The AQL query optimizer

AQL queries are sent through an optimizer before execution that creates an initial execution plan, looks for optimization opportunities, and applies them

AQL queries are parsed and planned. The optimizer might produce multiple execution plans for a single query. It then calculates the costs for all plans and picks the plan with the lowest total cost. This resulting plan is considered to be the optimal plan, which is then executed.

The optimizer is designed to only perform optimizations if they are safe, in the sense that an optimization should not modify the result of a query. A notable exception to this is that the optimizer is allowed to change the order of results for queries that do not explicitly specify how results should be sorted.

Execution plans

The explain command can be used to query the optimal executed plan or even all plans the optimizer has generated. Additionally, explain can reveal some more information about the optimizer’s view of the query.

Inspecting plans using the explain helper

The explain method of ArangoStatement as shown in the next chapters creates very verbose output. You can work on the output programmatically, or use this handsome tool that we created to generate a more human readable representation.

You may use it like this: (we disable syntax highlighting here)

var coll = db._create("test");
for (i = 0; i < 100; ++i) { db.test.save({ value: i }); }
var idx = db.test.ensureIndex({ type: "persistent", fields: [ "value" ] });
var explain = require("@arangodb/aql/explainer").explain;
explain("FOR i IN test FILTER i.value > 97 SORT i.value RETURN i.value", {colors:false});
Show output

Execution plans in detail

Let’s have a look at the raw json output of the same execution plan using the explain method of ArangoStatement:

var stmt = db._createStatement("FOR i IN test FILTER i.value > 97 SORT i.value RETURN i.value");
stmt.explain();
Show output

As you can see, the result details are very verbose. They are not covered in detail for brevity in the next sections. Instead, let’s take a closer look at the results step by step.

Execution nodes of a query

In general, an execution plan can be considered to be a pipeline of processing steps. Each processing step is carried out by a so-called execution node

The nodes attribute of the explain result contains these execution nodes in the execution plan. The output is still very verbose, so here’s a shorted form of it:

stmt.explain().plan.nodes.map(function (node) { return node.type; });
Show output

Note that the list of nodes might slightly change in future versions of ArangoDB if new execution node types get added or the optimizer create somewhat more optimized plans.

When a plan is executed, the query execution engine starts with the node at the bottom of the list (i.e. the ReturnNode).

The ReturnNode’s purpose is to return data to the caller. It does not produce data itself, but it asks the node above itself, which is the CalculationNode in our example. CalculationNodes are responsible for evaluating arbitrary expressions. In our example query, the CalculationNode evaluates the value of i.value, which is needed by the ReturnNode. The calculation is applied for all data the CalculationNode gets from the node above it, in our example the IndexNode.

Finally, all of this needs to be done for documents of collection test. This is where the IndexNode enters the game. It uses an index (thus its name) to find certain documents in the collection and ships it down the pipeline in the order required by SORT i.value. The IndexNode itself has a SingletonNode as its input. The sole purpose of a SingletonNode node is to provide a single empty document as input for other processing steps. It is always the end of the pipeline.

Here is a summary:

  • SingletonNode: produces an empty document as input for other processing steps.
  • IndexNode: iterates over the index on attribute value in collection test in the order required by SORT i.value.
  • CalculationNode: evaluates the result of the calculation i.value > 97 to true or false
  • CalculationNode: calculates return value i.value
  • ReturnNode: returns data to the caller

Optimizer rules used for a query

Note that in the example, the optimizer has optimized the SORT statement away. It can do it safely because there is a sorted persistent index on i.value, which it has picked in the IndexNode. As the index values are iterated over in sorted order anyway, the extra SortNode would have been redundant and was removed.

