com.hazelcast.org.apache.calcite.plan.volcano.OptimizeTask Maven / Gradle / Ivy
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to you under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.hazelcast.org.apache.calcite.plan.volcano;
import com.hazelcast.org.apache.calcite.plan.Convention;
import com.hazelcast.org.apache.calcite.plan.ConventionTraitDef;
import com.hazelcast.org.apache.calcite.plan.DeriveMode;
import com.hazelcast.org.apache.calcite.plan.RelTrait;
import com.hazelcast.org.apache.calcite.plan.RelTraitSet;
import com.hazelcast.org.apache.calcite.rel.PhysicalNode;
import com.hazelcast.org.apache.calcite.rel.RelNode;
import com.hazelcast.org.apache.calcite.util.trace.CalciteTrace;
import org.apiguardian.api.API;
import com.hazelcast.org.slf4j.Logger;
import java.util.ArrayList;
import java.util.HashSet;
import java.util.List;
import java.util.Set;
/**
* OptimizeTask represents the optimization task
* of VolcanoPlanner.
*
* How does it work?
*
* Let S# denote the seed physical operator in a RelSet after
* logical and physical rules transformation, P# denote the
* physical operator generated by passing down parent trait
* requirements, D# denote the physical operator generated by
* deriving from child delivered traitSets.
*
* The initial rel list state in a RelSet is as follows:
*
* cursor
* |
* V
* S1, S2
*
*
* When we create a task for RelSubset1, the task will immediately
* pass the subset's traitSet to seed operators, S1 and S2,
* now we have:
*
* cursor
* |
* V
* S1, S2, P1, P2
*
*
* The subset task will create a optimization task for the relnode
* pointed by cursor, and move cursor to next available physical
* operator S2. In the task for S1, it will continue optimize its
* child nodes, which are RelSubsets. After child inputs optimization
* is finished, S1 will derive new relnodes from delivered subsets
* in input RelSet. Once task for S1 is completed, we have:
*
* cursor
* |
* V
* S1, S2, P1, P2, D1
*
*
* The subset task continues scheduling task for S2, P1... until
* there is no more relnode created for the RelSet, then we have:
*
* cursor
* |
* V
* S1, S2, P1, P2, D1, D2, D3, null
*
*
* When a task for another RelSubset2 is created, the task will try
* to pass down the subset's traitSet to seed operator S1 and S2,
* now the RelSet looks like:
*
* cursor
* |
* V
* S1, S2, P1, P2, D1, D2, D3, P3, P4
*
*
* The process continues till there is no more subsets or relnodes
* created for the RelSet.
*/
@API(since = "1.23", status = API.Status.INTERNAL)
abstract class OptimizeTask {
static final Logger LOGGER = CalciteTrace.getPlannerTaskTracer();
static OptimizeTask create(RelNode node) {
if (node instanceof RelSubset) {
return new RelSubsetOptTask((RelSubset) node);
}
return new RelNodeOptTask(node);
}
final VolcanoPlanner planner;
final int id;
OptimizeTask(RelNode node) {
planner = (VolcanoPlanner) node.getCluster().getPlanner();
id = planner.nextTaskId++;
LOGGER.debug("Scheduled task(id={}) for {}", id, node);
}
abstract boolean hasSubTask();
abstract OptimizeTask nextSubTask();
abstract void execute();
/**
* Task State
*/
public enum State {
SCHEDULED,
EXECUTING,
COMPLETED
}
/**
* Task for optimizing RelNode.
*
* Optimize input RelSubsets and derive new RelNodes
* from child traitSets.
*/
static class RelNodeOptTask extends OptimizeTask {
final RelNode node;
int nextId = 0; // next child index
RelNodeOptTask(RelNode node) {
super(node);
this.node = node;
}
@Override boolean hasSubTask() {
int size = node.getInputs().size();
while (nextId < size) {
RelSubset subset = (RelSubset) node.getInput(nextId);
if (subset.taskState == null) {
// not yet scheduled
return true;
} else {
// maybe a cycle if it is not completed
nextId++;
}
}
return false;
}
@Override OptimizeTask nextSubTask() {
RelNode child = node.getInput(nextId++);
return new RelSubsetOptTask((RelSubset) child);
}
@Override void execute() {
if (!(node instanceof PhysicalNode)
|| ((PhysicalNode) node).getDeriveMode() == DeriveMode.PROHIBITED
|| !planner.isSeedNode(node)) {
LOGGER.debug("Completed task(id={}) for {}", id, node);
return;
}
PhysicalNode rel = (PhysicalNode) node;
DeriveMode mode = rel.getDeriveMode();
int arity = node.getInputs().size();
// for OMAKASE
List> inputTraits = new ArrayList<>(arity);
for (int i = 0; i < arity; i++) {
int childId = i;
if (mode == DeriveMode.RIGHT_FIRST) {
childId = arity - i - 1;
}
RelSubset input = (RelSubset) node.getInput(childId);
List traits = new ArrayList<>();
inputTraits.add(traits);
final int numSubset = input.set.subsets.size();
for (int j = 0; j < numSubset; j++) {
RelSubset subset = input.set.subsets.get(j);
if (!subset.isDelivered() || equalsSansConvention(
subset.getTraitSet(), rel.getCluster().traitSet())) {
// TODO: should use matching type to determine
// Ideally we should stop deriving new relnodes when the
// subset's traitSet equals with input traitSet, but
// in case someone manually builds a physical relnode
// tree, which is highly discouraged, without specifying
// correct traitSet, e.g.
