EmmyMiao87 commented on a change in pull request #1566: Add switch for cost
optimization
URL: https://github.com/apache/incubator-doris/pull/1566#discussion_r310422337
##########
File path: fe/src/main/java/org/apache/doris/planner/DistributedPlanner.java
##########
@@ -287,59 +287,73 @@ private PlanFragment createHashJoinFragment(HashJoinNode
node, PlanFragment righ
PlanFragment
leftChildFragment, long perNodeMemLimit,
ArrayList<PlanFragment>
fragments)
throws UserException {
- // broadcast: send the rightChildFragment's output to each node
executing
- // the leftChildFragment; the cost across all nodes is proportional to
the
- // total amount of data sent
- PlanNode rhsTree = rightChildFragment.getPlanRoot();
- long rhsDataSize = 0;
- long broadcastCost = 0;
- if (rhsTree.getCardinality() != -1 && leftChildFragment.getNumNodes()
!= -1) {
- rhsDataSize = Math.round((double) rhsTree.getCardinality() *
rhsTree.getAvgRowSize());
- broadcastCost = rhsDataSize * leftChildFragment.getNumNodes();
- }
- LOG.info("broadcast: cost=" + Long.toString(broadcastCost));
- LOG.info("card=" + Long.toString(rhsTree.getCardinality())
- + " row_size=" + Float.toString(rhsTree.getAvgRowSize())
- + " #nodes=" +
Integer.toString(leftChildFragment.getNumNodes()));
-
- // repartition: both left- and rightChildFragment are partitioned on
the
- // join exprs
- // TODO: take existing partition of input fragments into account to
avoid
- // unnecessary repartitioning
- PlanNode lhsTree = leftChildFragment.getPlanRoot();
- long partitionCost = 0;
- if (lhsTree.getCardinality() != -1 && rhsTree.getCardinality() != -1) {
- partitionCost = Math.round(
- (double) lhsTree.getCardinality() *
lhsTree.getAvgRowSize() + (double) rhsTree
- .getCardinality() * rhsTree.getAvgRowSize());
- }
- LOG.info("partition: cost=" + Long.toString(partitionCost));
- LOG.info(
- "lhs card=" + Long.toString(lhsTree.getCardinality()) + "
row_size=" + Float.toString(
- lhsTree.getAvgRowSize()));
- LOG.info(
- "rhs card=" + Long.toString(rhsTree.getCardinality()) + "
row_size=" + Float.toString(
- rhsTree.getAvgRowSize()));
- LOG.info(rhsTree.getExplainString());
-
boolean doBroadcast;
- // we do a broadcast join if
- // - we're explicitly told to do so
- // - or if it's cheaper and we weren't explicitly told to do a
partitioned join
- // - and we're not doing a full or right outer join (those require the
left-hand
- // side to be partitioned for correctness)
- // - and the expected size of the hash tbl doesn't exceed
perNodeMemLimit
- // we do a "<=" comparison of the costs so that we default to
broadcast joins if
- // we're unable to estimate the cost
- if (node.getJoinOp() != JoinOperator.RIGHT_OUTER_JOIN
- && node.getJoinOp() != JoinOperator.FULL_OUTER_JOIN
- && (perNodeMemLimit == 0 || Math.round(
- (double) rhsDataSize * PlannerContext.HASH_TBL_SPACE_OVERHEAD)
<= perNodeMemLimit)
- && (node.getInnerRef().isBroadcastJoin() ||
(!node.getInnerRef().isPartitionJoin()
- && broadcastCost <= partitionCost))) {
- doBroadcast = true;
+ if
(!ctx_.getRootAnalyzer().getContext().getSessionVariable().isDisableCostOptimization())
{
+ // broadcast: send the rightChildFragment's output to each node
executing
+ // the leftChildFragment; the cost across all nodes is
proportional to the
+ // total amount of data sent
+ PlanNode rhsTree = rightChildFragment.getPlanRoot();
+ long rhsDataSize = 0;
+ long broadcastCost = 0;
+ if (rhsTree.getCardinality() != -1 &&
leftChildFragment.getNumNodes() != -1) {
+ rhsDataSize = Math.round((double) rhsTree.getCardinality() *
rhsTree.getAvgRowSize());
+ broadcastCost = rhsDataSize * leftChildFragment.getNumNodes();
+ }
+ LOG.info("broadcast: cost=" + Long.toString(broadcastCost));
+ LOG.info("card=" + Long.toString(rhsTree.getCardinality())
+ + " row_size=" + Float.toString(rhsTree.getAvgRowSize())
+ + " #nodes=" +
Integer.toString(leftChildFragment.getNumNodes()));
+
+ // repartition: both left- and rightChildFragment are partitioned
on the
+ // join exprs
+ // TODO: take existing partition of input fragments into account
to avoid
+ // unnecessary repartitioning
+ PlanNode lhsTree = leftChildFragment.getPlanRoot();
+ long partitionCost = 0;
+ if (lhsTree.getCardinality() != -1 && rhsTree.getCardinality() !=
-1) {
+ partitionCost = Math.round(
+ (double) lhsTree.getCardinality() *
lhsTree.getAvgRowSize() + (double) rhsTree
+ .getCardinality() * rhsTree.getAvgRowSize());
+ }
+ LOG.info("partition: cost=" + Long.toString(partitionCost));
+ LOG.info(
+ "lhs card=" + Long.toString(lhsTree.getCardinality()) + "
row_size=" + Float.toString(
+ lhsTree.getAvgRowSize()));
+ LOG.info(
+ "rhs card=" + Long.toString(rhsTree.getCardinality()) + "
row_size=" + Float.toString(
+ rhsTree.getAvgRowSize()));
+ LOG.info(rhsTree.getExplainString());
+
+ // we do a broadcast join if
+ // - we're explicitly told to do so
+ // - or if it's cheaper and we weren't explicitly told to do a
partitioned join
+ // - and we're not doing a full or right outer join (those require
the left-hand
+ // side to be partitioned for correctness)
+ // - and the expected size of the hash tbl doesn't exceed
perNodeMemLimit
+ // we do a "<=" comparison of the costs so that we default to
broadcast joins if
+ // we're unable to estimate the cost
+ if (node.getJoinOp() != JoinOperator.RIGHT_OUTER_JOIN
+ && node.getJoinOp() != JoinOperator.FULL_OUTER_JOIN
+ && node.getJoinOp() != JoinOperator.RIGHT_ANTI_JOIN
+ && node.getJoinOp() != JoinOperator.RIGHT_SEMI_JOIN
+ && (perNodeMemLimit == 0 || Math.round(
+ (double) rhsDataSize *
PlannerContext.HASH_TBL_SPACE_OVERHEAD) <= perNodeMemLimit)
+ && (node.getInnerRef().isBroadcastJoin() ||
(!node.getInnerRef().isPartitionJoin()
+ && broadcastCost <= partitionCost))) {
+ doBroadcast = true;
+ } else {
+ doBroadcast = false;
+ }
} else {
- doBroadcast = false;
+ if (node.getJoinOp() != JoinOperator.RIGHT_OUTER_JOIN
+ && node.getJoinOp() != JoinOperator.RIGHT_ANTI_JOIN
+ && node.getJoinOp() != JoinOperator.RIGHT_SEMI_JOIN
+ && node.getJoinOp() != JoinOperator.FULL_OUTER_JOIN
+ && !node.getInnerRef().isPartitionJoin()) {
Review comment:
Does perNodeMemLimit need to be considered?
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@doris.apache.org
For additional commands, e-mail: dev-h...@doris.apache.org
