walterddr commented on code in PR #9064:
URL: https://github.com/apache/pinot/pull/9064#discussion_r928123071
##########
pinot-query-runtime/src/main/java/org/apache/pinot/query/runtime/operator/AggregateOperator.java:
##########
@@ -120,14 +108,17 @@ public String toExplainString() {
protected TransferableBlock getNextBlock() {
try {
cumulateAggregationBlocks();
- return new TransferableBlock(toResultBlock());
+ return toResultBlock();
} catch (Exception e) {
return TransferableBlockUtils.getErrorTransferableBlock(e);
}
}
- private BaseDataBlock toResultBlock()
+ private TransferableBlock toResultBlock()
throws IOException {
+ if (_upstreamErrorBlock != null) {
Review Comment:
This is tricky, so long-story short. the mailbox send/receive is only
special because of the "random distribution" exchange type.
without it. the connection will be fully connected (for data block)
- in either broadcast or hash distribution, all sender servers will send
data to all receiver servers
- in singleton distribution, the list of sender and receiver will only do
self-connection.
- where as in random distribution, we send data to one of the receiver
servers
however, for metadata block. EXACTLY BECAUSE the behavior of random
distribution, we need to broadcast the end-of-stream info to every receiver
since they might've still expect a data block to come if we dont do so.
so in summary
- error blocks are broadcast to all receivers.
- except for singleton, since they established the 1-to-1 relation at the
beginning.
--
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]