Additionally, the optimizer has done more work to generate an execution plan that avoids as much expensive operations as possible. Here is the list of optimizer rules that were applied to the plan:

stmt.explain().plan.rules;
Show output

Here is the meaning of these rules in context of this query:

  • move-calculations-up: Moves a CalculationNode and subqueries, when independent from the outer node, as far up in the processing pipeline as possible.
  • move-filters-up: Moves a FilterNode as far up in the processing pipeline as possible.
  • remove-redundant-calculations: Replaces references to variables with references to other variables that contain the exact same result. In the example query, i.value is calculated multiple times, but each calculation inside a loop iteration would produce the same value. Therefore, the expression result is shared by several nodes.
  • remove-unnecessary-calculations: Removes CalculationNodes whose result values are not used in the query. In the example this happens due to the remove-redundant-calculations rule having made some calculations unnecessary.
  • use-indexes: Use an index to iterate over a collection instead of performing a full collection scan. In the example case this makes sense, as the index can be used for filtering and sorting.
  • remove-filter-covered-by-index: Remove an unnecessary filter whose functionality is already covered by an index. In this case the index only returns documents matching the filter.
  • use-index-for-sort: Removes a SORT operation if it is already satisfied by traversing over a sorted index.

Note that some rules may appear multiple times in the list, with number suffixes. This is due to the same rule being applied multiple times, at different positions in the optimizer pipeline.

Also see the full list of optimizer rules below.

Collections used in a query

The list of collections used in a plan (and query) is contained in the collections attribute of a plan:

stmt.explain().plan.collections
Show output

The name attribute contains the name of the collection, and type is the access type, which can be either read or write.

Variables used in a query

The optimizer returns a list of variables used in a plan (and query). This list contains auxiliary variables created by the optimizer itself. You can ignore this list in most cases.

Cost of a query

For each plan the optimizer generates, it calculates the total cost. The plan with the lowest total cost is considered to be the optimal plan. Costs are estimates only, as the actual execution costs are unknown to the optimizer. Costs are calculated based on heuristics that are hard-coded into execution nodes. Cost values do not have any unit.

Retrieving all execution plans

To retrieve not just the optimal plan but a list of all plans the optimizer has generated, set the option allPlans to true:

This returns a list of all plans in the plans attribute instead of in the plan attribute:

stmt.explain({ allPlans: true });
Show output

Retrieving the plan as it was generated by the parser / lexer

To retrieve the plan which closely matches your query, you may turn off most optimization rules (i.e. cluster rules cannot be disabled if you’re running the explain on a cluster Coordinator) set the option rules to -all:

This returns an unoptimized plan in the plan:

stmt.explain({ optimizer: { rules: [ "-all" ] } });
Show output

Note that some optimizations are already done at parse time (i.e. evaluate simple constant calculation as 1 + 1)

Turning specific optimizer rules off

Optimizer rules can also be turned on or off individually, using the rules attribute. This can be used to enable or disable one or multiple rules. Rules that shall be enabled need to be prefixed with a +, rules to be disabled should be prefixed with a -. The pseudo-rule all matches all rules.

Rules specified in rules are evaluated from left to right, so the following works to turn on just the one specific rule:

stmt.explain({ optimizer: { rules: [ "-all", "+use-index-range" ] } });
Show output

By default, all rules are turned on. To turn off just a few specific rules, use something like this:

stmt.explain({ optimizer: { rules: [ "-use-index-range", "-use-index-for-sort" ] } });
Show output

The maximum number of plans created by the optimizer can also be limited using the maxNumberOfPlans attribute:

stmt.explain({ maxNumberOfPlans: 1 });
Show output

Optimizer statistics

The optimizer provides statistics as a part of an explain result. The following attributes are returned in the stats attribute:

  • plansCreated: The total number of plans created by the optimizer.
  • rulesExecuted: The number of rules executed. Note that an executed rule does not indicate that a plan has actually been modified by a rule.
  • rulesSkipped: The number of rules skipped by the optimizer.
  • executionTime: The (wall-clock) time in seconds needed to explain the query.
  • peakMemoryUsage: The maximum memory usage of the query during explain.

Warnings

For some queries, the optimizer may produce warnings. These are returned in the warnings attribute of the explain result:

var stmt = db._createStatement("FOR i IN 1..10 RETURN 1 / 0")
stmt.explain().warnings;
Show output

There is an upper bound on the number of warnings a query may produce. If that bound is reached, no further warnings are returned.

Optimization in a cluster

When you are running AQL in the cluster, the parsing of the query is done on the Coordinator. The Coordinator then chops the query into snippets, which are either to remain on the Coordinator or need to be distributed to the shards on the DB-Servers over the network. The cutting sites are interconnected via ScatterNodes, GatherNodes and RemoteNodes. These nodes mark the network borders of the snippets.