// EnumerableFilter [].ANY
// -> EnumerableMergeJoin [a].Hash[a]
// We should still be able to derive the correct traitSet
// for the dumb filter, even though the filter's traitSet
// should be derived from the MergeJoin when it is created.
// But if the subset's traitSet equals with the default
// empty traitSet sans convention (the default traitSet
// from cluster may have logical convention, NONE, which
// is not interesting), we are safe to ignore it, because
// a physical filter with non default traitSet, but has a
// input with default empty traitSet, e.g.
// EnumerableFilter [a].Hash[a]
// -> EnumerableProject [].ANY
// is definitely wrong, we should fail fast.
continue;
}
if (mode == DeriveMode.OMAKASE) {
traits.add(subset.getTraitSet());
} else {
RelNode newRel = rel.derive(subset.getTraitSet(), childId);
if (newRel != null && !planner.isRegistered(newRel)) {
RelSubset relSubset = planner.register(newRel, node);
assert relSubset.set == planner.getSubset(node).set;
}
}
}
if (mode == DeriveMode.LEFT_FIRST
|| mode == DeriveMode.RIGHT_FIRST) {
break;
}
}
if (mode == DeriveMode.OMAKASE) {
List relList = rel.derive(inputTraits);
for (RelNode relNode : relList) {
if (!planner.isRegistered(relNode)) {
planner.register(relNode, node);
}
}
}
LOGGER.debug("Completed task(id={}) for {}", id, node);
}
/**
* Returns whether the 2 traitSets are equal without Convention.
* It assumes they have the same traitDefs order.
*/
private boolean equalsSansConvention(RelTraitSet ts1, RelTraitSet ts2) {
assert ts1.size() == ts2.size();
for (int i = 0; i < ts1.size(); i++) {
RelTrait trait = ts1.getTrait(i);
if (trait.getTraitDef() == ConventionTraitDef.INSTANCE) {
continue;
}
if (!trait.equals(ts2.getTrait(i))) {
return false;
}
}
return true;
}
@Override public String toString() {
return "Task#" + id + ":{ " + node + " }";
}
}
/**
* Task for optimizing RelSubset.
*
* Pass down the trait requirements of current RelSubset
* and add enforcers to the new delivered subsets.
*/
static class RelSubsetOptTask extends OptimizeTask {
final RelSubset subset;
final Set deliveredSubsets = new HashSet<>();
RelSubsetOptTask(RelSubset subset) {
super(subset);
this.subset = subset;
subset.taskState = State.SCHEDULED;
propagateTraits();
}
private void propagateTraits() {
int size = subset.set.getSeedSize();
for (int i = 0; i < size; i++) {
RelNode rel = subset.set.rels.get(i);
if (!(rel instanceof PhysicalNode)
|| rel.getConvention() == Convention.NONE
|| rel.getTraitSet().satisfies(subset.getTraitSet())) {
continue;
}
RelNode node = ((PhysicalNode) rel).passThrough(
subset.getTraitSet());
if (node != null && !planner.isRegistered(node)) {
RelSubset newSubset = planner.register(node, subset);
deliveredSubsets.add(newSubset);
} else {
// TODO: should we consider stop trying propagation on node
// with the same traitset as phyNode?
assert true;
}
}
}
@Override boolean hasSubTask() {
return subset.set.hasNextPhysicalNode();
}
@Override OptimizeTask nextSubTask() {
RelNode rel = subset.set.nextPhysicalNode();
return new RelNodeOptTask(rel);
}
@Override void execute() {
subset.taskState = State.EXECUTING;
subset.set.addConverters(subset, true, false);
for (RelSubset delivered : deliveredSubsets) {
subset.set.addConverters(delivered, false, false);
}
subset.taskState = State.COMPLETED;
LOGGER.debug("Completed task(id={}) for {}", id, subset);
}
@Override public String toString() {
return "Task#" + id + ":{ " + subset + " }";
}
}
}
© 2015 - 2025 Weber Informatics LLC | Privacy Policy