Thomas Marshall has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/12297 )

Change subject: IMPALA-8138: Reintroduce rpc debugging options
......................................................................


Patch Set 2:

(3 comments)

This is not really ready for full review until 
https://gerrit.cloudera.org/#/c/12672/ is submitted and it is rebased on top, I 
just updated it so that reviewers of that patch can see what I've done for the 
async case.

http://gerrit.cloudera.org:8080/#/c/12297/1/be/src/runtime/krpc-backend-client.h
File be/src/runtime/krpc-backend-client.h:

http://gerrit.cloudera.org:8080/#/c/12297/1/be/src/runtime/krpc-backend-client.h@27
PS1, Line 27:
> Please add a class level comment.
Done


http://gerrit.cloudera.org:8080/#/c/12297/1/be/src/runtime/krpc-backend-client.h@28
PS1, Line 28:
> Should this class show that it is a decorated form of
 > ControlServiceProxy, perhaps by having ControlServiceProxy as part
 > of it's name?  ControlServiceProxyInjected? ControlServiceProxyDecorated?

Good idea. "Injected" and "Decorated" seem awkward to me, is it clear enough if 
I go with ImpalaControlServiceProxy?

 >
 > I think this change introduces a convention, that we expect
 > ControlServiceProxy methods to be overridden in KrpcBackendClient?
 > If that's so then a comment somewhere should say that. Is it in
 > fact now a mistake to use ControlServiceProxy?

Addressed in the class comment.

 >
 > It seems like KrpcBackendClient is mostly boilerplate. Did you
 > consider extending protoc-gen-krpc.cc to either generate the
 > decorated methods in ControlServiceProxy, or to generate
 > KrpcBackendClient?

As discussed, while this is an interesting idea, the Kudu people don't have 
much interest in directly adding such a feature to krpc, and for now at least 
the benefit of automating this boiler-plate is probably not worth the work.


http://gerrit.cloudera.org:8080/#/c/12297/1/be/src/runtime/query-state.h
File be/src/runtime/query-state.h:

http://gerrit.cloudera.org:8080/#/c/12297/1/be/src/runtime/query-state.h@314
PS1, Line 314: eProxy
> Will client_ be a more appropriate name ? Same for other places.
I assume that if I change the class name back to something ending in "Proxy" 
that you're fine leaving this as is?



--
To view, visit http://gerrit.cloudera.org:8080/12297
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I2046cb9dadf846ea90c04e95781b2bbde3325941
Gerrit-Change-Number: 12297
Gerrit-PatchSet: 2
Gerrit-Owner: Thomas Marshall <[email protected]>
Gerrit-Reviewer: Andrew Sherman <[email protected]>
Gerrit-Reviewer: Impala Public Jenkins <[email protected]>
Gerrit-Reviewer: Michael Ho <[email protected]>
Gerrit-Reviewer: Thomas Marshall <[email protected]>
Gerrit-Comment-Date: Thu, 07 Mar 2019 19:33:42 +0000
Gerrit-HasComments: Yes

Reply via email to