Optimize cp #906

DvirDukhan · 2020-02-05T13:50:42Z

Split multiple branched cartesian product into chain of dual branched cartesian products

jeffreylovitz · 2020-02-06T16:27:12Z

src/execution_plan/execution_plan_modify.c

-		for(int i = 0; i < op->childCount; i++) {
-			_OpBase_AddChild(parent, op->children[i]);
+		if(op->childCount == 1) {
+			_ExecutionPlan_ParentReplaceChild(op->parent, op, op->children[0]);


👍 Nice subtle optimization. We have to refactor this area in general, but it's nice to cut down on redundant alloc/frees until then!

jeffreylovitz · 2020-02-06T16:28:04Z

src/execution_plan/optimizations/optimize_cartesian_product.h

+ * 1. Might reduce memory consumption (store f records instead n^x) in each phase.
+ * 2. The filter overall runtime by order(s) of magnitude.*/


Minor rewording:

* 1. Potentially reduce memory consumption (storing only f records instead n^x) in each phase. * 2. Reduce the overall filter runtime by potentially order(s) of magnitude.*/

jeffreylovitz · 2020-02-06T16:28:31Z

src/execution_plan/optimizations/optimizer.c

@@ -24,6 +24,9 @@ void optimizePlan(ExecutionPlan *plan) {
 	/* Remove redundant SCAN operations. */
 	reduceScans(plan);

+	/* Try to optimize cartesian product */
+	optimizeCartesianProduct(plan);


It would be more efficient to call applyJoin before optimizeCartesianProduct, as the former may reduce the number of CPs with > 2 children.

I tend to think it would be better to reposition filters and only then try and "upgrade" them to a join operation.

jeffreylovitz · 2020-02-06T16:29:10Z

src/execution_plan/optimizations/optimize_cartesian_product.c

+#include "../ops/op_cartesian_product.h"
+#include "optimizations_util.h"
+
+#define NOT_RESOLVED -1


Can delete, defined in header.

jeffreylovitz · 2020-02-06T16:29:24Z

src/execution_plan/optimizations/optimize_cartesian_product.c

+#define NOT_RESOLVED -1
+
+// Tests to see if given filter can be re-positioned.
+static inline bool _applicableFilter(const FT_FilterNode *f) {


We can actually support all filters in this optimization, so this whole function can be deleted. This check is required for ApplyJoin because of memory inconsistencies when ValueHashJoin was operating on graph entities rather than scalars (an issue that could be resolved either in memory management or by adding ops like NodeHashJoin). It's not a concern here, though, as we don't alter the workings of Filter or CartesianProduct ops.

thanks
removed

jeffreylovitz · 2020-02-06T16:29:40Z

src/execution_plan/optimizations/optimize_cartesian_product.c

+	return filters;
+}
+
+static OpBase *_chain_cp_and_filter(const ExecutionPlan *plan, OpBase *lhs, OpBase *rhs,


Add explanatory comment.

jeffreylovitz · 2020-02-06T16:30:07Z

src/execution_plan/optimizations/optimize_cartesian_product.c

+			for(int j = 0; j < stream_count; j ++) raxFree(stream_entities[j]);
+			// Re-collect cartesian product streams.
+			stream_count = cp->childCount;
+			build_streams_from_op(cp, stream_entities, stream_count);


We only need to rebuild streams if we still have filters to evaluate. Perhaps if (i+1 == filter_count) break?

you merge streams (children) so the current number of streams and their structrure in not alligned with the current state of the execution plan.

Only if the loop is going to execute again, though - if we've just evaluated the last filter, this is unnecessary effort.

src/execution_plan/optimizations/optimize_cartesian_product.c

jeffreylovitz · 2020-02-06T16:30:48Z

src/execution_plan/optimizations/optimize_cartesian_product.c

+														  (OpBase *)filter_op);
+
+		if(cp->childCount == 0) {
+			// The entire Cartesian Product can be replaced with the join op.


