
-					limit -= 1;
+					count += 1;
 				}
 			} else {
 				break;
 			}
 		}
-		Ok(res)
+		Ok(count as usize)
 	}
 }

core/src/idx/planner/mod.rs

@@ -53,7 +53,6 @@ impl<'a> QueryPlanner<'a> {
 	) -> Result<(), Error> {
 		let mut is_table_iterator = false;
 		let mut is_knn = false;
-		let t = Arc::new(t);
 		match Tree::build(ctx, self.opt, txn, &t, self.cond, self.with).await? {
 			Some(tree) => {
 				is_knn = is_knn || !tree.knn_expressions.is_empty();
@@ -117,7 +116,7 @@ impl<'a> QueryPlanner<'a> {
 
 	fn add(
 		&mut self,
-		tb: Arc<Table>,
+		tb: Table,
 		irf: Option<IteratorRef>,
 		exe: InnerQueryExecutor,
 		it: &mut Iterator,

core/src/idx/planner/plan.rs

@@ -1,6 +1,7 @@
 use crate::err::Error;
 use crate::idx::ft::MatchRef;
-use crate::idx::planner::tree::{GroupRef, IndexRef, Node};
+use crate::idx::planner::tree::{GroupRef, IdiomPosition, IndexRef, Node};
+use crate::sql::statements::DefineIndexStatement;
 use crate::sql::with::With;
 use crate::sql::{Array, Idiom, Object};
 use crate::sql::{Expression, Operator, Value};
@@ -162,48 +163,53 @@ pub(super) enum Plan {
 	SingleIndexRange(IndexRef, UnionRangeQueryBuilder),
 }
 
-#[derive(Debug, Clone, Eq, PartialEq, Hash)]
+#[derive(Debug, Eq, PartialEq, Hash, Clone)]
-pub(crate) struct IndexOption(Arc<Inner>);
+pub(super) struct IndexOption {
-
+	/// A reference o the index definition
-#[derive(Debug, Eq, PartialEq, Hash)]
-pub(super) struct Inner {
 	ir: IndexRef,
-	id: Arc<Idiom>,
+	id: Idiom,
-	op: IndexOperator,
+	id_pos: IdiomPosition,
+	op: Arc<IndexOperator>,
 }
 
 #[derive(Debug, Eq, PartialEq, Hash)]
 pub(super) enum IndexOperator {
 	Equality(Value),
 	Union(Array),
+	Join(Vec<IndexOption>),
 	RangePart(Operator, Value),
 	Matches(String, Option<MatchRef>),
 	Knn(Array, u32),
 }
 
 impl IndexOption {
-	pub(super) fn new(ir: IndexRef, id: Arc<Idiom>, op: IndexOperator) -> Self {
+	pub(super) fn new(ir: IndexRef, id: Idiom, id_pos: IdiomPosition, op: IndexOperator) -> Self {
-		Self(Arc::new(Inner {
+		Self {
 			ir,
 			id,
-			op,
+			id_pos,
-		}))
+			op: Arc::new(op),
+		}
 	}
 
 	pub(super) fn require_distinct(&self) -> bool {
-		matches!(self.0.op, IndexOperator::Union(_))
+		matches!(self.op.as_ref(), IndexOperator::Union(_))
 	}
 
 	pub(super) fn ix_ref(&self) -> IndexRef {
-		self.0.ir
+		self.ir
 	}
 
 	pub(super) fn op(&self) -> &IndexOperator {
-		&self.0.op
+		self.op.as_ref()
 	}
 
 	pub(super) fn id_ref(&self) -> &Idiom {
-		self.0.id.as_ref()
+		&self.id
+	}
+
+	pub(super) fn id_pos(&self) -> IdiomPosition {
+		self.id_pos
 	}
 
 	fn reduce_array(v: &Value) -> Value {
@@ -215,7 +221,11 @@ impl IndexOption {
 		v.clone()
 	}
 
-	pub(crate) fn explain(&self, e: &mut HashMap<&str, Value>) {
+	pub(crate) fn explain(&self, ix_def: &[DefineIndexStatement]) -> Value {
+		let mut e = HashMap::new();
+		if let Some(ix) = ix_def.get(self.ir as usize) {
+			e.insert("index", Value::from(ix.name.0.to_owned()));
+		}
 		match self.op() {
 			IndexOperator::Equality(v) => {
 				e.insert("operator", Value::from(Operator::Equal.to_string()));
@@ -225,6 +235,15 @@ impl IndexOption {
 				e.insert("operator", Value::from("union"));
 				e.insert("value", Value::Array(a.clone()));
 			}
+			IndexOperator::Join(ios) => {
+				e.insert("operator", Value::from("join"));
+				let mut joins = Vec::with_capacity(ios.len());
+				for io in ios {
+					joins.push(io.explain(ix_def));
+				}
+				let joins = Value::from(joins);
+				e.insert("joins", joins);
+			}
 			IndexOperator::Matches(qs, a) => {
 				e.insert("operator", Value::from(Operator::Matches(*a).to_string()));
 				e.insert("value", Value::from(qs.to_owned()));
@@ -238,6 +257,7 @@ impl IndexOption {
 				e.insert("value", Value::Array(a.clone()));
 			}
 		};
+		Value::from(e)
 	}
 }
 
