Alex Behm has posted comments on this change. Change subject: IMPALA-4023: don't attach buffered tuple streams to batches ......................................................................
Patch Set 1: (2 comments) Changes make a lot of sense to me. http://gerrit.cloudera.org:8080/#/c/4448/1/be/src/runtime/buffered-tuple-stream-test.cc File be/src/runtime/buffered-tuple-stream-test.cc: Line 1185: } Add a simple test that checks pinned blocks are attached to a non-NULL batch when calling Close() on the stream. http://gerrit.cloudera.org:8080/#/c/4448/1/be/src/runtime/row-batch.h File be/src/runtime/row-batch.h: Line 129: bool ALWAYS_INLINE AtCapacity() { The meaning of AtCapacity() has slightly changed and I think it might be a problem, but you can convince me otherwise :) 1. I think we still need to check need_to_return_ 2. Before, when attaching a stream to a batch, that batch would always be considered at capacity. Now we are attaching pinned blocks, but those don't imply AtCapacity(). I'm worried about running into a scenario where we call GetNext() on an exec node before all those pinned blocks are unpinned/freed. -- To view, visit http://gerrit.cloudera.org:8080/4448 To unsubscribe, visit http://gerrit.cloudera.org:8080/settings Gerrit-MessageType: comment Gerrit-Change-Id: I6471422f86ce71e4c6ab277a276000051bc2e8ff Gerrit-PatchSet: 1 Gerrit-Project: Impala-ASF Gerrit-Branch: master Gerrit-Owner: Tim Armstrong <[email protected]> Gerrit-Reviewer: Alex Behm <[email protected]> Gerrit-Reviewer: Michael Ho Gerrit-Reviewer: Tim Armstrong <[email protected]> Gerrit-HasComments: Yes