// The entire Cartesian Product can be replaced with the new branch.

jeffreylovitz · 2020-02-06T16:31:04Z

src/execution_plan/optimizations/optimize_cartesian_product.c

+	// Try to re-position the additional filters.
+	uint pending_filters_count = array_len(pending_filters);
+	// If there was a modification to the execution plan, there will be less pending filters.
+	if(pending_filters_count < filter_count) {
+		for(uint i = 0; i < pending_filters_count; i++) {
+			OpFilter *additional_filter = pending_filters[i];
+			re_order_filter_op(plan, pending_filters_lookup_root, additional_filter);
+		}
+	}


See comment at 84.

jeffreylovitz · 2020-02-06T16:31:45Z

src/execution_plan/optimizations/optimizations_util.c

+
+#include "optimizations_util.h"
+
+void re_order_filter_op(ExecutionPlan *plan, OpBase *root, OpFilter *filter) {


We should use a naming convention for exposed functions - OptimizeUtils_MigrateFilterOp?

jeffreylovitz · 2020-02-06T16:32:04Z

src/execution_plan/optimizations/optimizations_util.c

+	FT_FilterNode *additional_filter_tree = filter->filterTree;
+
+	rax *references = FilterTree_CollectModified(additional_filter_tree);


I think a one-liner would be a bit more clear:
rax *references = FilterTree_CollectModified(filter->filterTree);

jeffreylovitz · 2020-02-06T16:32:17Z

src/execution_plan/optimizations/optimizations_util.c

+}
+
+// Returns true if the stream resolves all required entities.
+static bool _stream_resolves_entities(rax *stream_resolves, rax *entities_to_resolve) {


I think this function is equivalent to raxIsSubset in utils/rax_extensions.h.

yep. nice!
changed

jeffreylovitz · 2020-02-06T16:33:31Z

src/execution_plan/optimizations/optimizations_util.c

+	return resolved_all;
+}
+
+int relate_exp_to_stream(AR_ExpNode *exp, rax **stream_entities, int stream_count) {


Standardize naming conventions of exposed functions.

jeffreylovitz · 2020-02-06T16:33:38Z

src/execution_plan/optimizations/optimizations_util.c

+	return stream_num;
+}
+
+void build_streams_from_op(OpBase *op, rax **streams, int stream_count) {


Standardize naming conventions of exposed functions.

jeffreylovitz

This looks excellent, a few questions and suggestions in comments.

Additionally, what do you think of using the optimization stage to limit all Cartesian product ops to 2 branches? For cases like:
MATCH (a), (b), (c) RETURN *
We'll still currently have a 3-child CP op. This is functionally equivalent to a nested CP op, but I think the current approach is harder to read on execution plans and makes the op_cartesian_product code significantly different than other stream-joining ops. Thoughts?

jeffreylovitz · 2020-02-07T16:43:09Z

src/execution_plan/optimizations/apply_join.c

+		* When trying to replace a cartesian product which followed by multiple filters that can be resolved by the same two branches
+		* the application crashed on assert(lhs_resolving_stream != rhs_resolving_stream) since before the fix only one filter was used
+		* to create the hash join, and the other were ignored. */
+		// This stream is solved by a single cartesian product child and needs to be propogate up.


Nit: "propagated"

jeffreylovitz · 2020-02-07T16:43:34Z

src/execution_plan/optimizations/apply_join.c

 		if(lhs_resolving_stream == rhs_resolving_stream) {
-			pending_filters = array_append(pending_filters, filter_op);
+			OptimizeUtils_MigrateFilterOp(plan, cp->children[rhs_resolving_stream], filter_op);


👍 I think this is a lot tidier than the pending_filters approach.

jeffreylovitz · 2020-02-07T16:44:00Z

src/execution_plan/optimizations/apply_join.c

@@ -182,38 +129,24 @@ static void _reduce_cp_to_hashjoin(ExecutionPlan *plan, OpBase *cp) {
 			// The entire Cartesian Product can be replaced with the join op.
 			ExecutionPlan_ReplaceOp(plan, cp, value_hash_join);
 			OpBase_Free(cp);
-			pending_filters_lookup_root = value_hash_join;
-			break;
+			// Try to popagate up the remaining filters.


Let's expand this comment a little to emphasize that the optimization has been applied, and we'll now migrate all remaining filters and exit the loop.

jeffreylovitz · 2020-02-07T16:44:33Z

src/execution_plan/optimizations/apply_join.c

+			i++;
+			for(; i < filter_count; i++) {
+				OptimizeUtils_MigrateFilterOp(plan, value_hash_join, filter_ops[i]);
+			}
 		} else {
 			// The Cartesian Product still has a child operation; introduce the join op as another child.
 			ExecutionPlan_AddOp(cp, value_hash_join);
 			// Streams are no longer valid since cartesian product changed.
 			for(int j = 0; j < stream_count; j ++) raxFree(stream_entities[j]);


Can break here and avoid redundant OptimizeUtils_BuildStreamFromOp calls if (i+1 == filter_count).