The optimizer strives to reduce the amount of data transferred via these network interfaces by pushing FILTERs out to the shards, as it is vital to the query performance to reduce that data amount to transfer over the network links.

Some hops between Coordinators and DB-Servers are unavoidable. An example are user-defined functions (UDFs), which have to be executed on the Coordinator. If you cannot modify your query to have a lower amount of back and forth between sites, then try to lower the amount of data that has to be transferred between them. In case of UDFs, use effective FILTERs before calling them.

Using a cluster, there is a Site column if you explain a query. Snippets marked with DBS are executed on DB-Servers, COOR ones are executed on the respective Coordinator.

Query String (57 chars, cacheable: false):
 FOR doc IN test UPDATE doc WITH { updated: true } IN test

Execution plan:
 Id   NodeType          Site     Est.   Comment
  1   SingletonNode     DBS         1   * ROOT
  3   CalculationNode   DBS         1     - LET #3 = { "updated" : true }   
 13   IndexNode         DBS   1000000     - FOR doc IN test   /* primary index scan, index only, projections: `_key`, 5 shard(s) */    
  4   UpdateNode        DBS         0       - UPDATE doc WITH #3 IN test 
  7   RemoteNode        COOR        0       - REMOTE
  8   GatherNode        COOR        0       - GATHER 

Execution nodes

List of execution nodes

The following execution node types appear in the output of explain:

  • CalculationNode: Evaluates an expression. The expression result may be used by other nodes, e.g. FilterNode, EnumerateListNode, SortNode etc.

  • CollectNode: Aggregates its input and produces new output variables. This appears once per COLLECT statement.

  • EnumerateCollectionNode: Enumeration over documents of a collection (given in its collection attribute) without using an index.

  • EnumerateListNode: Enumeration over a list of (non-collection) values.

  • EnumerateViewNode: Enumeration over documents of a View.

  • FilterNode: Only lets values pass that satisfy a filter condition. Appears once per FILTER statement.

  • IndexNode: Enumeration over one or many indexes (given in its indexes attribute) of a collection. The index ranges are specified in the condition attribute of the node.

  • InsertNode: Inserts documents into a collection (given in its collection attribute). Appears exactly once in a query that contains an INSERT statement.

  • JoinNode: Replaces EnumerateCollectionNode and IndexNode for eligible join queries to perform optimized joins.

  • KShortestPathsNode: Indicates a traversal for k Shortest Paths (K_SHORTEST_PATHS in AQL).

  • KPathsNode: Indicates a traversal for k Paths (K_PATHS in AQL).

  • LimitNode: Limits the number of results passed to other processing steps. Appears once per LIMIT statement.

  • MaterializeNode: The presence of this node means that the query is not fully covered by indexes and therefore needs to involve the storage engine.

  • RemoveNode: Removes documents from a collection (given in its collection attribute). Appears exactly once in a query that contains a REMOVE statement.

  • ReplaceNode: Replaces documents in a collection (given in its collection attribute). Appears exactly once in a query that contains a REPLACE statement.

  • ReturnNode: Returns data to the caller. Appears in each read-only query at least once. Subqueries also contain ReturnNodes.

  • SingletonNode: The purpose of a SingletonNode is to produce an empty document that is used as input for other processing steps. Each execution plan contains exactly one SingletonNode as its top node.

  • ShortestPathNode: Indicates a traversal for a Shortest Path (SHORTEST_PATH in AQL).

  • SortNode: Performs a sort of its input values.

  • SubqueryEndNode: End of a spliced (inlined) subquery.

  • SubqueryNode: Executes a subquery.

  • SubqueryStartNode: Beginning of a spliced (inlined) subquery.

  • TraversalNode: Indicates a regular graph traversal, as opposed to a shortest path(s) traversal.

  • UpdateNode: Updates documents in a collection (given in its collection attribute). Appears exactly once in a query that contains an UPDATE statement.

  • UpsertNode: Upserts documents in a collection (given in its collection attribute). Appears exactly once in a query that contains an UPSERT statement.

List of cluster execution nodes

For queries in the cluster, the following additional nodes may appear in execution plans:

  • DistributeNode: Used on a Coordinator to fan-out data to one or multiple shards, taking into account a collection’s shard key.