@@ -391,23 +411,25 @@ impl UnionRangeQueryBuilder {
 #[cfg(test)]
 mod tests {
 	use crate::idx::planner::plan::{IndexOperator, IndexOption, RangeValue};
+	use crate::idx::planner::tree::IdiomPosition;
 	use crate::sql::{Array, Idiom, Value};
 	use crate::syn::Parse;
 	use std::collections::HashSet;
-	use std::sync::Arc;
 
 	#[test]
 	fn test_hash_index_option() {
 		let mut set = HashSet::new();
 		let io1 = IndexOption::new(
 			1,
-			Arc::new(Idiom::parse("test")),
+			Idiom::parse("test"),
+			IdiomPosition::Right,
 			IndexOperator::Equality(Value::Array(Array::from(vec!["test"]))),
 		);
 
 		let io2 = IndexOption::new(
 			1,
-			Arc::new(Idiom::parse("test")),
+			Idiom::parse("test"),
+			IdiomPosition::Right,
 			IndexOperator::Equality(Value::Array(Array::from(vec!["test"]))),
 		);
 

core/src/idx/planner/tree.rs

@@ -3,10 +3,11 @@ use crate::dbs::{Options, Transaction};
 use crate::err::Error;
 use crate::idx::planner::executor::KnnExpressions;
 use crate::idx::planner::plan::{IndexOperator, IndexOption};
+use crate::kvs;
 use crate::sql::index::{Distance, Index};
-use crate::sql::statements::DefineIndexStatement;
+use crate::sql::statements::{DefineFieldStatement, DefineIndexStatement};
 use crate::sql::{
-	Array, Cond, Expression, Idiom, Number, Operator, Part, Subquery, Table, Value, With,
+	Array, Cond, Expression, Idiom, Kind, Number, Operator, Part, Subquery, Table, Value, With,
 };
 use async_recursion::async_recursion;
 use std::collections::HashMap;
@@ -51,16 +52,26 @@ struct TreeBuilder<'a> {
 	txn: &'a Transaction,
 	table: &'a Table,
 	with: &'a Option<With>,
-	indexes: Option<Arc<[DefineIndexStatement]>>,
+	schemas: HashMap<Table, SchemaCache>,
+	idioms_indexes: HashMap<Table, HashMap<Idiom, LocalIndexRefs>>,
 	resolved_expressions: HashMap<Arc<Expression>, ResolvedExpression>,
-	resolved_idioms: HashMap<Arc<Idiom>, Arc<Idiom>>,
+	resolved_idioms: HashMap<Idiom, Node>,
-	idioms_indexes: HashMap<Arc<Idiom>, Option<Arc<Vec<IndexRef>>>>,
 	index_map: IndexesMap,
 	with_indexes: Vec<IndexRef>,
 	knn_expressions: KnnExpressions,
+	idioms_record_options: HashMap<Idiom, RecordOptions>,
 	group_sequence: GroupRef,
 }
 
+#[derive(Debug, Clone, Eq, PartialEq, Hash)]
+pub(super) struct RecordOptions {
+	locals: LocalIndexRefs,
+	remotes: RemoteIndexRefs,
+}
+
+pub(super) type LocalIndexRefs = Vec<IndexRef>;
+pub(super) type RemoteIndexRefs = Arc<Vec<(Idiom, LocalIndexRefs)>>;
+
 impl<'a> TreeBuilder<'a> {
 	fn new(
 		ctx: &'a Context<'_>,
@@ -79,28 +90,28 @@ impl<'a> TreeBuilder<'a> {
 			txn,
 			table,
 			with,
-			indexes: None,
+			schemas: Default::default(),
+			idioms_indexes: Default::default(),
 			resolved_expressions: Default::default(),
 			resolved_idioms: Default::default(),
-			idioms_indexes: Default::default(),
 			index_map: Default::default(),
 			with_indexes,
 			knn_expressions: Default::default(),
+			idioms_record_options: Default::default(),
 			group_sequence: 0,
 		}
 	}
 