The raxFree calls should be in the conditional block, as otherwise we'll double free in cleanup.

jeffreylovitz · 2020-02-07T16:45:04Z

src/execution_plan/optimizations/optimizations_util.c

+	ExecutionPlan_RemoveOp(plan, (OpBase *)filter);
+	ExecutionPlan_PushBelow(op, (OpBase *)filter);


Since we will get here fairly often with filters that don't actually need to be migrated, we can save ourselves some unnecessary work with a construction like:

if(op != filter->op.children[0]) { ExecutionPlan_RemoveOp(plan, (OpBase *)filter); ExecutionPlan_PushBelow(op, (OpBase *)filter); }

jeffreylovitz · 2020-02-07T16:46:06Z

src/execution_plan/optimizations/optimize_cartesian_product.c

+		uint rhs_resolving_stream = OptimizeUtils_RelateExpToStream(rhs, stream_entities, stream_count);
+		if(rhs_resolving_stream == NOT_RESOLVED) continue;
+
+		// This stream is solved by a single cartesian product child and needs to be propogate up.


Nit: "propagated"

jeffreylovitz · 2020-02-07T16:46:22Z

src/execution_plan/optimizations/optimize_cartesian_product.c

+			continue;
+		}
+
+		// Retrieve the relevent branch roots.


Nit: "relevant"

jeffreylovitz · 2020-02-07T16:46:41Z

src/execution_plan/optimizations/optimize_cartesian_product.c

+			// The entire Cartesian Product can be replaced with the new branch.
+			ExecutionPlan_ReplaceOp(plan, cp, filtered_cp_branch);
+			OpBase_Free(cp);
+			// Try to popagate up the remaining filters.


Nit: "propagate"

jeffreylovitz · 2020-02-07T16:47:05Z

src/execution_plan/optimizations/optimize_cartesian_product.c

+		} else {
+			// The Cartesian Product still has a child operation; introduce the new op as another child.
+			ExecutionPlan_AddOp(cp, filtered_cp_branch);
+			// Streams are no longer valid since cartesian product changed.


Can break here and avoid redundant OptimizeUtils_BuildStreamFromOp calls if (i+1 == filter_count).

jeffreylovitz · 2020-02-07T16:48:13Z

src/execution_plan/optimizations/optimize_cartesian_product.h

+ * trying to locate situations where a cartesian product of two disjoint streams
+ * can be filtered by the filter.


* locating situations where a new Cartesian Product of just two streams can resolve the filter.

jeffreylovitz · 2020-02-07T16:48:43Z

tests/flow/test_optimizations_plan.py

+                    ['Boaz'],
+                    ['Roi']]
+        resultset = graph.query(query).result_set
+        self.env.assertEqual(resultset, expected)


Let's add another test that triggers the (lhs_resolving_stream == rhs_resolving_stream) condition, such as:
MATCH (p1), (p2), (p3) WHERE p1.name <> p2.name AND ID(p1) <> ID(p2) RETURN DISTINCT p2.name ORDER BY p2.name

