Date: 2017-07-05 11:56:48 +0200
From: pavlos <<pavlos.katsogridakis>>
To: SQL devs <>
Version: 11.23.3 (Jun2016)
CC: @njnes, @yzchang

Last updated: 2017-08-02 10:21:31 +0200

Comment 25409

Date: 2017-07-05 11:56:48 +0200
From: pavlos <<pavlos.katsogridakis>>

User-Agent: Mozilla/5.0 (X11; Fedora; Linux x86_64; rv:54.0) Gecko/20100101 Firefox/54.0
Build Identifier:

Tried to run the tpch 03 query, taking a 10% sample on each table,
so I appended sample 0.1; to the query.
I expected the query to take a 10% sample of the tables and then run the
query. Instead, it executed the query and then it took a 10% sample.

Reproducible: Always

Steps to Reproduce:

1.Setup tpch with sf10
2.Append sample 0.1; to the 03.sql query

Comment 25413

Date: 2017-07-05 18:33:37 +0200
From: @njnes

indeed the sample operator (like limit) does it job on top of the result. Sampling per input should be done by hand.

Comment 25425

Date: 2017-07-06 11:14:36 +0200
From: @yzchang

Yeah, we got that sampling the base tables requires some manual work.

But how about the combined use of SAMPLE and LIMIT? Shouldn't SAMPLE be executed before LIMIT? Or should we actually disallow their combined use, since they both reduce the result set, only using different algorithms?

Comment 25539

Date: 2017-08-02 10:21:31 +0200
From: @sjoerdmullender

Fixed in Jul2017-SP1 release.