-	async fn lazy_cache_indexes(&mut self) -> Result<(), Error> {
+	async fn lazy_load_schema_resolver(
-		if self.indexes.is_none() {
+		&mut self,
-			let indexes = self
+		tx: &mut kvs::Transaction,
-				.txn
+		table: &Table,
-				.clone()
+	) -> Result<(), Error> {
-				.lock()
+		if self.schemas.contains_key(table) {
-				.await
+			return Ok(());
-				.all_tb_indexes(self.opt.ns(), self.opt.db(), &self.table.0)
-				.await?;
-			self.indexes = Some(indexes);
 		}
+		let l = SchemaCache::new(self.opt, table, tx).await?;
+		self.schemas.insert(table.clone(), l);
 		Ok(())
 	}
 
@@ -139,11 +150,8 @@ impl<'a> TreeBuilder<'a> {
 
 	async fn eval_idiom(&mut self, group: GroupRef, i: &Idiom) -> Result<Node, Error> {
 		// Check if the idiom has already been resolved
-		if let Some(i) = self.resolved_idioms.get(i) {
+		if let Some(node) = self.resolved_idioms.get(i).cloned() {
-			if let Some(Some(irs)) = self.idioms_indexes.get(i).cloned() {
+			return Ok(node);
-				return Ok(Node::IndexedField(i.clone(), irs));
-			}
-			return Ok(Node::NonIndexedField(i.clone()));
 		};
 
 		// Compute the idiom value if it is a param
@@ -154,42 +162,100 @@ impl<'a> TreeBuilder<'a> {
 			}
 		}
 
-		self.lazy_cache_indexes().await?;
+		let n = self.resolve_idiom(i).await?;
+		self.resolved_idioms.insert(i.clone(), n.clone());
 
-		let i = Arc::new(i.clone());
+		Ok(n)
-
-		self.resolved_idioms.insert(i.clone(), i.clone());
-
-		// Try to detect if it matches an index
-		if let Some(irs) = self.resolve_indexes(&i) {
-			return Ok(Node::IndexedField(i.clone(), irs));
-		}
-
-		Ok(Node::NonIndexedField(i))
 	}
 
-	fn resolve_indexes(&mut self, i: &Arc<Idiom>) -> Option<Arc<Vec<IndexRef>>> {
+	async fn resolve_idiom(&mut self, i: &Idiom) -> Result<Node, Error> {
-		let mut res = None;
+		let mut tx = self.txn.lock().await;
-		if let Some(indexes) = &self.indexes {
+		self.lazy_load_schema_resolver(&mut tx, self.table).await?;
-			let mut irs = Vec::new();
+
-			for ix in indexes.as_ref() {
+		// Try to detect if it matches an index
-				if ix.cols.len() == 1 && ix.cols[0].eq(i) {
+		if let Some(schema) = self.schemas.get(self.table).cloned() {
-					let ixr = self.index_map.definitions.len() as IndexRef;
+			let irs = self.resolve_indexes(self.table, i, &schema);
-					if let Some(With::Index(ixs)) = self.with {
-						if ixs.contains(&ix.name.0) {
-							self.with_indexes.push(ixr);
-						}
-					}
-					self.index_map.definitions.push(ix.clone());
-					irs.push(ixr);
-				}
-			}
 			if !irs.is_empty() {
-				res = Some(Arc::new(irs));
+				return Ok(Node::IndexedField(i.clone(), irs));
+			}
+			// Try to detect an indexed record field
+			if let Some(ro) = self.resolve_record_field(&mut tx, schema.fields.as_ref(), i).await? {
+				return Ok(Node::RecordField(i.clone(), ro));
 			}
 		}
-		self.idioms_indexes.insert(i.clone(), res.clone());
+		Ok(Node::NonIndexedField(i.clone()))
-		res
+	}
+
+	fn resolve_indexes(&mut self, t: &Table, i: &Idiom, schema: &SchemaCache) -> Vec<IndexRef> {
+		if let Some(m) = self.idioms_indexes.get(t) {
+			if let Some(irs) = m.get(i).cloned() {
+				return irs;
+			}
+		}
+		let mut irs = Vec::new();
+		for ix in schema.indexes.iter() {
+			if ix.cols.len() == 1 && ix.cols[0].eq(i) {
+				let ixr = self.index_map.definitions.len() as IndexRef;
+				if let Some(With::Index(ixs)) = self.with {
+					if ixs.contains(&ix.name.0) {
+						self.with_indexes.push(ixr);
+					}
+				}
+				self.index_map.definitions.push(ix.clone());
+				irs.push(ixr);
+			}
+		}
+		if let Some(e) = self.idioms_indexes.get_mut(t) {
+			e.insert(i.clone(), irs.clone());
+		} else {
+			self.idioms_indexes.insert(t.clone(), HashMap::from([(i.clone(), irs.clone())]));
+		}
+		irs
+	}
+
+	async fn resolve_record_field(
+		&mut self,
+		tx: &mut kvs::Transaction,
+		fields: &[DefineFieldStatement],
+		idiom: &Idiom,
+	) -> Result<Option<RecordOptions>, Error> {
+		for field in fields.iter() {
+			if let Some(Kind::Record(tables)) = &field.kind {
+				if idiom.starts_with(&field.name.0) {
+					let (local_field, remote_field) = idiom.0.split_at(field.name.0.len());
+					if remote_field.is_empty() {
+						return Ok(None);
+					}
+					let local_field = Idiom::from(local_field);
+					self.lazy_load_schema_resolver(tx, self.table).await?;
+					let locals;
+					if let Some(shema) = self.schemas.get(self.table).cloned() {
+						locals = self.resolve_indexes(self.table, &local_field, &shema);
+					} else {
+						return Ok(None);
+					}
+
+					let remote_field = Idiom::from(remote_field);
+					let mut remotes = vec![];
+					for table in tables {
+						self.lazy_load_schema_resolver(tx, table).await?;
+						if let Some(shema) = self.schemas.get(table).cloned() {
+							let remote_irs = self.resolve_indexes(table, &remote_field, &shema);
+							remotes.push((remote_field.clone(), remote_irs));
+						} else {
+							return Ok(None);
+						}
+					}
+					let ro = RecordOptions {
+						locals,
+						remotes: Arc::new(remotes),
+					};
+					self.idioms_record_options.insert(idiom.clone(), ro.clone());
+					return Ok(Some(ro));
+				}
+			}
+		}
+		Ok(None)
 	}
 