  • GatherNode: Used on a Coordinator to aggregate results from one or many shards into a combined stream of results. Parallelizes work for certain types of queries when there are multiple DB-Servers involved (shown as GATHER /* parallel */ in query explain).

  • RemoteNode: A RemoteNode performs communication with another ArangoDB instances in the cluster. For example, the cluster Coordinator needs to communicate with other servers to fetch the actual data from the shards. It does so via RemoteNodes. The data servers themselves might again pull further data from the Coordinator, and thus might also employ RemoteNodes. So, all of the above cluster relevant nodes are accompanied by a RemoteNode.

  • ScatterNode: Used on a Coordinator to fan-out data to one or multiple shards.

  • SingleRemoteOperationNode: Used on a Coordinator to directly work with a single document on a DB-Server that is referenced by its _key.

  • MultipleRemoteExecutionNode: Used to optimize bulk INSERT operations in cluster deployments, reducing the setup and shutdown overhead and the number of internal network requests.

Optimizer rules

List of optimizer rules

The following user-facing optimizer rules exist and are enabled by default unless noted otherwise. You can enable and disable optimizer rules except for a few required rules.

Some rules exist multiple times with number suffixes like -2, (e.g. remove-unnecessary-calculations-2). This is due to the same rule being applied multiple times at different optimization stages.

replace-function-with-index

This rule cannot be turned off.

Replace deprecated index functions such as FULLTEXT(), NEAR(), WITHIN(), or WITHIN_RECTANGLE() with a regular subquery.


inline-subqueries

Try to pull subqueries out into their surrounding scope, e.g. FOR x IN (FOR y IN collection FILTER y.value &gt;= 5 RETURN y.test) RETURN x.a becomes FOR tmp IN collection FILTER tmp.value &gt;= 5 LET x = tmp.test RETURN x.a.


replace-like-with-range

This rule cannot be turned off.

Replace LIKE() function with range scans where possible.


simplify-conditions

Replace parts in CalculationNode expressions with simpler expressions.


move-calculations-up

Move calculations up in the processing pipeline as far as possible (ideally out of enumerations) so they are not executed in loops if not required. It is quite common that this rule enables further optimizations.


move-filters-up

Move filters up in the processing pipeline as far as possible (ideally out of inner loops) so they filter results as early as possible.


remove-redundant-calculations

Replace references to redundant calculations (expressions with the exact same result) with a single reference, allowing other rules to remove no longer needed calculations.


remove-unnecessary-filters

Remove FILTER conditions that always evaluate to true.


remove-unnecessary-calculations

Remove all calculations whose result is not referenced in the query. This can be a consequence of applying other optimizations.


remove-redundant-sorts

Try to merge multiple SORT statements into fewer sorts.


optimize-subqueries

Apply optimizations to subqueries.

This rule adds a LIMIT statement to qualifying subqueries to make them return less data. It also modifies the result value of subqueries in case only the number of subquery results is checked later. This saves copying the document data from the subquery to the outer scope and may enable follow-up optimizations.


interchange-adjacent-enumerations

Try out permutations of FOR statements in queries that contain multiple loops, which may enable further optimizations by other rules.


move-calculations-up-2

Second pass of moving calculations up in the processing pipeline as far as possible, to pull them out of inner loops etc.


move-filters-up-2

Second pass of moving filters up in the processing pipeline as far as possible so they filter results as early as possible.


replace-equal-attribute-accesses

This rule is disabled by default.

Replace attribute accesses that are equal due to a filter statement with the same value. This might enable other optimizations later on.


remove-redundant-sorts-2

Second pass of trying to merge multiple SORT statements into fewer sorts.


remove-sort-rand-limit-1

Remove SORT RAND() LIMIT 1 constructs by moving the random iteration into EnumerateCollectionNode.