* Support float inputs for modulo computations (#895) * Add references * Update mkdocs.yml * Update References.md * fixed ast mapping for path filter (#896) * fixed ast mapping for path filter * renamed tests. open a new redis graph client per tests Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * added mutex per matrix (#898) * added mutex per matrix * fixed PR comments * updated license headers to 2020 (#902) * Preserve order of op's children array when introducing index scans (#912) * Fix memory leaks (#917) * Fix memory leaks * Use original logic for MarkWriter * added filter tree clone (#915) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Optimize cp (#906) * added optimize cp * fixed a bug. added test * add optimization_util files * fixed PR comments * fixed PR comments * added multiple branch cp optimization * fixed PR comment * added in place replacement at ExecutionPlan_RemoveOp * added test for cp optimization and semi apply * fixed PR comments * Added non existing entity runtime error (#919) * Added non existing entity runtime error * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * RediSearch query error reporting (#925) * added filter tree compaction (#922) * added filter tree compaction * fixed PR comments * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Handle both possible execution orders in concurrent rename test (#926) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Refferencing a RedisGraph javascript client library (#928) * Improve client libraries sections on both readme and clients.md Clients libraries are subject to repetitive edition, it seems better to have a concise formatting * refferencing my own contribution to clients libraries Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Rust client (#931) * Update README.md * Update clients.md * intoduced fpClone in OpBase (#930) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Apply deplete match stream (#913) * Deplete Apply op's match stream for every left-hand Record * Disallow ExpandInto ops on variables beneath Apply ops * Add tests * PR fixes * Argument op holds one Record * PR fixes * Improve logic for building OpArgument modifies arrays * standardize logic for building tmp ExecutionPlans * Remove sub-ExecutionPlan logic * Fix CondTraverseReset routine * Fix ExpandIntoReset * Fix variable-length QueryGraph pointer * Always reset match branch * Add explanatory comment * PR fixes * PR fixes * raxValues returns void pointer array Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * increased allowed parameters count to UNIT_MAX. fixed a bug in query_ctx (#933) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * introduce mapping between record entries and projected columns (#936) * added arr clone with cb (#937) * added arr clone with cb * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Leak fixes (#935) * Close Redis key handles * Fix memory leak in checking whether procedures are read-only * Improve explanatory comment Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * label matrix should be fetch right before eval (#938) * More sensible function signature for SIValue_Free (#941) * Resolve memory leaks on Path SIValues (#940) * Resolve memory leaks on Path SIValues * Update op_delete.c * Update op_cond_var_len_traverse.c Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * GraphBLAS 3.2.0 (#942) * GraphBLAS 3.2.0 * updated makefiles * Remove unnecessary GraphBLAS build flag Co-authored-by: Jeffrey Lovitz <jeffrey.lovitz@gmail.com> * Simplify OpAggregate logic, remove unnecessary struct members (#947) * Simplify OpAggregate logic, remove unnecessary struct members * Update op_aggregate.c * Update op_aggregate.c * Update op_aggregate.c * Update op_aggregate.c Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Add compile-time error for unsupported AST node types (#944) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * version bump 2.0.2 Co-authored-by: Jeffrey Lovitz <jeffrey.lovitz@gmail.com> Co-authored-by: Guy Korland <gkorland@gmail.com> Co-authored-by: DvirDukhan <dvir@redislabs.com> Co-authored-by: Sceat <11330271+Sceat@users.noreply.github.com>

* Support float inputs for modulo computations (#895) * Add references * Update mkdocs.yml * Update References.md * fixed ast mapping for path filter (#896) * fixed ast mapping for path filter * renamed tests. open a new redis graph client per tests Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * added mutex per matrix (#898) * added mutex per matrix * fixed PR comments * updated license headers to 2020 (#902) * Preserve order of op's children array when introducing index scans (#912) * Fix memory leaks (#917) * Fix memory leaks * Use original logic for MarkWriter * added filter tree clone (#915) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Optimize cp (#906) * added optimize cp * fixed a bug. added test * add optimization_util files * fixed PR comments * fixed PR comments * added multiple branch cp optimization * fixed PR comment * added in place replacement at ExecutionPlan_RemoveOp * added test for cp optimization and semi apply * fixed PR comments * Added non existing entity runtime error (#919) * Added non existing entity runtime error * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * RediSearch query error reporting (#925) * added filter tree compaction (#922) * added filter tree compaction * fixed PR comments * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Handle both possible execution orders in concurrent rename test (#926) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Refferencing a RedisGraph javascript client library (#928) * Improve client libraries sections on both readme and clients.md Clients libraries are subject to repetitive edition, it seems better to have a concise formatting * refferencing my own contribution to clients libraries Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Rust client (#931) * Update README.md * Update clients.md * intoduced fpClone in OpBase (#930) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Apply deplete match stream (#913) * Deplete Apply op's match stream for every left-hand Record * Disallow ExpandInto ops on variables beneath Apply ops * Add tests * PR fixes * Argument op holds one Record * PR fixes * Improve logic for building OpArgument modifies arrays * standardize logic for building tmp ExecutionPlans * Remove sub-ExecutionPlan logic * Fix CondTraverseReset routine * Fix ExpandIntoReset * Fix variable-length QueryGraph pointer * Always reset match branch * Add explanatory comment * PR fixes * PR fixes * raxValues returns void pointer array Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * increased allowed parameters count to UNIT_MAX. fixed a bug in query_ctx (#933) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * introduce mapping between record entries and projected columns (#936) * added arr clone with cb (#937) * added arr clone with cb * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Leak fixes (#935) * Close Redis key handles * Fix memory leak in checking whether procedures are read-only * Improve explanatory comment Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * label matrix should be fetch right before eval (#938) * More sensible function signature for SIValue_Free (#941) * Resolve memory leaks on Path SIValues (#940) * Resolve memory leaks on Path SIValues * Update op_delete.c * Update op_cond_var_len_traverse.c Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * GraphBLAS 3.2.0 (#942) * GraphBLAS 3.2.0 * updated makefiles * Remove unnecessary GraphBLAS build flag Co-authored-by: Jeffrey Lovitz <jeffrey.lovitz@gmail.com> * Simplify OpAggregate logic, remove unnecessary struct members (#947) * Simplify OpAggregate logic, remove unnecessary struct members * Update op_aggregate.c * Update op_aggregate.c * Update op_aggregate.c * Update op_aggregate.c Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Add compile-time error for unsupported AST node types (#944) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * removed unused operations and types (#950) * version bump 2.0.3 Co-authored-by: Jeffrey Lovitz <jeffrey.lovitz@gmail.com> Co-authored-by: Guy Korland <gkorland@gmail.com> Co-authored-by: DvirDukhan <dvir@redislabs.com> Co-authored-by: Sceat <11330271+Sceat@users.noreply.github.com>