 	async fn eval_expression(&mut self, group: GroupRef, e: &Expression) -> Result<Node, Error> {
@@ -210,23 +276,25 @@ impl<'a> TreeBuilder<'a> {
 				let left = Arc::new(self.eval_value(group, l).await?);
 				let right = Arc::new(self.eval_value(group, r).await?);
 				let mut io = None;
-				if let Some((id, irs)) = left.is_indexed_field() {
+				if let Some((id, local_irs, remote_irs)) = left.is_indexed_field() {
-					io = self.lookup_index_option(
+					io = self.lookup_index_options(
-						irs.as_slice(),
 						o,
 						id,
 						&right,
 						&exp,
 						IdiomPosition::Left,
+						local_irs,
+						remote_irs,
 					)?;
-				} else if let Some((id, irs)) = right.is_indexed_field() {
+				} else if let Some((id, local_irs, remote_irs)) = right.is_indexed_field() {
-					io = self.lookup_index_option(
+					io = self.lookup_index_options(
-						irs.as_slice(),
 						o,
 						id,
 						&left,
 						&exp,
 						IdiomPosition::Right,
+						local_irs,
+						remote_irs,
 					)?;
 				} else if let Some(id) = left.is_non_indexed_field() {
 					self.eval_knn(id, &right, &exp)?;
@@ -236,7 +304,7 @@ impl<'a> TreeBuilder<'a> {
 				let re = ResolvedExpression {
 					group,
 					exp: exp.clone(),
-					io: io.clone(),
+					io,
 					left: left.clone(),
 					right: right.clone(),
 				};
@@ -246,11 +314,41 @@ impl<'a> TreeBuilder<'a> {
 		}
 	}
 