The RocksDB storage engine doesn't allow to seek random documents efficiently. This optimization picks a pseudo-random document based on a limited number of seeks within the collection's key range, selecting a random start key in the key range, and then going a few steps before or after that.


remove-collect-variables

Remove INTO and AGGREGATE clauses from COLLECT statements if the result is not used.


propagate-constant-attributes

Insert constant values into FILTER conditions, replacing dynamic attribute values.


remove-data-modification-out-variables

Avoid setting the pseudo-variables OLD and NEW if they are not used in data modification queries.


replace-or-with-in

Combine multiple OR equality conditions on the same variable or attribute with an IN condition.


remove-redundant-or

Combine multiple OR conditions for the same variable or attribute into a single condition.


geo-index-optimizer

Utilize geo-spatial indexes.


use-indexes

Use indexes to iterate over collections, replacing EnumerateCollectionNode with IndexNode in the query plan.


remove-filter-covered-by-index

Replace or remove FilterNode if the filter conditions are already covered by IndexNode.


remove-unnecessary-filters-2

Second pass of removing FILTER conditions that always evaluate to true.


use-index-for-sort

Use indexes to avoid SORT operations, removing SortNode from the query plan.


sort-in-values

Use a binary search for in-list lookups with a logarithmic complexity instead of the default linear complexity in-list lookup if the comparison array on the right-hand side of an IN operator is pre-sorted by an extra function call.


optimize-traversal-last-element-access

Transform accesses to the last vertex or edge of the path output variable (p.vertices[-1] and p.edges[-1]) emitted by AQL traversals (FOR v, e, p IN ...) with accesses to the vertex or edge variable (v and e). This can avoid computing the path variable at all and enable further optimizations that are not possible on the path variable p.


optimize-traversals

Try to move FILTER conditions into TraversalNode for early pruning of results, apply traversal projections, and avoid calculating edge and path output variables that are not declared in the query for the AQL traversal.


optimize-enumerate-path-filters

Move FILTER conditions on the path output variable into the path search


optimize-paths

Check how the output variables of K_PATHS, K_SHORTEST_PATHS, and ALL_SHORTEST_PATHS path search graph algorithms are used and avoid loading the vertex documents if they are not accessed in the query.


remove-filter-covered-by-traversal

Replace or remove FilterNode if the filter conditions are already covered by TraversalNode.


handle-arangosearch-views

This rule cannot be turned off.

Appears whenever an arangosearch or search-alias View is accessed in a query.


arangosearch-constrained-sort

Make nodes of type EnumerateViewNode aware of SORT with a subsequent LIMIT when using Views to reduce memory usage and avoid unnecessary sorting that has already been carried out by ArangoSearch internally.


remove-unnecessary-calculations-2

Second pass of removing all calculations whose result is not referenced in the query. This can be a consequence of applying other optimizations


remove-redundant-path-var

Avoid computing the variables emitted by AQL traversals if they are declared but unused in the query, or only used in filters that are pulled into the traversal, significantly reducing overhead.


optimize-cluster-single-document-operations

Only available in cluster deployments.

Let a Coordinator work with a document directly if you reference a document by its _key. In this case, no AQL is executed on the DB-Servers.


optimize-cluster-multiple-document-operations

Only available in cluster deployments.

For bulk INSERT operations in cluster deployments, avoid unnecessary overhead that AQL queries typically require for the setup and shutdown in clusters, as well as for the internal batching.

This optimization also decreases the number of HTTP requests to the DB-Servers.

The following patterns are recognized:

  • FOR doc IN @docs INSERT doc INTO collection, where @docs is a bind parameter with an array of documents to be inserted
  • FOR doc IN [ { … }, { … }, … ] INSERT doc INTO collection, where the FOR loop iterates over an array of input documents known at query compile time
  • LET docs = [ { … }, { … }, … ] FOR doc IN docs INSERT doc INTO collection, where the docs variable is a static array of input documents known at query compile time

If a query has such a pattern, and all of the following restrictions are met, then the optimization is triggered:

  • There are no following RETURN nodes (including any RETURN OLD or RETURN NEW)
  • The FOR loop is not contained in another outer FOR loop or subquery
  • There are no other operations (e.g. LET, FILTER) between FOR and INSERT
  • INSERT is not used on a SmartGraph edge collection
  • The FOR loop iterates over a constant, deterministic expression

The optimization then replaces the InsertNode and EnumerateListNode with a MultipleRemoteExecutionNode in the query execution plan, which takes care of inserting all documents into the collection in one go. Further optimizer rules are skipped if the optimization is triggered.


move-calculations-down

Move calculations down in the processing pipeline as far as possible (below FILTER, LIMIT and SUBQUERY nodes) so they are executed as late as possible and not before their results are required.


fuse-filters

Merges adjacent FILTER nodes together into a single FILTER node.


cluster-one-shard

Enterprise Edition only

Only available in cluster deployments.

