Tim Armstrong has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/15244 )

Change subject: IMPALA-5904: (part 2) Fix more TSAN bugs
......................................................................


Patch Set 4:

(1 comment)

Thanks for the fixes. One last issue..

http://gerrit.cloudera.org:8080/#/c/15244/4/bin/tsan-suppressions.txt
File bin/tsan-suppressions.txt:

http://gerrit.cloudera.org:8080/#/c/15244/4/bin/tsan-suppressions.txt@34
PS4, Line 34: # Accesses to KuduScanNode::done_ are intentionally racey; see 
IMPALA-4880 and the
> Yeah, we could, but looking at IMPALA-4880 it seems to be done this way int
I think the argument about optimistic checking also works for atomic. I think 
it's fine with the volatile, it would just be better to get it TSAN-clean at 
the same time as everything else.

Just to confirm, I checked that atomic operations with store/release generate 
the same instructions: https://godbolt.org/z/qA3KF2

If the functions are inlined into a bigger function the atomic operations will 
come with different compiler barriers, but even if the generated code was 
different, I doubt any perf effects would be significant when the check is done 
once per scan range.



--
To view, visit http://gerrit.cloudera.org:8080/15244
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I01feb40417dc5ea64ccb0c1044cfc3eed8508476
Gerrit-Change-Number: 15244
Gerrit-PatchSet: 4
Gerrit-Owner: Sahil Takiar <[email protected]>
Gerrit-Reviewer: Impala Public Jenkins <[email protected]>
Gerrit-Reviewer: Sahil Takiar <[email protected]>
Gerrit-Reviewer: Tim Armstrong <[email protected]>
Gerrit-Comment-Date: Tue, 25 Feb 2020 20:02:06 +0000
Gerrit-HasComments: Yes

Reply via email to