+	#[allow(clippy::too_many_arguments)]
+	fn lookup_index_options(
+		&mut self,
+		o: &Operator,
+		id: &Idiom,
+		node: &Node,
+		exp: &Arc<Expression>,
+		p: IdiomPosition,
+		local_irs: LocalIndexRefs,
+		remote_irs: Option<RemoteIndexRefs>,
+	) -> Result<Option<IndexOption>, Error> {
+		if let Some(remote_irs) = remote_irs {
+			let mut remote_ios = Vec::with_capacity(remote_irs.len());
+			for (id, irs) in remote_irs.iter() {
+				if let Some(io) = self.lookup_index_option(irs.as_slice(), o, id, node, exp, p)? {
+					remote_ios.push(io);
+				} else {
+					return Ok(None);
+				}
+			}
+			if let Some(ir) = self.lookup_join_index_ref(local_irs.as_slice()) {
+				let io = IndexOption::new(ir, id.clone(), p, IndexOperator::Join(remote_ios));
+				return Ok(Some(io));
+			}
+			return Ok(None);
+		}
+		let io = self.lookup_index_option(local_irs.as_slice(), o, id, node, exp, p)?;
+		Ok(io)
+	}
+
 	fn lookup_index_option(
 		&mut self,
 		irs: &[IndexRef],
 		op: &Operator,
-		id: Arc<Idiom>,
+		id: &Idiom,
 		n: &Node,
 		e: &Arc<Expression>,
 		p: IdiomPosition,
@@ -263,10 +361,10 @@ impl<'a> TreeBuilder<'a> {
 					Index::Search {
 						..
 					} => Self::eval_matches_operator(op, n),
-					Index::MTree(_) => self.eval_indexed_knn(e, op, n, id.clone())?,
+					Index::MTree(_) => self.eval_indexed_knn(e, op, n, id)?,
 				};
 				if let Some(op) = op {
-					let io = IndexOption::new(*ir, id, op);
+					let io = IndexOption::new(*ir, id.clone(), p, op);
 					self.index_map.options.push((e.clone(), io.clone()));
 					return Ok(Some(io));
 				}
@@ -274,6 +372,19 @@ impl<'a> TreeBuilder<'a> {
 		}
 		Ok(None)
 	}
+
+	fn lookup_join_index_ref(&self, irs: &[IndexRef]) -> Option<IndexRef> {
+		for ir in irs {
+			if let Some(ix) = self.index_map.definitions.get(*ir as usize) {
+				match &ix.index {
+					Index::Idx | Index::Uniq => return Some(*ir),
+					_ => {}
+				};
+			}
+		}
+		None
+	}
+
 	fn eval_matches_operator(op: &Operator, n: &Node) -> Option<IndexOperator> {
 		if let Some(v) = n.is_computed() {
 			if let Operator::Matches(mr) = op {
@@ -288,14 +399,14 @@ impl<'a> TreeBuilder<'a> {
 		exp: &Arc<Expression>,
 		op: &Operator,
 		n: &Node,
-		id: Arc<Idiom>,
+		id: &Idiom,
 	) -> Result<Option<IndexOperator>, Error> {
 		if let Operator::Knn(k, d) = op {
 			if let Node::Computed(v) = n {
 				let vec: Vec<Number> = v.as_ref().try_into()?;
 				self.knn_expressions.insert(
 					exp.clone(),
-					(*k, id, Arc::new(vec), d.clone().unwrap_or(Distance::Euclidean)),
+					(*k, id.clone(), Arc::new(vec), d.clone().unwrap_or(Distance::Euclidean)),
 				);
 				if let Value::Array(a) = v.as_ref() {
 					match d {
@@ -310,13 +421,13 @@ impl<'a> TreeBuilder<'a> {
 		Ok(None)
 	}
 
-	fn eval_knn(&mut self, id: Arc<Idiom>, val: &Node, exp: &Arc<Expression>) -> Result<(), Error> {
+	fn eval_knn(&mut self, id: &Idiom, val: &Node, exp: &Arc<Expression>) -> Result<(), Error> {
 		if let Operator::Knn(k, d) = exp.operator() {
 			if let Node::Computed(v) = val {
 				let vec: Vec<Number> = v.as_ref().try_into()?;
 				self.knn_expressions.insert(
 					exp.clone(),
-					(*k, id, Arc::new(vec), d.clone().unwrap_or(Distance::Euclidean)),
+					(*k, id.clone(), Arc::new(vec), d.clone().unwrap_or(Distance::Euclidean)),
 				);
 			}
 		}
@@ -367,6 +478,23 @@ pub(super) struct IndexesMap {
 	pub(super) definitions: Vec<DefineIndexStatement>,
 }
 
+#[derive(Clone)]
+struct SchemaCache {
+	indexes: Arc<[DefineIndexStatement]>,
+	fields: Arc<[DefineFieldStatement]>,
+}
+
+impl SchemaCache {
+	async fn new(opt: &Options, table: &Table, tx: &mut kvs::Transaction) -> Result<Self, Error> {
+		let indexes = tx.all_tb_indexes(opt.ns(), opt.db(), table).await?;
+		let fields = tx.all_tb_fields(opt.ns(), opt.db(), table).await?;
+		Ok(Self {
+			indexes,
+			fields,
+		})
+	}
+}
+
 pub(super) type GroupRef = u16;
 
 #[derive(Debug, Clone, Eq, PartialEq, Hash)]
@@ -378,8 +506,9 @@ pub(super) enum Node {
 		right: Arc<Node>,
 		exp: Arc<Expression>,
 	},
-	IndexedField(Arc<Idiom>, Arc<Vec<IndexRef>>),
+	IndexedField(Idiom, Vec<IndexRef>),
-	NonIndexedField(Arc<Idiom>),
+	RecordField(Idiom, RecordOptions),
+	NonIndexedField(Idiom),
 	Computed(Arc<Value>),
 	Unsupported(String),
 }
@@ -393,28 +522,31 @@ impl Node {
 		}
 	}
 
