impala-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tim Armstrong (Code Review)" <ger...@cloudera.org>
Subject [Impala-ASF-CR] IMPALA-5483: Automatically disable codegen for small queries
Date Wed, 14 Jun 2017 01:01:11 GMT
Tim Armstrong has posted comments on this change.

Change subject: IMPALA-5483: Automatically disable codegen for small queries
......................................................................


Patch Set 3:

1. Pretty big. The single node optimisation defaults to 100 rows total, this is set to 50,000
rows per node (e.g. 500,000 on a 10 node system).

2. Agreed. This checks the scan node estimates so is always disabled if there are a lot of
rows to be scanned (or stats are unavailable). The main case where this could cause a major
regression is an exploding many-to-many join, e.g. when two smallish tables are scanned but
it blows up.

3. Yeah I think the model breaks down when the cost of codegen is non-linear with the size
of the input exprs, but I think then we just need to fix the codegen. The inlining changes
help but we still have cases like IMPALA-5296.

-- 
To view, visit http://gerrit.cloudera.org:8080/7153
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I273bcee58641f5b97de52c0b2caab043c914b32e
Gerrit-PatchSet: 3
Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-Owner: Tim Armstrong <tarmstrong@cloudera.com>
Gerrit-Reviewer: Michael Ho
Gerrit-Reviewer: Michael Ho <kwho@cloudera.com>
Gerrit-Reviewer: Tim Armstrong <tarmstrong@cloudera.com>
Gerrit-HasComments: No

Mime
View raw message