Offload the entire query to the DB-Server (except the client communication via a Coordinator). This saves all the back and forth that normally exists in regular cluster queries, benefitting traversals and joins in particular.

Only for eligible queries in the OneShard deployment mode as well as for queries that only involve collection(s) with a single shard (and identical sharding in case of multiple collections, e.g. via distributeShardsLike). Queries involving V8 / JavaScript (e.g. user-defined AQL functions) or SmartGraphs cannot be optimized.


cluster-lift-constant-for-disjoint-graph-nodes

Enterprise Edition only

This rule cannot be turned off.

Only available in cluster deployments.

Detect SmartGraph traversals with a constant start vertex to prepare follow-up optimizations that can determine the shard location and push down calculations to a DB-Server.


distribute-in-cluster

This rule cannot be turned off.

Only available in cluster deployments.

Appears if query parts get distributed in a cluster.


smart-joins

Enterprise Edition only

Only available in cluster deployments.

Reduce inter-node joins to server-local joins. This rule is only employed when joining two collections with identical sharding setup via their shard keys.


scatter-in-cluster

This rule cannot be turned off.

Only available in cluster deployments.

Appears if nodes of the types ScatterNode, GatherNode, and RemoteNode are inserted into a distributed query plan.


scatter-satellite-graphs

Enterprise Edition only

Only available in cluster deployments.

Execute nodes of the types TraversalNode, ShortestPathNode, and KShortestPathsNode on a DB-Server instead of on a Coordinator if the nodes operate on SatelliteGraphs, removing the need to transfer data for these nodes.


remove-satellite-joins

Enterprise Edition only

Only available in cluster deployments.

Optimize nodes of the types ScatterNode, GatherNode, and RemoteNode for SatelliteCollections and SatelliteGraphs away. Execute the respective query parts on each participating DB-Server independently, so that the results become available locally without network communication. Depends on the remove-unnecessary-remote-scatter rule.


remove-distribute-nodes

Enterprise Edition only

Only available in cluster deployments.

Combine multiples nodes of type DistributeNode into one if two adjacent DistributeNode nodes share the same input variables and therefore can be optimized into a single DistributeNode.


distribute-offset-info-to-cluster

Enterprise Edition only

This rule cannot be turned off.

Only available in cluster deployments.

Push the calculation of search highlighting information to DB-Servers where the data for determining the offsets is stored.


distribute-filtercalc-to-cluster

Only available in cluster deployments.

Move filters up in a distributed execution plan. Filters are moved as far up in the plan as possible to make result sets as small as possible, as early as possible.


distribute-sort-to-cluster

Only available in cluster deployments.

Move sort operations up in a distributed query. Sorts are moved as far up in the query plan as possible to make result sets as small as possible, as early as possible.


move-filters-into-enumerate

Move filters on non-indexed collection attributes into IndexNode or EnumerateCollectionNode to allow early pruning of non-matching documents. This optimization can help to avoid a lot of temporary document copies. The optimization can also be applied to enumerations over non-collection array.


remove-unnecessary-remote-scatter

Only available in cluster deployments.

Avoid distributing calculations and handle them centrally if a RemoteNode is followed by a ScatterNode, and the ScatterNode is only followed by calculations or a SingletonNode.


undistribute-remove-after-enum-coll

Only available in cluster deployments.

Push nodes of type RemoveNode into the same query part that enumerates over the documents of a collection. This saves inter-cluster roundtrips between the EnumerateCollectionNode and the RemoveNode. It includes simple UPDATE and REPLACE operations that modify multiple documents and do not use LIMIT.


collect-in-cluster

Only available in cluster deployments.

Perform the heavy processing for COLLECT statements on DB-Servers and only light-weight aggregation on a Coordinator. Both sides get a CollectNode in the query plan.


sort-limit

Make SORT aware of a subsequent LIMIT to enable optimizations internal to the SortNode that allow to reduce memory usage and, in many cases, improve the sorting speed.

A SortNode needs to be followed by a LimitNode with no intervening nodes that may change the element count (e.g. a FilterNode which cannot be moved before the sort, or a source node like EnumerateCollectionNode).