-	pub(super) fn is_indexed_field(&self) -> Option<(Arc<Idiom>, Arc<Vec<IndexRef>>)> {
+	pub(super) fn is_indexed_field(
-		if let Node::IndexedField(id, irs) = self {
+		&self,
-			Some((id.clone(), irs.clone()))
+	) -> Option<(&Idiom, LocalIndexRefs, Option<RemoteIndexRefs>)> {
-		} else {
+		match self {
-			None
+			Node::IndexedField(id, irs) => Some((id, irs.clone(), None)),
+			Node::RecordField(id, ro) => Some((id, ro.locals.clone(), Some(ro.remotes.clone()))),
+			_ => None,
 		}
 	}
 
-	pub(super) fn is_non_indexed_field(&self) -> Option<Arc<Idiom>> {
+	pub(super) fn is_non_indexed_field(&self) -> Option<&Idiom> {
 		if let Node::NonIndexedField(id) = self {
-			Some(id.clone())
+			Some(id)
 		} else {
 			None
 		}
 	}
 }
 
-#[derive(Clone, Copy)]
+#[derive(Clone, Copy, Debug, Eq, PartialEq, Hash)]
-enum IdiomPosition {
+pub(super) enum IdiomPosition {
 	Left,
 	Right,
 }
+
 impl IdiomPosition {
 	// Reverses the operator for non-commutative operators
 	fn transform(&self, op: &Operator) -> Operator {

core/src/sql/expression.rs

@@ -185,7 +185,7 @@ impl Expression {
 			Operator::NoneInside => fnc::operate::inside_none(&l, &r),
 			Operator::Outside => fnc::operate::outside(&l, &r),
 			Operator::Intersects => fnc::operate::intersects(&l, &r),
-			Operator::Matches(_) => fnc::operate::matches(ctx, txn, doc, self).await,
+			Operator::Matches(_) => fnc::operate::matches(ctx, opt, txn, doc, self, l, r).await,
 			Operator::Knn(_, _) => fnc::operate::knn(ctx, opt, txn, doc, self).await,
 			_ => unreachable!(),
 		}

lib/tests/matches.rs

@@ -37,11 +37,11 @@ async fn select_where_matches_using_index() -> Result<(), Error> {
 						},
 						operation: 'Iterate Index'
 					},
-						{
+					{
-							detail: {
+						detail: {
-								type: 'Memory'
+							type: 'Memory'
-							},
+						},
-							operation: 'Collector'
+						operation: 'Collector'
 					}
 			]",
 	);

