Package org.elasticsearch.xpack.esql.plugin

Class ComputeService

java.lang.Object

org.elasticsearch.xpack.esql.plugin.ComputeService

public class ComputeService
extends Object

Once query is parsed and validated it is scheduled for execution by org.elasticsearch.xpack.esql.plugin.ComputeService#execute This method is responsible for splitting physical plan into coordinator and data node plans.

Coordinator plan is immediately executed locally (using org.elasticsearch.xpack.esql.plugin.ComputeService#runCompute) and is prepared to collect and merge pages from data nodes into the final query result.

Data node plan is passed to org.elasticsearch.xpack.esql.plugin.DataNodeComputeHandler#startComputeOnDataNodes that is responsible for

• Determining list of nodes that contain shards referenced by the query with org.elasticsearch.xpack.esql.plugin.DataNodeRequestSender#searchShards
• Each node in the list processed in org.elasticsearch.xpack.esql.plugin.DataNodeComputeHandler#startComputeOnDataNodes in order to
  • Open ExchangeSink on the target data node and link it with local ExchangeSource for the query using `internal:data/read/esql/open_exchange` transport request.
  • Start data node plan execution on the target data node using `indices:data/read/esql/data` transport request
  • While coordinator plan executor is running it will read data from ExchangeSource that will poll pages from linked ExchangeSink on target data nodes or notify them that data set is already completed (for example when running FROM * | LIMIT 10 type of query) or query is canceled using `internal:data/read/esql/exchange` transport requests.

Field Summary

Fields

Modifier and Type	Field	Description
static final String	CLUSTER_ACTION_NAME
static final String	DATA_ACTION_NAME
static final String	DATA_DESCRIPTION
static final String	REDUCE_DESCRIPTION

Constructor Summary

Constructors

Constructor	Description
ComputeService(TransportActionServices transportActionServices, EnrichLookupService enrichLookupService, LookupFromIndexService lookupFromIndexService, ThreadPool threadPool, BigArrays bigArrays, BlockFactory blockFactory, OperatorFactoryRegistry operatorFactoryRegistry, FormatReaderRegistry formatReaderRegistry)

Method Summary

Modifier and Type	Method	Description
static boolean	canSkipSplitDiscovery(PhysicalPlan plan, FormatReaderRegistry formatReaderRegistry)	Returns true when split discovery (Phase 2 footer reads) can be skipped because every fragment in plan that references an ExternalRelation is a pure ungrouped Aggregate directly over that relation, the relation's sourceMetadata already contains complete (non-partial) statistics, and the fragment's aggregate functions are metadata-pushable for that source's format reader (i.e.
EsqlFlags	createFlags()
void	execute(String sessionId, CancellableTask rootTask, EsqlFlags flags, PhysicalPlan physicalPlan, Configuration configuration, FoldContext foldContext, EsqlExecutionInfo execInfo, PlanTimeProfile planTimeProfile, ActionListener<Result> listener)
void	executePlan(String sessionId, CancellableTask rootTask, EsqlFlags flags, PhysicalPlan physicalPlan, Configuration configuration, FoldContext foldContext, EsqlExecutionInfo execInfo, String profileQualifier, ActionListener<Result> listener, Supplier<ExchangeSink> exchangeSinkSupplier, Map<String,EsqlExecutionInfo.Cluster.Status> initialClusterStatuses, PlanTimeProfile planTimeProfile)
static ReductionPlan	reductionPlan(PlannerSettings plannerSettings, EsqlFlags flags, Configuration configuration, FoldContext foldCtx, ExchangeSinkExec originalPlan, boolean runNodeLevelReduction, boolean reduceNodeLateMaterialization, PlanTimeProfile planTimeProfile)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

DATA_DESCRIPTION

public static final String DATA_DESCRIPTION

See Also:

• Constant Field Values

REDUCE_DESCRIPTION

public static final String REDUCE_DESCRIPTION

See Also:

• Constant Field Values

DATA_ACTION_NAME

public static final String DATA_ACTION_NAME

See Also:

• Constant Field Values

CLUSTER_ACTION_NAME

public static final String CLUSTER_ACTION_NAME

See Also:

• Constant Field Values

Constructor Details

ComputeService

public ComputeService(TransportActionServices transportActionServices,
 EnrichLookupService enrichLookupService,
 LookupFromIndexService lookupFromIndexService,
 ThreadPool threadPool,
 BigArrays bigArrays,
 BlockFactory blockFactory,
 OperatorFactoryRegistry operatorFactoryRegistry,
 FormatReaderRegistry formatReaderRegistry)

Method Details

canSkipSplitDiscovery

public static boolean canSkipSplitDiscovery(PhysicalPlan plan,
 FormatReaderRegistry formatReaderRegistry)

Returns true when split discovery (Phase 2 footer reads) can be skipped because every fragment in plan that references an ExternalRelation is a pure ungrouped Aggregate directly over that relation, the relation's sourceMetadata already contains complete (non-partial) statistics, and the fragment's aggregate functions are metadata-pushable for that source's format reader (i.e. AggregatePushdownSupport.canPushAggregates(java.util.List<org.elasticsearch.xpack.esql.core.expression.Expression>, java.util.List<org.elasticsearch.xpack.esql.core.expression.Expression>) returns YES).

When eligible, the local physical optimizer uses sourceMetadata statistics to evaluate the aggregates (COUNT/MIN/MAX) without reading any data files. Skipping discovery for non-pushable aggregates (e.g. SUM, AVG) would force the data node to fall back to a single-path read with no slice queue, which is typically less parallel than per-row-group splits — hence the explicit pushability check.

This method is conservative: if the registry has no reader for the source type, if any fragment containing an ExternalRelation does not match the Aggregate -> ExternalRelation shape, or if any required statistic is missing, it returns false so that normal split discovery proceeds.

execute

public void execute(String sessionId,
 CancellableTask rootTask,
 EsqlFlags flags,
 PhysicalPlan physicalPlan,
 Configuration configuration,
 FoldContext foldContext,
 EsqlExecutionInfo execInfo,
 PlanTimeProfile planTimeProfile,
 ActionListener<Result> listener)

executePlan

public void executePlan(String sessionId,
 CancellableTask rootTask,
 EsqlFlags flags,
 PhysicalPlan physicalPlan,
 Configuration configuration,
 FoldContext foldContext,
 EsqlExecutionInfo execInfo,
 String profileQualifier,
 ActionListener<Result> listener,
 Supplier<ExchangeSink> exchangeSinkSupplier,
 Map<String,EsqlExecutionInfo.Cluster.Status> initialClusterStatuses,
 PlanTimeProfile planTimeProfile)

reductionPlan

public static ReductionPlan reductionPlan(PlannerSettings plannerSettings,
 EsqlFlags flags,
 Configuration configuration,
 FoldContext foldCtx,
 ExchangeSinkExec originalPlan,
 boolean runNodeLevelReduction,
 boolean reduceNodeLateMaterialization,
 PlanTimeProfile planTimeProfile)

createFlags

public EsqlFlags createFlags()