walterddr commented on code in PR #11912:
URL: https://github.com/apache/pinot/pull/11912#discussion_r1377683135
##########
pinot-query-planner/src/main/java/org/apache/pinot/query/planner/logical/PlanFragmenter.java:
##########
@@ -126,49 +124,65 @@ public PlanNode visitExchange(ExchangeNode node, Context
context) {
if (!isPlanFragmentSplitter(node)) {
return process(node, context);
}
- int currentPlanFragmentId = context._previousPlanFragmentId;
- int nextPlanFragmentId = ++context._currentPlanFragmentId;
- // Set previous PlanFragment ID in the context to be the next PlanFragment
ID to be used by the child node.
- context._previousPlanFragmentId = nextPlanFragmentId;
- PlanNode nextPlanFragmentRoot = node.getInputs().get(0).visit(this,
context);
+ // Split the ExchangeNode to a MailboxReceiveNode and a MailboxSendNode,
where MailboxReceiveNode is the leave node
+ // of the current PlanFragment, and MailboxSendNode is the root node of
the next PlanFragment.
+ int receiverPlanFragmentId = context._currentPlanFragmentId;
+ int senderPlanFragmentId = context._nextPlanFragmentId.getAndIncrement();
Review Comment:
nit: visitor pattern is single threaded so i dont think we need to have
atomic integer here.
this "getAndIncrement" is still confusing --> if we are handling one context
per visit upon hitting an exchange then this is the only variable that's cross
all fragments: IIUC, the only reason this works is b/c the nextPlanFragmentId
is passed by reference so it is not really contain within the context?
##########
pinot-query-planner/src/main/java/org/apache/pinot/query/planner/logical/PlanFragmenter.java:
##########
@@ -126,49 +124,65 @@ public PlanNode visitExchange(ExchangeNode node, Context
context) {
if (!isPlanFragmentSplitter(node)) {
return process(node, context);
}
- int currentPlanFragmentId = context._previousPlanFragmentId;
- int nextPlanFragmentId = ++context._currentPlanFragmentId;
- // Set previous PlanFragment ID in the context to be the next PlanFragment
ID to be used by the child node.
- context._previousPlanFragmentId = nextPlanFragmentId;
- PlanNode nextPlanFragmentRoot = node.getInputs().get(0).visit(this,
context);
+ // Split the ExchangeNode to a MailboxReceiveNode and a MailboxSendNode,
where MailboxReceiveNode is the leave node
+ // of the current PlanFragment, and MailboxSendNode is the root node of
the next PlanFragment.
+ int receiverPlanFragmentId = context._currentPlanFragmentId;
+ int senderPlanFragmentId = context._nextPlanFragmentId.getAndIncrement();
+
context._planFragmentIdToChildrenMap.computeIfAbsent(receiverPlanFragmentId, k
-> new ArrayList<>())
+ .add(senderPlanFragmentId);
+
+ // Create a new context for the next PlanFragment with the new
PlanFragment ID.
+ Context nextPlanFragmentContext =
+ new Context(senderPlanFragmentId, context._nextPlanFragmentId,
context._planFragmentIdToRootNodeMap,
+ context._planFragmentIdToChildrenMap);
+ PlanNode nextPlanFragmentRoot = node.getInputs().get(0).visit(this,
nextPlanFragmentContext);
+
+ // Create a MailboxSendNode as the root node of the next PlanFragment.
PinotRelExchangeType exchangeType = node.getExchangeType();
RelDistribution.Type distributionType = node.getDistributionType();
// NOTE: Only HASH_DISTRIBUTED requires distribution keys
// TODO: Revisit ExchangeNode creation logic to avoid using
HASH_DISTRIBUTED with empty distribution keys
List<Integer> distributionKeys =
distributionType == RelDistribution.Type.HASH_DISTRIBUTED ?
node.getDistributionKeys() : null;
-
- PlanNode mailboxSender =
- new MailboxSendNode(nextPlanFragmentId,
nextPlanFragmentRoot.getDataSchema(), currentPlanFragmentId,
+ MailboxSendNode mailboxSendNode =
+ new MailboxSendNode(senderPlanFragmentId,
nextPlanFragmentRoot.getDataSchema(), receiverPlanFragmentId,
distributionType, exchangeType, distributionKeys,
node.getCollations(), node.isSortOnSender());
- PlanNode mailboxReceiver =
- new MailboxReceiveNode(currentPlanFragmentId,
nextPlanFragmentRoot.getDataSchema(), nextPlanFragmentId,
- distributionType, exchangeType, distributionKeys,
node.getCollations(), node.isSortOnSender(),
- node.isSortOnReceiver(), mailboxSender);
- mailboxSender.addInput(nextPlanFragmentRoot);
-
- context._planFragmentIdToRootNodeMap.put(nextPlanFragmentId,
- new PlanFragment(nextPlanFragmentId, mailboxSender, new
PlanFragmentMetadata(), new ArrayList<>()));
- if
(!context._planFragmentIdToChildrenMap.containsKey(currentPlanFragmentId)) {
- context._planFragmentIdToChildrenMap.put(currentPlanFragmentId, new
ArrayList<>());
- }
-
context._planFragmentIdToChildrenMap.get(currentPlanFragmentId).add(nextPlanFragmentId);
-
- return mailboxReceiver;
+ mailboxSendNode.addInput(nextPlanFragmentRoot);
+ context._planFragmentIdToRootNodeMap.put(senderPlanFragmentId,
+ new PlanFragment(senderPlanFragmentId, mailboxSendNode, new
PlanFragmentMetadata(), new ArrayList<>()));
+
+ // Return the MailboxReceiveNode as the leave node of the current
PlanFragment.
+ return new MailboxReceiveNode(receiverPlanFragmentId,
nextPlanFragmentRoot.getDataSchema(), senderPlanFragmentId,
+ distributionType, exchangeType, distributionKeys,
node.getCollations(), node.isSortOnSender(),
+ node.isSortOnReceiver(), mailboxSendNode);
}
private boolean isPlanFragmentSplitter(PlanNode node) {
return ((ExchangeNode) node).getExchangeType() !=
PinotRelExchangeType.SUB_PLAN;
}
public static class Context {
+ final int _currentPlanFragmentId;
+ final AtomicInteger _nextPlanFragmentId;
+ final Map<Integer, PlanFragment> _planFragmentIdToRootNodeMap;
+ final Map<Integer, List<Integer>> _planFragmentIdToChildrenMap;
+
+ public Context() {
Review Comment:
why do we need this public constructor? no one is using it right?
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]