lib/tests/planner.rs

@@ -1914,3 +1914,367 @@ async fn select_with_in_operator_multiple_indexes() -> Result<(), Error> {
 	assert_eq!(format!("{:#}", tmp), format!("{:#}", val));
 	Ok(())
 }
+
+#[tokio::test]
+async fn select_with_record_id_link_no_index() -> Result<(), Error> {
+	let dbs = new_ds().await?;
+	let ses = Session::owner().with_ns("test").with_db("test");
+	//
+	let sql = "
+		DEFINE FIELD name ON TABLE t TYPE string;
+		DEFINE FIELD t ON TABLE i TYPE record(t);
+		CREATE t:1 SET name = 'h';
+		CREATE t:2 SET name = 'h';
+		CREATE i:A SET t = t:1;
+		CREATE i:B SET t = t:2;
+		SELECT * FROM i WHERE t.name = 'h';
+		SELECT * FROM i WHERE t.name = 'h' EXPLAIN;
+	";
+	let mut res = dbs.execute(&sql, &ses, None).await?;
+	//
+	assert_eq!(res.len(), 8);
+	skip_ok(&mut res, 6)?;
+	//
+	let tmp = res.remove(0).result?;
+	let val = Value::parse(
+		r#"[
+				{ "id": i:A, "t": t:1 },
+				{ "id": i:B, "t": t:2 }
+			]"#,
+	);
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", val));
+	//
+	let tmp = res.remove(0).result?;
+	let val = Value::parse(
+		r#"[
+				{
+					detail: {
+						table: 'i'
+					},
+					operation: 'Iterate Table'
+				},
+				{
+					detail: {
+						reason: 'NO INDEX FOUND'
+					},
+					operation: 'Fallback'
+				},
+				{
+					detail: {
+						type: 'Memory'
+					},
+					operation: 'Collector'
+				}
+			]"#,
+	);
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", val));
+	//
+	Ok(())
+}
+
+#[tokio::test]
+async fn select_with_record_id_link_index() -> Result<(), Error> {
+	let dbs = new_ds().await?;
+	let ses = Session::owner().with_ns("test").with_db("test");
+	//
+	let sql = "
+		DEFINE INDEX i_t_id ON TABLE i COLUMNS t;
+		DEFINE INDEX t_name_idx ON TABLE t COLUMNS name;
+		DEFINE FIELD name ON TABLE t TYPE string;
+		DEFINE FIELD t ON TABLE i TYPE record(t);
+		CREATE t:1 SET name = 'h';
+		CREATE t:2 SET name = 'h';
+		CREATE i:A SET t = t:1;
+		CREATE i:B SET t = t:2;
+		SELECT * FROM i WHERE t.name = 'h' EXPLAIN;
+		SELECT * FROM i WHERE t.name = 'h';
+	";
+	let mut res = dbs.execute(&sql, &ses, None).await?;
+	//
+	assert_eq!(res.len(), 10);
+	skip_ok(&mut res, 8)?;
+	//
+	let expected = Value::parse(
+		r#"[
+				{ "id": i:A, "t": t:1 },
+				{ "id": i:B, "t": t:2 }
+			]"#,
+	);
+	//
+	let tmp = res.remove(0).result?;
+	let val = Value::parse(
+		r#"[
+					{
+						detail: {
+							plan: {
+								index: 'i_t_id',
+								joins: [
+									{
+										index: 't_name_idx',
+										operator: '=',
+										value: 'h'
+									}
+								],
+								operator: 'join'
+							},
+							table: 'i'
+						},
+						operation: 'Iterate Index'
+					},
+					{
+						detail: {
+							type: 'Memory'
+						},
+						operation: 'Collector'
+					}
+				]"#,
+	);
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", val));
+	//
+	let tmp = res.remove(0).result?;
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", expected));
+	//
+	Ok(())
+}
+
+#[tokio::test]
+async fn select_with_record_id_link_unique_index() -> Result<(), Error> {
+	let dbs = new_ds().await?;
+	let ses = Session::owner().with_ns("test").with_db("test");
+	//
+	let sql = "
+		DEFINE INDEX i_t_unique_id ON TABLE i COLUMNS t UNIQUE;
+		DEFINE INDEX t_name_idx ON TABLE t COLUMNS name;
+		DEFINE FIELD name ON TABLE t TYPE string;
+		DEFINE FIELD t ON TABLE i TYPE record(t);
+		CREATE t:1 SET name = 'h';
+		CREATE t:2 SET name = 'h';
+		CREATE i:A SET t = t:1;
+		CREATE i:B SET t = t:2;
+		SELECT * FROM i WHERE t.name = 'h' EXPLAIN;
+		SELECT * FROM i WHERE t.name = 'h';
+	";
+	let mut res = dbs.execute(&sql, &ses, None).await?;
+	//
+	assert_eq!(res.len(), 10);
+	skip_ok(&mut res, 8)?;
+	//
+	let expected = Value::parse(
+		r#"[
+				{ "id": i:A, "t": t:1 },
+				{ "id": i:B, "t": t:2 }
+			]"#,
+	);
+	//
+	let tmp = res.remove(0).result?;
+	let val = Value::parse(
+		r#"[
+					{
+						detail: {
+							plan: {
+								index: 'i_t_unique_id',
+								joins: [
+									{
+										index: 't_name_idx',
+										operator: '=',
+										value: 'h'
+									}
+								],
+								operator: 'join'
+							},
+							table: 'i'
+						},
+						operation: 'Iterate Index'
+					},
+					{
+						detail: {
+							type: 'Memory'
+						},
+						operation: 'Collector'
+					}
+				]"#,
+	);
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", val));
+	//
+	let tmp = res.remove(0).result?;
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", expected));
+	//
+	Ok(())
