impala-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Dan Hecht (Code Review)" <>
Subject [Impala-ASF-CR] IMPALA-4856: Port data stream service to KRPC
Date Wed, 01 Nov 2017 21:48:21 GMT
Dan Hecht has posted comments on this change. ( )

Change subject: IMPALA-4856: Port data stream service to KRPC

Patch Set 6:


Note to self: remaining files: krpc-data-stream-{mgr,recvr}.cc
File be/src/runtime/
PS6, Line 89:     "Number of datastream service processing threads");
how are these defaults chosen?
File be/src/runtime/row-batch.h:
PS6, Line 50: outbound
I think we should say something about KRPC to at least give that hint. maybe:

A KRPC outbound row batch...
PS6, Line 60: sizeof(int32_t)
sizeof(tuple_offsets_[0]) seems clearer and more robust
PS6, Line 354:   /// it is ignored. This function does not Reset().
we should preserve this comment when removing the thrift variant. So you could just put the
new decl here now so we don't forget that.
PS6, Line 424:   ///
nit: i don't think we generally have all these line breaks between parameter comments.
PS6, Line 426:  .
delete space
PS6, Line 444: nput_
PS6, Line 447: input_
PS6, Line 537:   std::string compression_scratch_;
this seems like a hack and we could do something simpler, but let's leave it alone for now.
File be/src/runtime/
PS6, Line 241:   // as sidecars to the RpcController.
this comment was probably meant to be deleted?
File common/protobuf/data_stream_service.proto:
PS6, Line 29: fragment
isn't this the id of the instance?  The comment in KrpcDataStreamSender is clearer, let's
copy that:
  /// Sender instance id, unique within a fragment.
  int sender_id_;
PS6, Line 59:   // Id of this fragment in its role as a sender.
File common/protobuf/row_batch.proto:
PS3, Line 32: = 2;
> That's the tuple data sent as sidecar. Clarified in the new comments.
My point is that writing it like 'tuple_data' doesn't make sense since it's not a field in
this struct. You should just write:
Size of the tuple data (sent as a sidecar) in bytes ...
File common/protobuf/row_batch.proto:
PS6, Line 32: epeated int32 row_tuples = 2;
why is this needed? i don't see it used. The size of it is used, though it seems like even
that can be inferred from the descriptors.

To view, visit
To unsubscribe, visit

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: Ic0b8c1e50678da66ab1547d16530f88b323ed8c1
Gerrit-Change-Number: 8023
Gerrit-PatchSet: 6
Gerrit-Owner: Michael Ho <>
Gerrit-Reviewer: Dan Hecht <>
Gerrit-Reviewer: Michael Ho <>
Gerrit-Reviewer: Mostafa Mokhtar <>
Gerrit-Reviewer: Sailesh Mukil <>
Gerrit-Comment-Date: Wed, 01 Nov 2017 21:48:21 +0000
Gerrit-HasComments: Yes

  • Unnamed multipart/alternative (inline, 8-Bit, 0 bytes)
View raw message