The optimizer may choose not to apply the rule if it decides that it offers little or no benefit. In particular, it does not apply the rule if the input size is very small or if the output from the LimitNode is similar in size to the input. In exceptionally rare cases, this rule could result in some small slowdown. If observed, you can disable the rule for the affected query at the cost of increased memory usage.


reduce-extraction-to-projection

Modify EnumerationCollectionNode and IndexNode that would have extracted entire documents to only return a projection of each document.

Projections are limited to at most 5 different document attributes by default. The maximum number of projected attributes can optionally be adjusted by setting the maxProjections hint for an AQL FOR operation since ArangoDB 3.9.1.


restrict-to-single-shard

Only available in cluster deployments.

Restrict operations to a single shard instead of applying them for all shards if a collection operation (IndexNode or a data modification node) only affects a single shard.

This optimization can be applied for queries that access a collection only once in the query, and that do not use traversals, shortest path queries, and that do not access collection data dynamically using the DOCUMENT(), FULLTEXT(), NEAR() or WITHIN() AQL functions. Additionally, the optimizer can only apply this optimization if it can safely determine the values of all the collection's shard keys from the query, and when the shard keys are covered by a single index (this is always true if the shard key is the default _key).


optimize-count

Optimize subqueries to use an optimized code path for counting documents.

The requirements are that the subquery result must only be used with the COUNT() or LENGTH() AQL function and not for anything else. The subquery itself must be read-only (no data modification subquery), not use nested FOR loops, no LIMIT statement, and no FILTER condition or calculation that requires accessing document data. Accessing index data is supported for filtering but not for further calculations.


parallelize-gather

Only available in cluster deployments.

Apply an optimization to execute Coordinator GatherNode nodes in parallel. These notes cannot be parallelized if they depend on a TraversalNode, except for certain Disjoint SmartGraph traversals where the traversal can run completely on the local DB-Server.


decay-unnecessary-sorted-gather

Only available in cluster deployments.

Avoid merge-sorting results on a Coordinator if they are all from a single shard and fully sorted by a DB-Server already.


push-subqueries-to-dbserver

Enterprise Edition only

Only available in cluster deployments.

Execute subqueries entirely on a DB-Server if possible. Subqueries need to contain exactly one distribute/gather section, and only one collection access or traversal, shortest path, or k-shortest paths query.


late-document-materialization-arangosearch

Try to read from the underlying collections of a View as late as possible if the involved attributes are covered by the View index.


late-document-materialization

Try to read from collections as late as possible if the involved attributes are covered by inverted indexes.


push-limit-into-index

Push limit into index node.


batch-materialize-documents

Batch document lookup from indexes.


late-materialization-offset-info

Enterprise Edition only

Get the search highlighting offsets as late as possible to avoid unnecessary reads.


join-index-nodes

Join adjacent index nodes and replace them with a join node in case the indexes qualify for it.


push-down-late-materialization

Push down late materialization.


materialize-into-separate-variable

This rule cannot be turned off.

Introduce a separate variable for late materialization.


optimize-projections

Remove projections that are no longer used and store projection results in separate output registers.


remove-unnecessary-calculations-4

Fourth pass of removing all calculations whose result is not referenced in the query. This can be a consequence of applying other optimizations


async-prefetch

Allow query execution nodes to asynchronously prefetch the next batch while processing the current batch, allowing parts of the query to run in parallel. This is only possible for certain operations in a query.


immutable-search-condition

Optimize immutable search condition for nested loops, we don't need to make real search many times, if we can cache results in bitset


splice-subqueries

This rule cannot be turned off.

Appears if subqueries are spliced into the surrounding query, reducing overhead for executing subqueries by inlining the execution. This mainly benefits queries which execute subqueries very often that only return a few results at a time.

This optimization is performed on all subqueries and is applied after all other optimizations.


Additional optimizations applied

Scan-Only Optimization

If a query iterates over a collection (for filtering or counting) but does not need the actual document values later, the optimizer can apply a “scan-only” optimization for EnumerateCollectionNode and IndexNode node types. In this case, it does not build up a result with the document data at all, which may reduce work significantly. In case the document data is actually not needed later on, it may be sensible to remove it from query strings so the optimizer can apply the optimization.

