Michael Smith has posted comments on this change. ( http://gerrit.cloudera.org:8080/20719 )
Change subject: IMPALA-12433: Share buffers among channels in KrpcDataStreamSender ...................................................................... Patch Set 8: (10 comments) http://gerrit.cloudera.org:8080/#/c/20719/8//COMMIT_MSG Commit Message: http://gerrit.cloudera.org:8080/#/c/20719/8//COMMIT_MSG@12 PS8, Line 12: This patch switches to use a single buffer per channel. This is Please also discuss the partitioned case, as that seems to be handled differently. http://gerrit.cloudera.org:8080/#/c/20719/8//COMMIT_MSG@18 PS8, Line 18: calculation has several issues (see IMPALA-12433). Is this supposed to point to IMPALA-12594? http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/benchmarks/row-batch-serialize-benchmark.cc File be/src/benchmarks/row-batch-serialize-benchmark.cc: http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/benchmarks/row-batch-serialize-benchmark.cc@345 PS8, Line 345: // TODO: cleanup What cleanup is still needed? http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/catalog/catalog-util.cc File be/src/catalog/catalog-util.cc: http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/catalog/catalog-util.cc@358 PS8, Line 358: compressor->Close(); Please note why this was added in the commit message (i.e. what was missed before). http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/runtime/krpc-data-stream-sender.h File be/src/runtime/krpc-data-stream-sender.h: http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/runtime/krpc-data-stream-sender.h@172 PS8, Line 172: RowBatch* src, OutboundRowBatch* dest, bool compress, int num_receivers = 1); Please update the function comment to describe how 'compress' should be used. http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/runtime/krpc-data-stream-sender.cc File be/src/runtime/krpc-data-stream-sender.cc: http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/runtime/krpc-data-stream-sender.cc@123 PS8, Line 123: // backing the in-fligh RPC or kept as reserve if there is no ongoing RPC. The actual typo: in-flight http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/runtime/krpc-data-stream-sender.cc@128 PS8, Line 128: // unpartitioned case the shared outbound_batch_ holds the data for the the in-fligh RPC. typo: in-flight http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/runtime/krpc-data-stream-sender.cc@185 PS8, Line 185: Status TransmitData(std::unique_ptr<OutboundRowBatch>* outbound_batch, bool swap_batch); Please update the comment to describe when swap_batch should be true/false. http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/runtime/krpc-data-stream-sender.cc@244 PS8, Line 244: // TODO: replace this with a queue. Schedule another RPC callback in the Does this TODO still make sense? http://gerrit.cloudera.org:8080/#/c/20719/8/be/src/runtime/krpc-data-stream-sender.cc@601 PS8, Line 601: // At this point the previous RPC must be already finished and the previous buffer Is there a DCHECK we could add here? -- To view, visit http://gerrit.cloudera.org:8080/20719 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-Project: Impala-ASF Gerrit-Branch: master Gerrit-MessageType: comment Gerrit-Change-Id: I64854a350a9dae8bf3af11c871882ea4750e60b3 Gerrit-Change-Number: 20719 Gerrit-PatchSet: 8 Gerrit-Owner: Csaba Ringhofer <[email protected]> Gerrit-Reviewer: Impala Public Jenkins <[email protected]> Gerrit-Reviewer: Michael Smith <[email protected]> Gerrit-Comment-Date: Tue, 12 Dec 2023 22:31:02 +0000 Gerrit-HasComments: Yes