+}
+#[tokio::test]
+async fn select_with_record_id_link_unique_remote_index() -> Result<(), Error> {
+	let dbs = new_ds().await?;
+	let ses = Session::owner().with_ns("test").with_db("test");
+	//
+	let sql = "
+		DEFINE INDEX i_t_id ON TABLE i COLUMNS t;
+		DEFINE INDEX t_name_unique_idx ON TABLE t COLUMNS name UNIQUE;
+		DEFINE FIELD name ON TABLE t TYPE string;
+		DEFINE FIELD t ON TABLE i TYPE record(t);
+		CREATE t:1 SET name = 'a';
+		CREATE t:2 SET name = 'b';
+		CREATE i:A SET t = t:1;
+		CREATE i:B SET t = t:2;
+		SELECT * FROM i WHERE t.name IN ['a', 'b'] EXPLAIN;
+		SELECT * FROM i WHERE t.name IN ['a', 'b'];
+	";
+	let mut res = dbs.execute(&sql, &ses, None).await?;
+	//
+	assert_eq!(res.len(), 10);
+	skip_ok(&mut res, 8)?;
+	//
+	let expected = Value::parse(
+		r#"[
+				{ "id": i:A, "t": t:1 },
+				{ "id": i:B, "t": t:2 }
+			]"#,
+	);
+	//
+	let tmp = res.remove(0).result?;
+	let val = Value::parse(
+		r#"[
+					{
+						detail: {
+							plan: {
+								index: 'i_t_id',
+								joins: [
+									{
+										index: 't_name_unique_idx',
+										operator: 'union',
+										value: [
+											'a',
+											'b'
+										]
+									}
+								],
+								operator: 'join'
+							},
+							table: 'i'
+						},
+						operation: 'Iterate Index'
+					},
+					{
+						detail: {
+							type: 'Memory'
+						},
+						operation: 'Collector'
+					}
+				]"#,
+	);
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", val));
+	//
+	let tmp = res.remove(0).result?;
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", expected));
+	//
+	Ok(())
+}
+
+#[tokio::test]
+async fn select_with_record_id_link_full_text_index() -> Result<(), Error> {
+	let dbs = new_ds().await?;
+	let ses = Session::owner().with_ns("test").with_db("test");
+	//
+	let sql = "
+		DEFINE ANALYZER name TOKENIZERS class FILTERS lowercase,ngram(1,128);
+		DEFINE INDEX t_name_search_idx ON TABLE t COLUMNS name SEARCH ANALYZER name BM25 HIGHLIGHTS;
+		DEFINE INDEX i_t_id ON TABLE i COLUMNS t;
+		DEFINE FIELD name ON TABLE t TYPE string;
+		DEFINE FIELD t ON TABLE i TYPE record(t);
+		CREATE t:1 SET name = 'Hello World';
+		CREATE i:A SET t = t:1;
+		SELECT * FROM i WHERE t.name @@ 'world' EXPLAIN;
+		SELECT * FROM i WHERE t.name @@ 'world';
+	";
+	let mut res = dbs.execute(&sql, &ses, None).await?;
+
+	assert_eq!(res.len(), 9);
+	skip_ok(&mut res, 7)?;
+	//
+	let tmp = res.remove(0).result?;
+	let val = Value::parse(
+		r#"[
+				{
+					detail: {
+						plan: {
+							index: 'i_t_id',
+							joins: [
+								{
+									index: 't_name_search_idx',
+									operator: '@@',
+									value: 'world'
+								}
+							],
+							operator: 'join'
+						},
+						table: 'i'
+					},
+					operation: 'Iterate Index'
+				},
+				{
+					detail: {
+						type: 'Memory'
+					},
+					operation: 'Collector'
+				}
+			]"#,
+	);
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", val));
+	//
+	let tmp = res.remove(0).result?;
+	let val = Value::parse(r#"[{ "id": i:A, "t": t:1}]"#);
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", val));
+	//
+	Ok(())
+}
+
+#[tokio::test]
+async fn select_with_record_id_link_full_text_no_record_index() -> Result<(), Error> {
+	let dbs = new_ds().await?;
+	let ses = Session::owner().with_ns("test").with_db("test");
+	//
+	let sql = "
+		DEFINE ANALYZER name TOKENIZERS class FILTERS lowercase,ngram(1,128);
+		DEFINE INDEX t_name_search_idx ON TABLE t COLUMNS name SEARCH ANALYZER name BM25 HIGHLIGHTS;
+		DEFINE FIELD name ON TABLE t TYPE string;
+		DEFINE FIELD t ON TABLE i TYPE record(t);
+		CREATE t:1 SET name = 'Hello World';
+		CREATE i:A SET t = t:1;
+		SELECT * FROM i WHERE t.name @@ 'world' EXPLAIN;
+		SELECT * FROM i WHERE t.name @@ 'world';
+	";
+	let mut res = dbs.execute(&sql, &ses, None).await?;
+
+	assert_eq!(res.len(), 8);
+	skip_ok(&mut res, 6)?;
+	//
+	let tmp = res.remove(0).result?;
+	let val = Value::parse(
+		r#"[
+					{
+						detail: {
+							table: 'i'
+						},
+						operation: 'Iterate Table'
+					},
+					{
+						detail: {
+							reason: 'NO INDEX FOUND'
+						},
+						operation: 'Fallback'
+					},
+					{
+						detail: {
+							type: 'Memory'
+						},
+						operation: 'Collector'
+					}
+			]"#,
+	);
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", val));
+	//
+	let tmp = res.remove(0).result?;
+	let val = Value::parse(r#"[{ "id": i:A, "t": t:1}]"#);
+	assert_eq!(format!("{:#}", tmp), format!("{:#}", val));
+	//
+	Ok(())
+}