* Support float inputs for modulo computations (#895) * Add references * Update mkdocs.yml * Update References.md * fixed ast mapping for path filter (#896) * fixed ast mapping for path filter * renamed tests. open a new redis graph client per tests Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * added mutex per matrix (#898) * added mutex per matrix * fixed PR comments * updated license headers to 2020 (#902) * Preserve order of op's children array when introducing index scans (#912) * Fix memory leaks (#917) * Fix memory leaks * Use original logic for MarkWriter * added filter tree clone (#915) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Optimize cp (#906) * added optimize cp * fixed a bug. added test * add optimization_util files * fixed PR comments * fixed PR comments * added multiple branch cp optimization * fixed PR comment * added in place replacement at ExecutionPlan_RemoveOp * added test for cp optimization and semi apply * fixed PR comments * Added non existing entity runtime error (#919) * Added non existing entity runtime error * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * RediSearch query error reporting (#925) * added filter tree compaction (#922) * added filter tree compaction * fixed PR comments * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Handle both possible execution orders in concurrent rename test (#926) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Refferencing a RedisGraph javascript client library (#928) * Improve client libraries sections on both readme and clients.md Clients libraries are subject to repetitive edition, it seems better to have a concise formatting * refferencing my own contribution to clients libraries Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Rust client (#931) * Update README.md * Update clients.md * intoduced fpClone in OpBase (#930) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Apply deplete match stream (#913) * Deplete Apply op's match stream for every left-hand Record * Disallow ExpandInto ops on variables beneath Apply ops * Add tests * PR fixes * Argument op holds one Record * PR fixes * Improve logic for building OpArgument modifies arrays * standardize logic for building tmp ExecutionPlans * Remove sub-ExecutionPlan logic * Fix CondTraverseReset routine * Fix ExpandIntoReset * Fix variable-length QueryGraph pointer * Always reset match branch * Add explanatory comment * PR fixes * PR fixes * raxValues returns void pointer array Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * increased allowed parameters count to UNIT_MAX. fixed a bug in query_ctx (#933) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * introduce mapping between record entries and projected columns (#936) * added arr clone with cb (#937) * added arr clone with cb * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Leak fixes (#935) * Close Redis key handles * Fix memory leak in checking whether procedures are read-only * Improve explanatory comment Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * label matrix should be fetch right before eval (#938) * More sensible function signature for SIValue_Free (#941) * Resolve memory leaks on Path SIValues (#940) * Resolve memory leaks on Path SIValues * Update op_delete.c * Update op_cond_var_len_traverse.c Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * GraphBLAS 3.2.0 (#942) * GraphBLAS 3.2.0 * updated makefiles * Remove unnecessary GraphBLAS build flag Co-authored-by: Jeffrey Lovitz <jeffrey.lovitz@gmail.com> * Simplify OpAggregate logic, remove unnecessary struct members (#947) * Simplify OpAggregate logic, remove unnecessary struct members * Update op_aggregate.c * Update op_aggregate.c * Update op_aggregate.c * Update op_aggregate.c Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Add compile-time error for unsupported AST node types (#944) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * removed unused operations and types (#950) * remove GraphBLAS cacheing from CI (#953) * version bump Co-authored-by: Jeffrey Lovitz <jeffrey.lovitz@gmail.com> Co-authored-by: Guy Korland <gkorland@gmail.com> Co-authored-by: DvirDukhan <dvir@redislabs.com> Co-authored-by: Sceat <11330271+Sceat@users.noreply.github.com>

