impala-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tianyi Wang (Code Review)" <>
Subject [Impala-ASF-CR] IMPALA-5789: Add always false flag in bloom filter
Date Mon, 09 Oct 2017 21:07:01 GMT
Tianyi Wang has posted comments on this change. ( )

Change subject: IMPALA-5789: Add always_false flag in bloom filter

Patch Set 2:

File be/src/runtime/coordinator-filter-state.h:
PS2, Line 57: class Coordinator::FilterState {
> Could you add a comment up here explaining the different meanings of 'alway
These implications does not apply to other parts of the code. We are essentially doing an
OR(a list of bloom_filters) here so OR(empty list) should be false. And always_true short-circuits
the evaluation since future filters can no longer make any difference.  In comparison, in
scanner we need to apply multiple bloom_filters to a row, which is an AND, so any non-yet-received
filter is treated as true. And if one filter returns false we terminate the evaluation.

If we use disabled_ and is_inited_ we need additional synchronization between these two set
of flags, and the two sets of flags have no difference. And I think using always_false and
always_true here is algebraically more natural than is_inited_ and disabled_.
File be/src/runtime/
PS2, Line 1121:     swap(rpc_params.bloom_filter, aggregated_filter);
> What are the possible states after this swap() ? Since we're not explicitly
A check is added. (It could be always_false as well)
PS2, Line 1123: size()
> Since you changed the mem tracking from size() to capacity() in L1120 and L
You are right. Done.

To view, visit
To unsubscribe, visit

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: If680240a3cd4583fc97c3192177d86d9567c4f8d
Gerrit-Change-Number: 8170
Gerrit-PatchSet: 2
Gerrit-Owner: Tianyi Wang <>
Gerrit-Reviewer: Sailesh Mukil <>
Gerrit-Reviewer: Tianyi Wang <>
Gerrit-Reviewer: Tim Armstrong <>
Gerrit-Comment-Date: Mon, 09 Oct 2017 21:07:01 +0000
Gerrit-HasComments: Yes

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