xingtanzjr commented on a change in pull request #5291:
URL: https://github.com/apache/iotdb/pull/5291#discussion_r830638330
##########
File path:
server/src/main/java/org/apache/iotdb/db/mpp/sql/planner/plan/DistributionPlanner.java
##########
@@ -29,7 +39,214 @@ public DistributionPlanner(Analysis analysis,
LogicalQueryPlan logicalPlan) {
this.logicalPlan = logicalPlan;
}
+ public PlanNode rewriteSource() {
+ SourceRewriter rewriter = new SourceRewriter();
+ return rewriter.visit(logicalPlan.getRootNode(), new
DistributionPlanContext());
+ }
+
+ public PlanNode addExchangeNode(PlanNode root) {
+ ExchangeNodeAdder adder = new ExchangeNodeAdder();
+ return adder.visit(root, new NodeGroupContext());
+ }
+
public DistributedQueryPlan planFragments() {
return null;
}
+
+ private class SourceRewriter extends
SimplePlanNodeRewriter<DistributionPlanContext> {
+
+ //TODO: (xingtanzjr) implement the method visitDeviceMergeNode()
+ public PlanNode visitDeviceMerge(TimeJoinNode node,
DistributionPlanContext context) {
+ return null;
+ }
+
+ public PlanNode visitTimeJoin(TimeJoinNode node, DistributionPlanContext
context) {
+ TimeJoinNode root = (TimeJoinNode) node.clone();
+
+ // Step 1: Get all source nodes. For the node which is not source, add
it as the child of
+ // current TimeJoinNode
+ List<SeriesScanNode> sources = new ArrayList<>();
+ for (PlanNode child : node.getChildren()) {
+ if (child instanceof SeriesScanNode) {
+ // If the child is SeriesScanNode, we need to check whether this
node should be seperated
+ // into several splits.
+ SeriesScanNode handle = (SeriesScanNode) child;
+ Set<DataRegion> dataDistribution =
+ analysis.getPartitionInfo(handle.getSeriesPath(),
handle.getTimeFilter());
+ // If the size of dataDistribution is m, this SeriesScanNode should
be seperated into m
+ // SeriesScanNode.
+ for (DataRegion dataRegion : dataDistribution) {
+ SeriesScanNode split = (SeriesScanNode) handle.clone();
+ split.setDataRegion(dataRegion);
+ sources.add(split);
+ }
+ } else if (child instanceof SeriesAggregateScanNode) {
+ // TODO: (xingtanzjr) We should do the same thing for
SeriesAggregateScanNode. Consider to
+ // make SeriesAggregateScanNode
+ // and SeriesScanNode to derived from the same parent Class because
they have similar
+ // process logic in many scenarios
+ } else {
+ // In a general logical query plan, the children of TimeJoinNode
should only be
+ // SeriesScanNode or SeriesAggregateScanNode
+ // So this branch should not be touched.
+ root.addChild(visit(child, context));
+ }
+ }
+
+ // Step 2: For the source nodes, group them by the DataRegion.
+ Map<DataRegion, List<SeriesScanNode>> sourceGroup = new HashMap<>();
+ sources.forEach(
+ source -> {
+ List<SeriesScanNode> group =
+ sourceGroup.containsKey(source.getDataRegion())
+ ? sourceGroup.get(source.getDataRegion())
+ : new ArrayList<>();
+ group.add(source);
+ sourceGroup.put(source.getDataRegion(), group);
+ });
+
+ // Step 3: For the source nodes which belong to same data region, add a
TimeJoinNode for them
+ // and make the
+ // new TimeJoinNode as the child of current TimeJoinNode
+ sourceGroup.forEach(
+ (dataRegion, seriesScanNodes) -> {
+ if (seriesScanNodes.size() == 1) {
+ root.addChild(seriesScanNodes.get(0));
+ } else {
+ // We clone a TimeJoinNode from root to make the params to be
consistent
+ TimeJoinNode parentOfGroup = (TimeJoinNode) root.clone();
+ seriesScanNodes.forEach(parentOfGroup::addChild);
+ root.addChild(parentOfGroup);
+ }
+ });
Review comment:
I added a TODO here to remind us the optimization. Because this code is
not really completed and let's complete it in later PR
##########
File path:
server/src/main/java/org/apache/iotdb/db/mpp/sql/planner/plan/DistributionPlanner.java
##########
@@ -29,7 +39,214 @@ public DistributionPlanner(Analysis analysis,
LogicalQueryPlan logicalPlan) {
this.logicalPlan = logicalPlan;
}
+ public PlanNode rewriteSource() {
+ SourceRewriter rewriter = new SourceRewriter();
+ return rewriter.visit(logicalPlan.getRootNode(), new
DistributionPlanContext());
+ }
+
+ public PlanNode addExchangeNode(PlanNode root) {
+ ExchangeNodeAdder adder = new ExchangeNodeAdder();
+ return adder.visit(root, new NodeGroupContext());
+ }
+
public DistributedQueryPlan planFragments() {
return null;
}
+
+ private class SourceRewriter extends
SimplePlanNodeRewriter<DistributionPlanContext> {
+
+ //TODO: (xingtanzjr) implement the method visitDeviceMergeNode()
+ public PlanNode visitDeviceMerge(TimeJoinNode node,
DistributionPlanContext context) {
+ return null;
+ }
+
+ public PlanNode visitTimeJoin(TimeJoinNode node, DistributionPlanContext
context) {
+ TimeJoinNode root = (TimeJoinNode) node.clone();
+
+ // Step 1: Get all source nodes. For the node which is not source, add
it as the child of
+ // current TimeJoinNode
+ List<SeriesScanNode> sources = new ArrayList<>();
+ for (PlanNode child : node.getChildren()) {
+ if (child instanceof SeriesScanNode) {
+ // If the child is SeriesScanNode, we need to check whether this
node should be seperated
+ // into several splits.
+ SeriesScanNode handle = (SeriesScanNode) child;
+ Set<DataRegion> dataDistribution =
+ analysis.getPartitionInfo(handle.getSeriesPath(),
handle.getTimeFilter());
+ // If the size of dataDistribution is m, this SeriesScanNode should
be seperated into m
+ // SeriesScanNode.
+ for (DataRegion dataRegion : dataDistribution) {
+ SeriesScanNode split = (SeriesScanNode) handle.clone();
+ split.setDataRegion(dataRegion);
+ sources.add(split);
+ }
+ } else if (child instanceof SeriesAggregateScanNode) {
+ // TODO: (xingtanzjr) We should do the same thing for
SeriesAggregateScanNode. Consider to
+ // make SeriesAggregateScanNode
+ // and SeriesScanNode to derived from the same parent Class because
they have similar
+ // process logic in many scenarios
+ } else {
+ // In a general logical query plan, the children of TimeJoinNode
should only be
+ // SeriesScanNode or SeriesAggregateScanNode
+ // So this branch should not be touched.
+ root.addChild(visit(child, context));
+ }
+ }
+
+ // Step 2: For the source nodes, group them by the DataRegion.
+ Map<DataRegion, List<SeriesScanNode>> sourceGroup = new HashMap<>();
+ sources.forEach(
+ source -> {
+ List<SeriesScanNode> group =
+ sourceGroup.containsKey(source.getDataRegion())
+ ? sourceGroup.get(source.getDataRegion())
+ : new ArrayList<>();
+ group.add(source);
+ sourceGroup.put(source.getDataRegion(), group);
+ });
Review comment:
Done
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