If the optimization is applied, it shows up as scan only in an AQL query’s execution plan for an EnumerateCollectionNode or an IndexNode.

Index-Only Optimization

The optimizer can apply an “index-only” optimization for AQL queries that can satisfy the retrieval of all required document attributes directly from an index.

This optimization is triggered if an index is used that covers all required attributes of the document used later on in the query. If applied, it saves retrieving the actual document data (which would require an extra lookup by the storage engine), but instead builds the document data solely from the index values found. It only applies when using up to 5 (or maxProjections) attributes from the document, and only if the rest of the document data is not used later on in the query.

The optimization is available for the following index types: primary, edge, and persistent.

If the optimization is applied, it shows up as index only in an AQL query’s execution plan for an IndexNode.

Filter Projections Optimizations

Introduced: v3.10.0

If an index is used that does not cover all required attributes for the query, but if it is followed by filter conditions that only access attributes that are part of the index, then an optimization is applied, to only fetch matching documents. “Part of the index” here means, that all attributes referred to in the post-filter conditions are contained in the fields or storedValues attributes of the index definition.

For example, the optimization is applied in the following case:

  • There is a persistent index on the attributes [ "value1", "value2" ] (in this order), or there is a persistent index on just ["value1"] and with a storedValues definition of ["value2"].
  • There is a filter condition on value1 that can use the index, and a filter condition on value2 that cannot use the index (post-filter condition).

Example query:

FOR doc IN collection
  FILTER doc.value1 == @value1   /* uses the index */
  FILTER ABS(doc.value2) != @value2   /* does not use the index */
  RETURN doc

This query’s execution plan looks as follows:

Execution plan:
 Id   NodeType        Est.   Comment
  1   SingletonNode      1   * ROOT
  8   IndexNode          0     - FOR doc IN collection   /* persistent index scan (filter projections: `value2`) */    FILTER (ABS(doc.`value2`) != 2)   /* early pruning */   
  7   ReturnNode         0       - RETURN doc

Indexes used:
 By   Name                      Type         Collection   Unique   Sparse   Cache   Selectivity   Fields                   Ranges
  8   idx_1737498319258648576   persistent   collection   false    false    false       99.96 %   [ `value1`, `value2` ]   (doc.`value1` == 1)

The first filter condition is transformed to an index lookup, as you can tell from the persistent index scan comment and the Indexes used section that shows the range doc.`value` == 1. The post-filter condition FILTER ABS(doc.value2) != 2 can be recognized as such by the early pruning comment that follows it.

The filter projections mentioned in the above execution plan is an indicator of the optimization being triggered.

Instead of fetching the full documents from the storage engine for all index entries that matched the index lookup condition, only those that also satisfy the index lookup post-filter condition are fetched. If the post-filter condition filters out a lot of documents, this optimization can significantly speed up queries that produce large result sets from index lookups but filter many of the documents away with post-filter conditions.

Note that the optimization can also be combined with regular projections, e.g. for the following query that returns a specific attribute from the documents only:

FOR doc IN collection
  FILTER doc.value1 == @value1   /* uses the index */
  FILTER ABS(doc.value2) != @value2   /* does not use the index */
  RETURN doc.value3

That query’s execution plan combines projections from the index for the post-filter condition (filter projections) as well as regular projections (projections) for the processing parts of the query that follow the post-filter condition:

Execution plan:
 Id   NodeType          Est.   Comment
  1   SingletonNode        1   * ROOT
  9   IndexNode         5000     - FOR doc IN collection   /* persistent index scan (filter projections: `value2`) (projections: `value3`) */    FILTER (ABS(doc.`value2`) != 2)   /* early pruning */
  7   CalculationNode   5000       - LET #5 = doc.`value3`   /* attribute expression */   /* collections used: doc : collection */
  8   ReturnNode        5000       - RETURN #5

Indexes used:
 By   Name                      Type         Collection   Unique   Sparse   Cache   Selectivity   Fields                   Ranges
  9   idx_1737498319258648576   persistent   collection   false    false    false       99.96 %   [ `value1`, `value2` ]   (doc.`value1` == 1)

The optimization is most effective for queries in which many documents would be selected by the index lookup condition, but many are filtered out by the post-filter condition.