* Support float inputs for modulo computations (#895) * Add references * Update mkdocs.yml * Update References.md * fixed ast mapping for path filter (#896) * fixed ast mapping for path filter * renamed tests. open a new redis graph client per tests Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * added mutex per matrix (#898) * added mutex per matrix * fixed PR comments * updated license headers to 2020 (#902) * Preserve order of op's children array when introducing index scans (#912) * Fix memory leaks (#917) * Fix memory leaks * Use original logic for MarkWriter * added filter tree clone (#915) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Optimize cp (#906) * added optimize cp * fixed a bug. added test * add optimization_util files * fixed PR comments * fixed PR comments * added multiple branch cp optimization * fixed PR comment * added in place replacement at ExecutionPlan_RemoveOp * added test for cp optimization and semi apply * fixed PR comments * Added non existing entity runtime error (#919) * Added non existing entity runtime error * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * RediSearch query error reporting (#925) * added filter tree compaction (#922) * added filter tree compaction * fixed PR comments * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Handle both possible execution orders in concurrent rename test (#926) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Refferencing a RedisGraph javascript client library (#928) * Improve client libraries sections on both readme and clients.md Clients libraries are subject to repetitive edition, it seems better to have a concise formatting * refferencing my own contribution to clients libraries Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Rust client (#931) * Update README.md * Update clients.md * intoduced fpClone in OpBase (#930) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Apply deplete match stream (#913) * Deplete Apply op's match stream for every left-hand Record * Disallow ExpandInto ops on variables beneath Apply ops * Add tests * PR fixes * Argument op holds one Record * PR fixes * Improve logic for building OpArgument modifies arrays * standardize logic for building tmp ExecutionPlans * Remove sub-ExecutionPlan logic * Fix CondTraverseReset routine * Fix ExpandIntoReset * Fix variable-length QueryGraph pointer * Always reset match branch * Add explanatory comment * PR fixes * PR fixes * raxValues returns void pointer array Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * increased allowed parameters count to UNIT_MAX. fixed a bug in query_ctx (#933) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * introduce mapping between record entries and projected columns (#936) * added arr clone with cb (#937) * added arr clone with cb * fixed PR comments Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Leak fixes (#935) * Close Redis key handles * Fix memory leak in checking whether procedures are read-only * Improve explanatory comment Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * label matrix should be fetch right before eval (#938) * More sensible function signature for SIValue_Free (#941) * Resolve memory leaks on Path SIValues (#940) * Resolve memory leaks on Path SIValues * Update op_delete.c * Update op_cond_var_len_traverse.c Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * GraphBLAS 3.2.0 (#942) * GraphBLAS 3.2.0 * updated makefiles * Remove unnecessary GraphBLAS build flag Co-authored-by: Jeffrey Lovitz <jeffrey.lovitz@gmail.com> * Simplify OpAggregate logic, remove unnecessary struct members (#947) * Simplify OpAggregate logic, remove unnecessary struct members * Update op_aggregate.c * Update op_aggregate.c * Update op_aggregate.c * Update op_aggregate.c Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * Add compile-time error for unsupported AST node types (#944) Co-authored-by: Roi Lipman <swilly22@users.noreply.github.com> * removed unused operations and types (#950) * remove GraphBLAS cacheing from CI (#953) * switch from OR AND semiring to ANY PAIR (#955) * version bump Co-authored-by: Jeffrey Lovitz <jeffrey.lovitz@gmail.com> Co-authored-by: Guy Korland <gkorland@gmail.com> Co-authored-by: DvirDukhan <dvir@redislabs.com> Co-authored-by: Sceat <11330271+Sceat@users.noreply.github.com>